Changelog

Reporting a claim and hearing from your carrier before a renewal are two of the moments that decide whether a policyholder stays. You can now try both conversations yourself, exactly as a policyholder would meet them. Claims intake takes a loss one question at a time, asks about injuries first, and adapts to whether it was a car, a home, or a liability claim. Pre-renewal review catches the new car, the new driver, or the renovation that changed what they need, and hears a rate concern while there is still time to do something about it.

On the banking page, the application example now opens the agent it describes: the one that replaces the application form and learns who is serious before anyone picks up the phone.

57 new agents are ready to adapt across three industries: retail and e-commerce, hotels and hospitality, and banking and financial services. They cover the moments that decide whether someone comes back, from an abandoned cart or a delivery that went wrong, to how a stay actually felt, to the account that was opened and never funded. Browse retail, hotels, or banking, or filter the whole agent library by industry.

• If your customers are policyholders, shoppers, guests, or account holders, your industry now has a page of its own. Each one walks through what a conversation captures at every stage of that relationship, from the first quote or basket to the renewal, and links straight to the agents to start from: insurance, retail, hotels, and banking.
• Grok 4.7 is now the xAI option wherever you pick a model, replacing Grok 4.6. Workspaces already set to 4.6 move across on their own, so an agent running on xAI keeps going without a trip back to settings.

Completion flows can now send participants to booking pages or other destination URLs with the context those pages need already filled in. Names and emails from sign-in or invitations, plus campaign tags and CRM IDs from the conversation link, carry forward alongside answers captured during the conversation. Participants do not repeat themselves, attribution stays intact, and the destination can personalize what happens next. Known identity and metadata are available immediately, so the handoff does not wait for conversation analysis. Ask the assistant in Design to set it up, or see how to pass participant details to the destination.

Updated the dependencies behind Perspective to keep your workspace protected against newly disclosed vulnerabilities.

Sensei is our AI conversation quality coach. It reviews real conversations and production signals to catch failures. This time, it found a participant who returned to a conversation completed due to inactivity, answered the agent's last question, and got silence. Sensei traced the failure and gave us the evidence to fix it. Participants can now pick up where they left off after the system completes a conversation due to inactivity, while deliberately completed conversations stay completed.

• If Meta takes longer to finish connecting your WhatsApp Business number, Perspective now retries instead of leaving setup stuck. If the number cannot be used, workspace settings tell you why and let you choose another or go back to Perspective's number.
• The assistant in Design no longer repeatedly fails valid updates to titles, branding, and other agent settings.

• Your Perspective agents can now use your own WhatsApp Business number instead of Perspective's shared number. Connect your Meta account in workspace settings and choose the number. Connecting it does not disable the WhatsApp Business app, so you can continue using the app on that number. Participants see your business throughout the conversation, from the first message through every reply. Learn how to set up WhatsApp.
• Participants can now attach files during live voice conversations without leaving the call. Each file stays with the transcript for you to open later, and the agent acknowledges receiving it without reading its contents.

• A previous WhatsApp conversation tied to a deleted agent no longer blocks someone from starting a newly invited conversation. Replies to the old conversation now explain that there is no active conversation instead of returning a generic error.
• The assistant in Design now reliably finds existing conversations when you ask it to review them, instead of incorrectly saying that none exist.

• Finishing the welcome conversation or joining a workspace now takes you straight into your workspace when the closing card ends.
• When work is locked after your workspace runs out of credits, Add credits now shows exactly how many you need before checkout. Packages and tiers that cannot unlock everything are clearly flagged, so you do not buy credits and wonder why your conversations or design sessions are still locked.
• After the assistant in Design builds your outline, it now explains what participants will experience and offers one focused set of next steps matched to where you are. You can keep adjusting, try the conversation, choose what happens after completion, reach participants, or set up follow-ups without guessing what to ask next.
• You can now pull the sidebar open or closed by dragging its edge. A quick flick works like a drawer, and the width you chose is still there the next time you open it.
• Participants who try a different background can now spot the scene you chose as the default and return to it in one tap, on desktop or mobile.

Sensei continuously coaches our agents toward mastery by reviewing real conversations and production signals. It caught every issue below and gave us the evidence to fix each one.

• Choice fields now stay ready for reporting and routing. When a participant answers a fixed-choice field, completed conversations record one of the options you defined instead of free-form prose. Unanswered fields remain empty.
• Conversation transcripts now keep the order in which people actually spoke, even when a participant's device clock is out of sync. Summaries and analysis no longer see an agent reply before the answer that prompted it.
• Voice conversations now stay in the selected language, answer directly without narrating ordinary next steps, and stay in character after a quiet moment. They also wait when someone answers in text and can still wrap up after a silence reminder.

Behind-the-scenes improvements to keep your workspace running smoothly.

You told us it was hard to find your way back to the Design page, because Edit was hidden inside the three-dot menu on the results page. The right-side panel you already use in Design is now visible in the results and invite pages too, with Edit Agent at the top and the agent's outline, preview, settings and automations right below it. You can change anything about an agent without going back to Design. The panel also has a Share section: Send opens the invite view, and Copy Link grabs your agent's link so you can share it right away.

• On iPhone, tapping Type on the conversation page hid the reply box behind the keyboard. It now stays above the keyboard. Closing the file picker without picking a file no longer leaves the box stuck open either. It goes back to the Type or Talk choice.
• When the assistant in Design can't load a website or form, it now tells you within a minute or two. Before, it could sit there for a long time and then fail with nothing to show.
• Asking the assistant in Design to send participants back to Prodege now always adds the step to your completion flow. Before, it could quietly skip it. Also, a very short secret key saved in your integration settings is no longer shown back in full.
• Creating a concierge from the Try Concierge page no longer logs an error in your browser when the last details fail to load. The concierge, its link and the claim button all work as before.

• Looking for a ready agent to try or adapt? The agent library now lives under Agents in the main navigation. Browse by type, industry, outcome, or use case, and every agent-type page links straight into its full collection.
• Wondering how accents work when you pick a voice? The conversation settings guide now covers it: the accent comes with the voice, why you pin a language first, and how to search or filter the list by accent.

You can now send participants back to the panel that recruited them. Connect Prodege, the first panel company we support, in your workspace integration settings with the keys they issued you, then add a step to your completion flow for each outcome you report, complete or disqualified. Perspective then gives you the project URL to set on the Prodege side, and signs each participant's return link for you, so paid respondents get credited or screened out the moment they finish.

• The fast Google option for your text interviewer is now Gemini 3.8 Flash, at the same price as the one it replaces. Workspaces already set to Gemini 3.7 Flash move across on their own, so none of them quietly drop back to the Perspective default.
• Ask the assistant in Design to review a conversation and it now sees what the interviewer already knew about that participant before the first question: the context carried in the link, the invite, or a Slack profile, each labelled with where it came from. It could only read the messages before, so it had no way to tell a question the interviewer genuinely had to ask from one it should have skipped.

Restarting a voice conversation now connects right away, instead of waiting for the previous call to finish closing.

Behind-the-scenes improvements to keep your workspace running smoothly.

Pro workspaces get a new White Label section in agent settings. One switch removes the Perspective attribution line from conversations, link previews and participant emails. The other removes the try-it-yourself card that invites participants to sign up for Perspective once a conversation ends, so it closes on your own thank-you instead. Turn both on and the conversation is yours end to end.

• Replies in text conversations arrive about twice as fast, so a participant spends less time watching a typing indicator.
• Opening the full agents list is easier. The whole Agents header in the sidebar is now a link, with an arrow that shows on hover, and both it and the filter beside it have tooltips.

• Fixed the assistant in Design failing on every message.
• Opening a deleted insight, or one from another agent, crashed the chat page for good. It now says the insight is unavailable and carries on.
• Connecting a brand-new account through the Perspective MCP connector used to dead-end. A request to create an agent came back as an error with nothing in it you could act on. Your assistant now hands you a link to finish setting up, joining your team's workspace if they already have one or starting your own, then picks the request back up once you're done.

Updated underlying dependencies to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

We've reduced delays after a conversation ends. Results are prepared sooner, and participants can move on to the next page sooner.

• Pasting an image into chat now attaches it once, instead of adding two copies.
• Fixed an error that sometimes interrupted text conversations, agent design, and analysis sessions after a reply appeared.
• When you remove a workspace member, their agents move to the workspace owner. The name and picture shown in those conversations now stay the same.

Agent hand-off: the welcome Concierge seamlessly hands off the conversation to the design agent.

Contact Us is now a full-page conversation. Closing it takes you back to where you came from on the site, or to the homepage.

Loading time right after sign-up is now shorter.

• Copy to clipboard in analysis sessions now pastes as readable text and tables for answers that include tables, charts, or reports, with no raw data dumped in front of the text.
• The welcome card that greets participants no longer fades in twice at the start of an embedded conversation. It enters once and stays put.
• Participants in right-to-left languages such as Hebrew and Arabic who return to an unfinished conversation now see the message inviting them to pick up where they left off aligned to the right edge, like the rest of the screen.

The Outline version dropdown actually drops down again. A generation failure used to leave it stuck shut, so you could see the version label but not open the list.

Perspective AI MCP now exposes the Agent Library, so your coding or building agent can browse existing agent templates and use them as a starting point. Each match includes a shareable page, a live demo when one exists, and a ready-made brief to create your own agent.

• Calls to participants can now come from a number your workspace owns. Before you dial, the Call participants screen shows the exact calling number. You can also check it on the Phone Number page in workspace settings.
• Interested customers can also bring their own phone number. Contact us to set it up while we work on adding setup to workspace settings.

• Your agents now use the context you pass with a conversation, such as the participant's source page, intent, and use case, to guide the questions that follow. The participant's answers and your Outline still take priority.
• Finishing a conversation now gets participants to their next destination faster. In our own onboarding flow, we measured that the hand-off took less than half the time.

Participants on iPhone can attach a photo or document at the start of a conversation again.

You can now email yourself the summary of any finished conversation. Open the conversation's actions menu and choose Email summary to me. Useful when the original summary went to a previous owner's inbox, or when you simply want a copy in yours.

• In text conversations, participants can now attach a photo or document before typing anything. The attach button shows as soon as the reply box appears, and a file alone is enough to start the conversation. When your agent asks for an ID or an insurance card, nobody has to work out how to send it.
• Picking a voice is easier. The Voice for Conversations picker under Language & Voice is now searchable, and you can filter it by accent with a tap. Each voice reads like a person: its name, traits like gender, age, and tone, a flag for its accent, and a count of the other languages it speaks.
• Workspace Admins now receive the same account emails as the Owner: low credits, out of credits, credits purchased, and plan changes. An admin who runs the workspace day to day no longer hits an out-of-credits wall without warning.

You can now transfer ownership of an agent to a teammate. Open the agent's settings, look under Advanced, and choose Transfer ownership next to the owner's name. The agent and all of its conversations move together. If the agent's Access is set to Owner & Collaborators, you lose access to it once it's transferred.

• Automations are now one click away on the Design page. They have their own entry in the side panel menu next to the design chat, right under Settings, instead of a collapsed section at the bottom of Settings.
• Onboarding now runs full screen. When you sign up or join a workspace, the conversation used to sit in a small box over a greyed-out app.
• The My agents button on the website now opens right away. It used to take a few seconds while Perspective worked out which workspace page to send you to. Now you don't have to wait.

Deleting an agent now stops its scheduled digests. Before, one last digest could still go out after the delete and use credits.

Updated underlying dependencies to keep your workspace secure.

You can now write your own button text for invitation emails, so the invite asks people for exactly what you want. The compose step also shows the email as it will arrive, so you can check it without sending yourself a test first.

The sign-up flow is now significantly faster, with the first screens appearing right away.

Updated underlying dependencies to keep your workspace secure.

Workspace Model settings now include a Thinking control for text interviews. Slide from Speed to Depth when you want quicker turns or richer follow-ups.

• Text interviews now run on the latest models from each provider. Older options are gone from the picker; if a workspace still pointed at one, it moves to the closest current model instead of falling back to the default. Models are grouped as Recommended, Fast, and Smart so the choice stays easy.
• Text conversations are up to 9x faster by default, which makes for a much better experience.

• On the welcome screen, Talk and Type stay side by side in languages with longer labels, and choosing Type no longer stretches the box out of shape.
• Finishing onboarding with only a first name now continues to your destination instead of sending you through the welcome conversation again.

Visiting /agents now opens the Agent Library.

• When you remove someone from a workspace, their agents and conversations now move to the workspace owner without changing who took part in each conversation.
• When you share an agent in Slack, the message now tags the agent owner instead of showing their name as plain text.

The opening greeting is up to 5x faster. Since 24 August it had been pausing to think before it spoke, so participants watched an empty bubble for several seconds, and on some agents far longer. The first word of a conversation now lands in about a second, at exactly the moment a conversation can least afford to lose someone.

You can now bring an interview that happened outside Perspective into one of your agents. Ask your AI assistant to import the transcript from an in-person meeting or a Zoom, Meet, Teams, or Gong call, and it lands in the agent and gets analyzed like any conversation Perspective ran itself. A backlog of past interviews can join the same insights instead of sitting in documents. Setup is in the MCP docs.

Old documentation links work again. Several rounds of restructuring had left bookmarks and search results pointing at pages that no longer existed. Those now open the current page, along with older use-case and blog category links.

You can now ask us a question from any page on this site. A conversation opens in the corner, so questions about pricing, features, or whether Perspective fits your use case get answered on the spot instead of becoming a form and a wait.

A participant whose microphone can't start, because there isn't one, another app is using it, or the operating system blocks it, used to land on a raw browser error with nothing to do but close it. They now get a clear explanation in their own language and a Switch to Text button, so the conversation carries on in writing instead of ending there. The same fallback covers a microphone that drops mid-conversation, which used to show a misleading connection error.

• Coming back to a conversation you already started now opens with a proper greeting instead of a pop-up you have to dismiss. The welcome screen recognizes you by name and offers Continue or Start over right where the Type and Talk buttons normally sit, so on a phone or inside an embedded conversation nothing is covering the screen while you decide. Participants land in the conversation they actually meant to be in, and more of them finish the one they came back for.

• Sharing a conversation link now produces a preview that reads like an invitation rather than a plain card. It shows who is asking, roughly how long the conversation takes, a Start conversation button, and Perspective attribution unless you have branding switched off. Anyone deciding whether to click can see what they are agreeing to before they open it.

• Participants working in Spanish, German, Russian, Hebrew, Japanese and the other eleven languages the conversation screen supports were still meeting two stray English words: the Type label beside the Talk button, and the prompt to resume a voice call. Both now appear in the participant's own language, and the control grows to fit longer words instead of letting them run underneath the Talk button.

• Reopening a conversation you had already begun could drop you straight back into it without saying so, which meant answers you gave expecting a fresh start were added to the earlier conversation instead. Returning now always asks which one you want first.

You can now drag the edge of the sidebar to make it wider. It remembers your choice, so long agent and session names are easier to read. You can also resize it with your keyboard.

• When you open a Perspective link inside another app, such as LinkedIn, the LinkedIn and Microsoft sign-in buttons are visible again.
• The bold, italic, list, and link buttons in invitation messages now stay at the bottom of the message box, even when the message is short. You can click anywhere in the box to start typing.
• A host's title on the welcome card now stays visible whenever there is enough room. It was sometimes cut off even on wide screens.

Conversation links shared in Slack now preview like any other link. Your agent's name is the clickable title, the opening message sits under it, and the preview image is scaled to fit instead of the oversized one Slack used to show. The preview still includes the Start in Slack and Open in Browser buttons.

Perspective agents got better at running conversations. Each one now follows an established method for the kind of conversation it is having, instead of general advice. Concretely: a discovery conversation now asks what people actually did instead of what they think they would do, a survey explains a question in its own words when someone does not follow it, and an agent making a case gets to it as soon as it hears a real concern. New agents get all of this automatically. Existing agents pick it up too, but their own saved guidelines still take priority, so ask Perspective to update the outline once to bring an older agent fully in line, which matters most for an Advocate built around a fixed order of stages.

• Taking a conversation that requires participants to sign in no longer signs you out of Perspective. Trying out your own conversation used to mean signing out of the Perspective dashboard first, or opening a private window to work around it. Now the conversation signs you in on its own and leaves your dashboard session alone.
• Your credit balance now updates the moment it changes, everywhere it appears: the workspace switcher in the sidebar and the credit usage report. A page left open while credits were bought or used kept showing the old numbers, so a purchase could look like it never arrived.
• Changing your plan on another device no longer leaves the page showing your old plan next to your new balance.

Voice conversations now put closed questions on screen. When your agent asks a participant to pick from a list, give a rating, or answer yes or no, the options appear as tappable buttons while the call keeps going. Participants can tap an answer or still say it out loud, whichever is easier. A rating or one option out of six is exactly the kind of short answer that gets misheard when spoken, so making it a tap means fewer participants stall partway through.

Ask for an accent in a design session and Perspective now handles it end to end. Say "make the voice Australian" and it points you to English, where those voices live, explains what pinning the conversation language means for your participants, and confirms before it changes anything.

Hardened data handling to keep participant details private.

Viewing results and jumping between conversations, design sessions, and analysis sessions all take a little less time, with fewer loading screens along the way.

Small fix to the agent card above the chat input in design sessions. If saving your first agent failed, the card could still appear as though the agent was ready to copy, invite, or test. It now waits until an agent has actually saved.

While you design an agent, it now stays pinned just above the chat input instead of scrolling away in the thread. You get a compact card with Copy, Invite, and Test: open it for the full preview, title, and share link. Updates stay in the conversation until the new version is ready, so the card always shows the last finished agent, not a half-built one.

• Fixed an issue in the design session where the conversation kept saying it was still designing your agent after the attempt had already failed. The error now appears right away instead of only once the assistant finishes writing.
• Fixed the avatar in the preview people see when you share a conversation link. Missing avatars from a Google account could show up as your initials instead.

The Trust Assessment card on the conversation results page has a fresh look: it matches the design of the cards around it, follows your workspace theme, and fits comfortably on phones, so the score and recommendations are easier to scan wherever you read them.

Fixed the Perspective MCP connector in Claude Cowork. The connector described its tools in an older schema format that stricter surfaces such as Cowork rejected before any tool could run, while regular Claude chat kept working. The tools now use the current standard, so they load and run normally again, with nothing to reconfigure on your side.

Updated underlying dependencies to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

Arabic and Hebrew conversations can now use a genuinely Arabic or Hebrew voice, with the Arabic voices spanning Modern Standard, Egyptian, Levantine, Gulf and Moroccan. Participants in those markets hear someone who sounds like them. These voices are marked Beta, since pronunciation is still being refined.

• English voices now span the accents people actually speak, Scottish, Irish, Nigerian and Jamaican among them, in a list short enough to read all the way through. Languages like Tagalog now offer genuinely native voices too.
• Conversation language and voice now sit together in one Language & Voice section, language first, because the language decides which voices you can choose from.

Grok 4.6, xAI's newest flagship, replaces Grok 4.5 in Perspective Agents. Select it from the model menu when you build an agent or run an analysis, or choose it as your workspace text interviewer model. Anything still set to an older version of Grok moves over on its own, with nothing for you to update.

Small fix in workspace settings: long pages scrolled inside an inner panel, putting the scrollbar in the middle of the screen instead of at the edge of the window. Settings pages now scroll normally, like the rest of the app.

Behind-the-scenes improvements to keep your workspace running smoothly.

Fixed an issue where leaving a voice conversation could skip the closing chime. Participants now hear when the call has ended.

Voice conversations now open and close with a chime: a short rising tone the moment the line goes live, and a falling one when the conversation ends. Participants hear when the agent starts listening and when the conversation is over, the way a phone call tells you when it has been picked up and when it has ended.

Fix in voice conversations and phone calls: a cough, a knock, or background noise no longer counts as an answer, and if something cuts the agent off while the participant stays silent, it now carries on from where it stopped.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Fix in workspace integrations: turning an integration off now stops Perspective from using it on its own. Exports to Google Docs, Notion, and Confluence pause, as does anything the AI sends to Slack, HubSpot, Salesforce, or Gmail, until you turn it back on.
• Fix in voice conversations: in some cases the agent carried on talking after a participant jumped in. It now stops as soon as they speak, so they can cut in the way they would with a person.

Updated underlying dependencies to keep your workspace secure.

• The agent-ready card in the design session now shows the title and welcome message participants see when they open your link, plus the outline version that change produced. You see a new card each time your agent is generated or changed, so it is clear what you are about to share and which version you are looking at.
• New guides for importing and exporting conversations. They cover the file types and transcript formats we accept, the size limits, what lands in an export, and how to move conversations between agents or workspaces.

Fixed a small visual glitch in your agent's settings: focus outlines under Language, Voice & Tone, and Advanced were clipped at the edges. They now show in full.

• Restored and expanded the available voices for Argentinian Spanish and Australian English accents. You now have dozens to pick from in each accent, and you can preview them before you choose.
• Fixed some edge cases where the agent got cut off while speaking longer answers, especially closing summaries. It now has room to finish, and picks up where it left off if it ever does run long.
• Fixed voice conversations that could start up without any sound, or stop hearing the participant midway. If a conversation cannot connect at all, participants are now told and offered the text version.
• File attachments now work with OpenAI GPT models. PDFs and text files were not getting processed properly, which caused errors in design sessions.
• We now show accurate token usage in design and analysis sessions. Hover the small dial next to the model picker to see how full the conversation is.

Updated underlying dependencies to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

Voice conversations can now speak in three new accents: Australian English, Indian English, and Argentinian Spanish. Pick one as the conversation language and preview the matching voices before you choose, or just describe the voice you want while designing your agent in chat. Participants hear someone who sounds like them, not a generic machine voice. Phone calls keep the standard voice for now.

• Starting a voice conversation feels smoother, with the call controls easing into place instead of jumping.
• Your workspace menu is tighter and tidier on a phone, so more of your agents and sessions fit on screen.
• Conversations keep moving when the AI is slow to respond, so participants spend less time waiting.

• A few small fixes to the workspace menu on a phone, which now opens cleanly instead of covering what you were looking at.
• Participants starting a voice conversation now see the start button in their own language.

Updated underlying dependencies to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Voice conversations now open with the written conversation on screen, so participants can read along while the agent speaks: closed captions, in effect. The words are right there for anyone hard of hearing, in a noisy place, or following along in a second language. Giving the persona the full screen is still one tap away.
• Design your agent in another language and the conversation language is set for you: participants get conversations, and noticeably more accurate voice transcription, in their language without you touching a setting. When your audience's language is only likely, the agent offers a short suggestion instead of deciding for you, and it leaves any language you've set yourself alone.

• Fine-tuned how voice conversations handle overlap: on the rare occasion a participant and the agent speak at once, the agent gives way and listens from the very first word.
• Asking for a text-only conversation while designing your agent now lands exactly as you mean it: participants simply get a text chat on the web, while phone calls stay spoken.
• A small fix to the recent conversations feed: date dividers now scroll away with the list instead of stacking under the page title.

Updated underlying dependencies to keep your workspace secure.

• Import transcripts or recordings from meetings held in other apps into a Perspective agent. Your team can view everything in one place and analyze those conversations alongside conversations conducted by Perspective agents. Learn more.

• Export conversations conducted by Perspective agents for reporting, collaboration, governance, or further analysis in the tools your organization already uses. Learn more.

• Large imports and exports run in the background, so you can keep working while your data is prepared and receive an email when it's ready.

• Agent design and Analysis Sessions in Perspective now use GPT-5.6 Terra by default. It can infer your intent and work through complex requests with less step-by-step prompting, from shaping a new agent to finding meaningful insights across conversations.

• Keep up with every release by following us on X. The changelog now includes a Follow on X button alongside the RSS option.

Updated underlying dependencies to keep your workspace secure.

• Voice and phone conversations are transcribed by a new speech model with much higher accuracy, and it now gets the conversation's context to work from: what your agent is about, so words specific to your topic come through, and background noise or an unclear moment no longer turns into words your participant never said. What someone actually told you arrives in their own words, not an approximation of them.
• Hiding Perspective branding is clearer now. Turning on Remove Perspective branding also takes the "Perspective AI Agent" line under the message box with it, and the setting is honored in the design page's Agent view and the conversation preview, not just in the live conversation. We'd still love it if you told everyone about us anyway.
• Voice conversations now open with Murmur, and the persona sits a little smaller and closer to the message box so it shares the screen with the conversation instead of dominating it. Jelly and Weave are still one tap away.

• The persona animation ran at whatever rate the participant's screen refreshed, so on a high-refresh-rate display it moved at double its intended speed and read as frantic while the agent was speaking. Every screen now shows the same, intended pace.
• Testing your agent before you share it now shows exactly what participants will see. The Agent view on the design page and the conversation preview were both falling back to the older two-screen text and voice experience instead of the single conversation view.

• Text and voice now share one chat surface instead of switching participants between separate views. The conversation stays put while the typing box transforms into compact floating call controls, and hanging up returns straight to typing with the transcript still in place. Above the controls, an animated persona reacts to whoever is speaking. Tap to cycle through Jelly, Murmur, and Weave, or hide the transcript and give the persona the full conversation view.
• New automation templates make it easy to enable daily and weekly insight automations for every agent or only the agents you choose.

Voice-only conversations now open directly in voice mode. If a call fails, participants can try again instead of landing on a text path they cannot use.

• An agentic sidebar. The app shell is rebuilt from scratch around your agents, matching the way the rest of the product now works. If you spend your day in Slack, Cursor, Codex, or Claude Code, it will feel immediately familiar: one quiet column that holds your work, collapses to a rail when you want the room, and reopens with a click or a keystroke. It lists your workspace's active agents by default, with drafts and ready ones a filter away: you can re-sort or switch state, type, and ownership right in the sidebar, or open the full list with Show all. Getting back into something that is running no longer starts with a search.
• Below your agents, every session you have had with Perspective, newest first: the design sessions where you built an agent and the analysis sessions where you worked through what came back. Page through the whole history in place and carry on where you stopped, so returning to yesterday's thinking is a click rather than a hunt.

Navigation is simpler alongside it. Activity is now Conversations, and help, feedback, and settings sit together under your profile. Single-key shortcuts work the same way everywhere, with the key shown in the tooltip when the sidebar is collapsed. Your workspace and its credit balance moved to the bottom and stay visible in the collapsed rail, so you can see what is left without opening anything.

• The single-letter navigation shortcuts no longer fire while a menu or dialog is open. A letter typed into one of those stays there instead of jumping you to another page.
• A small one on the design page: the Outline view now stays greyed out until an outline exists, rather than opening an empty panel.

The analysis agent can now turn a refinement into an automation. When you refine an automation-generated insight in an Analysis Session and the change is reusable, the agent offers to update the source automation or spin up a new one, proposing its name, schedule, instructions, and filters and changing nothing until you confirm.

• Automation-generated insights now remember where they came from. An insight created by a scheduled automation keeps its link to that automation, so you can open the automation's canvas straight from the insight to review or adjust it instead of hunting for it.
• Daily automation runs are quieter. A scheduled run now reviews recent insights before analyzing new conversations and creates one only when there is materially new, changed, or actionable information, so a run that finds nothing valuable finishes cleanly instead of adding noise.
• Insight cards now show the exact last-updated date and time, kept visible on desktop and in a dedicated row on mobile.
• The automation canvas back button now appears only when there is somewhere to return to.

Your automated daily and weekly insight updates are now cleaner and easier to scan. Each update separates completion metrics (active conversations, completed, and completion rate) from qualification metrics (qualified, unqualified, and qualification rate), shows the right chart for each, adds a short findings section tied to your study's own question, and ends with evidence-based action items. It is faster to see how a study is performing and know what to do next.

• Running a daily or weekly insight automation on demand now analyzes the intended time window, so a manual run reflects the same period it would on its schedule.
• You can now stop an analysis and immediately ask a revised question. Previously, stopping and quickly resubmitting could fail with an error; cancellations now settle before the next question begins.

Updated underlying dependencies to keep your workspace secure.

• The workspace Automations page used to show summary cards, and changing most settings meant asking the design agent in chat. Now, click any automation on that page to open its canvas. There you can see its full details, status, and errors, then directly edit its instructions, conversation filters, schedule, recipients, delivery destination, or participant email content. Participant emails include a preview, while webhooks show what information they send. You can understand, update, and troubleshoot an automation without leaving the Automations page.
• Scheduled automations previously started only at their set time, so checking a new setup could mean waiting until the next day or week. Now, Run now starts the automation immediately without moving its next scheduled time. You can verify a recent change while it is still fresh or start time-sensitive work as soon as you need it. Previous or Active Runs shows whether each run completed, was skipped, or failed, and opens any conversation or Insights it produced, so you can investigate problems without hunting for the result elsewhere.

• Insights are now a timeline, not a flat feed. Workspace Insights and each agent's Highlights and Pages were previously long lists of cards. They now join Updates in a dated timeline grouped under Today, Yesterday, and older dates. The current date stays visible as you scroll, and compact search and filters leave more room for the results themselves.
• Analysis Sessions, Results, and workspace Insights previously opened their content in different sidebars. They now share the same canvas and editing tools, with sources, citations, and exports close at hand, so you can review the evidence or make changes without learning a different workflow on each page.

• The template gallery is now organized by team, and four new teams join the lineup: Customer Experience, Digital, Rev Ops, and People & HR. Pick your team and land on templates curated for the work you do.
• Template browse pages now take you deeper instead of sideways. A team's page (take Marketing) groups its templates by the business outcome you're chasing, an industry's page by the job to be done, with tiles up top that jump straight into each group. Other teams and categories moved to the bottom of the page, still one click away.

Saved transcripts of voice and phone conversations now match what was actually said. When a participant talked over the agent, the saved transcript could show that exchange out of order, and the agent's "are you still there?" reminder could be recorded twice. Both are stored correctly now.

Updated underlying dependencies to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Fixed a rare case where an inaudible voice answer could stop a conversation before the participant finished.
• Google Docs exports now guide you when a permission is missing. If you skipped a required Google permission while connecting, the export previously failed with a vague "Unknown error." Now it names what to approve and helps you reconnect, and it surfaces the real reason when something else goes wrong. Successful exports are also shared with you automatically, so the new document is ready to open right away.

Updated underlying dependencies to keep your workspace secure.

Every new agent now comes with a daily and a weekly insight update that lands straight in your workspace's Insights feed. The daily update covers the day at a glance, the most important learnings, and any action items. The weekly update reviews the week and calls out new patterns and emerging trends. Days with no activity are simply skipped. On existing agents, turn the updates on by sending "Setup default insight automations" in your design chat. You can also change when the updates run, or create custom ones with their own instructions and conversation filters.

Routine security and dependency updates to keep your data safe.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Designing an agent now keeps everything in one place. The Outline, a live preview of your agent, and its settings open in a side panel right next to the design chat, with a compact menu to switch between them. The panel reopens where you left off, you can drag it to the width that suits your screen, and on smaller screens it opens as an overlay.
• The Conversation Language setting has moved into its own Language section in your agent's settings, so it is no longer tucked under Voice & Tone. Learn more.

• After a recent update, a saved phone conversation could show only the participant's side of the call. The agent's spoken replies are now recorded again, so the whole conversation is in your transcripts: every word your participants heard is part of the record.
• When a participant ends a voice or phone call while the agent is mid-sentence, the words they heard now stay in the saved transcript instead of being lost.
• Opening a blog article occasionally failed with a server error. Articles now load reliably.

Analysis sessions can now cite saved Insights, not just conversations. Select an Insight citation to open the exact saved finding, report, chart, or table beside your analysis, with a link back to the session that created it. You can follow the thread of evidence without losing your place.

• Saved Insights now behave consistently across Analysis, workspace Insights, Highlights, and Pages. The title, version, citations, and actions stay together on each saved Insight, and every list supports search plus Published and Drafts filters, so the next action is in the same place wherever you opened it.
• Full-study exploration now works across more than 1,000 conversations, includes meaningful in-progress conversations, and checks several discovery lenses plus a final coverage gap before synthesis. Broad findings can be tested against the full study instead of relying only on a small relevance sample.
• Automation cards are easier to scan and edit. Destinations and daily or weekly schedules appear in the title, running or failing status stays close by, and long webhook addresses keep both ends visible with the full address available on focus or hover. For chat-based edits, describe the change on the card before chat opens; Perspective carries your request, destination, and trigger into the session. Webhook edits keep their secure configuration dialog.

• Citations in saved Insights now open the correct conversation message, including older citations and conversations where participants chose an on-screen answer.
• The New insight and New automation dialogs now close immediately when dismissed or after you choose an agent.
• Insight lists now refill after publishing moves an item out of view, so loading later pages does not skip the next saved Insight.

• You can now set the language your conversations run in. Perspective opens and conducts the conversation in that language, and voice calls, phone calls, and spoken answers are tuned to listen for it, so participants in any of 57 languages are understood accurately instead of being misheard or transcribed in the wrong one. Auto-detect stays the default for mixed audiences. Learn more.
• While you set up an agent, you can now ask Perspective for a ready-to-paste embed snippet for your site, pull up your share and preview links, publish or unpublish the agent, and check what any setting is currently set to, all right in the chat instead of hunting through the settings sidebar.

• Magic Summary now shows its progress and writes the summary out as it goes, so a longer analysis no longer sits on a blank screen feeling stuck. You see it working, and you can start reading the moment the first lines land.
• Insight charts now stay on-theme and readable everywhere they appear, and you can recolor a chart yourself: pick from the product's theme colors and map them to the exact items in your chart. Shared and published views stay read-only.

Voice conversations now save and show the transcript exactly as it was spoken. Replies the participant never actually heard no longer slip into the record or split a turn in two, and the saved conversation stays in the right order.

A small polish to the dictation experience introduced in interview conversations: the input now holds to a single line that keeps pace with your speech instead of growing tall, and the send and attach buttons stay in place beside it even on a longer message.

• Participants answering by text can now speak their responses instead of typing. The same voice-to-text mic appears right in the chat input, so anyone who would rather talk can, without switching into a full voice conversation. Less effort to answer means more people finish, and you get richer answers in their own words.
• Perspective now works with Salesforce. The leads, updates, and context from your conversations can flow straight into your CRM: a qualified conversation becomes a new lead, the right opportunity gets updated, and accounts stay current, without anyone copying data over by hand.
• Now you can ask Perspective to make sense of all your saved insights at once. In an analysis session, compare themes across time periods, analysts, or conversation types, or track down every mention of a topic, without waiting for it to read each insight one by one. Every finding comes back with citations checked against your real insights, and when it can't ground an answer it says so instead of guessing.

The Insurance template library now covers listening, not just capture. Six new agent templates run adaptive AI conversations that go deeper than a static survey can: a Post-Claim Experience Survey that asks more careful questions after a denied claim than after a smooth repair, plus ready-made templates for why policyholders switch, quote abandonment, agent and broker experience, app and digital experience feedback, and concept testing. Browse the templates.

Uploading a workspace logo or avatar no longer crops the edges by default. Fit now shows your complete image and Fill covers the frame, repositioning stays within view, and zoom responds smoothly to external mouse wheels instead of jumping straight to the minimum.

• Your AI assistant can now analyze your Perspective conversations, not just manage them. From ChatGPT, Claude, Cursor, or any client connected to Perspective MCP, ask for a quick health check on an agent (its current phase, conversation volume, participant feedback, and remaining credits), pull up the insights already saved in your analysis sessions, run counts, breakdowns, and trends across conversations, ask "what do people say about X?" and get answers with cited excerpts, or launch a deep transcript exploration for exact phrases, comparisons, and close reads that reports back with sources.
• Participant invite messages now come with a rich text editor in the invite view: add bold, italic, lists, and links, and your invitation arrives formatted the way you wrote it.

Fixed a bug where line breaks in invite messages collapsed into a single block, so multi-paragraph messages arrive readable again.

• Starting a new interview is now instant, with a plus icon. Handy for demoing or documenting a conversation, or restarting one during a support call, without an extra confirmation step in the way.
• Credit usage now shows a single clear bar broken down by session type, instead of per-row percentages, and no longer says credits are needed when your balance already covers it.

Small fixes to template search: the "no matches" message no longer hides behind the see-all link, and the results panel no longer clips on the gallery page.

A new Credit usage view breaks your spend down by conversations, design, and analysis. Find it in the workspace switcher or a dedicated Settings page.

• Template search now opens with popular collections and top picks, narrows instantly as you type, and matches by what a template does, not just its title. It no longer navigates you away mid-keystroke on template pages.
• Insurance templates now speak the market's language: the quote template is Insurance Quote, insurance has its own gallery section and industry filter, and two new templates cover claims (FNOL) and policy renewals.
• GPT-5.6 Terra and Luna are now available. Terra is the new recommended default for design and analysis sessions, and Luna is a faster option for text interviews.

Fixed a text glitch on public use-case pages where a bolded phrase followed by punctuation could leave a stray asterisk and bleed the underline into the next word.

The Perspective design agent now sees how your agent is doing in the real world. Ask "where does this stand?" and it answers from live data: the project's current phase (design, test, share, or live), how many conversations have come in, the completion rate, and your remaining credits. It can also read the insights you've drawn from those conversations in your analysis sessions, list them, pull up any one in full, and fold what they say into its suggestions, so refining your agent builds on your own findings instead of guesswork. When an insight needs editing, it points you straight to its analysis session.

Hardened behind-the-scenes file handling to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

The welcome teaser on Float embeds, the message bubble that greets visitors above the launcher, is now yours to shape without touching code. Turn it off, choose how long it waits before appearing, or mute its chime, either from the embed customizer on the Invite page or just by asking the AI assistant. Publish your changes and they reach embeds already live on your site automatically, so the greeting evolves with you instead of being frozen at install time. See the embedding guide for details.

Attaching an image could leave the assistant unable to reply: once a screenshot or photo entered the conversation, every following turn failed, both while building a Perspective Agent and in analysis sessions. Images now come through fine and the conversation carries on.

• A new Insights page gathers every insight from all your Perspective Agents into one place. Filter by type or by draft and published state, search by title, description, or agent, and browse the whole list without opening each one individually. Draft insights publish right from their card: turn one into a highlight or a page, or export it. The list updates the instant you publish, unpublish, or delete, so there is nothing to refresh.
• Grok 4.5, xAI's newest flagship, is now one of the AI models you can choose in Perspective.

• Building a Perspective Agent now holds up when your connection wavers. Refresh the page, switch devices, or hit a brief network hiccup while a reply is streaming in, and it keeps going instead of disappearing. You can also reliably stop a reply in progress.
• The Perspective Agents on our own site, like the contact, Ask Perspective, and newsletter ones, now load more smoothly.
• Added docs for the welcome teaser on Float embeds, a message bubble that greets visitors above the launcher. Read more in the embed API reference.

• Fixed two display glitches in the chat. While you build a Perspective Agent, finishing an in-chat step like picking a Slack channel or connecting an integration no longer makes the previous reply appear twice as the next one streams in. And during a long analysis, a brief connection hiccup no longer scrambles or duplicates the answer until you refresh.
• On iPhone, the progress bar in a conversation could get stuck overlapping the first line of your Perspective Agent's messages, drawing a colored line through the text. It now stays in the header where it belongs and never draws over the conversation.

• The template gallery now meets you where you are. Search up front, or browse from your own vantage point: your role, your industry, the outcome you're after, or the form tool you're replacing. Every path lands on ready-to-use templates you can try live, so instead of scrolling one long list hoping something fits, you start from a proven agent that already matches your situation and launch in minutes.
• Voice and phone conversations listen more sharply. Spoken email addresses, phone numbers, and other letter-and-number details come through more accurately, pauses and background noise are handled more gracefully, and the agent responds more naturally when you jump in mid-sentence, so participants can just talk the way they would to a person.
• Choosing how a conversation looks is simpler: the appearance menu now offers a two-way Light and Dark switch that opens already matching the theme you're seeing, whether that comes from your device or the page hosting the conversation. Pick the other mode to override it; pick back and the conversation follows its default again. The separate System option is gone.

The conversation page now lets participants start a brand-new conversation with your agent in one tap. When your agent allows multiple conversations per participant, a Start new button appears right on the page, and it works both ways: begin again after finishing a conversation, or end the current one midway and start over. A quick confirmation guards the switch, since the current conversation wraps up for good. Each conversation is saved as its own record, so the previous one stays in your results with its full transcript and every new one adds another response instead of replacing the last.

Setting up a webhook automation through a connected AI assistant now keeps endpoints and credentials out of the chat. The assistant shares a secure Perspective link where you enter the endpoint URL and any authentication header after signing in, and the automation stays off until that's saved, so secrets never travel through the conversation. Delivery-failure messages shown in the assistant are scrubbed too, so a failed webhook never echoes its address back into the chat.

• Fixed a flicker on conversation pages with a background scene: the page could paint plain and have the scene photo pop in a beat later, and hovering scene options in the appearance menu flashed the backdrop between previews. Scenes now appear only once fully loaded, with a smooth fade on first visit and instantly on return visits, and embedded conversations hold their loading placeholder until the page is fully drawn.
• Long titles could stretch cards in the analysis sessions list past the edge of the page. Titles now trim cleanly, and every card stays inside the list.

A new Automations page brings every automation in your workspace into one place, across all your research projects. Search and filter by trigger, channel, delivery status, and health to find the one you need, then edit, delete, preview, or turn automations on and off without opening each project one by one. Failing automations are surfaced up front, so a delivery that quietly broke no longer hides inside a single project until someone notices.

Your research list is now My agents, at a clearer /agents address that reads as what it actually is: the list of your agents, instead of a generic "Home." Existing /list bookmarks keep working, and a new keyboard shortcut jumps you straight there.

Bullet and numbered lists now render as you type in every chat input across Perspective. Start a line with a dash or "1." and it becomes a real, indented list; Enter continues the numbering, and Enter on an empty item ends the list. Your message keeps that formatting after you send it, so a list arrives as a list, whether you're designing an agent, analyzing results, or replying in a text conversation.

• Conversations embedded on your site now adapt to the height of the frame you give them, not just its width. In a compact widget or a short embed, the header and controls scale down instead of crowding out the conversation. Voice conversations get a compact control dock at short heights, and participants can tuck it away into a slim strip to read the transcript, then bring the controls back with a tap.
• The docs home is now a self-service hub. Two start-here guides sit up top, each topic card previews its top articles with a count of everything inside, and when you still can't find an answer, you can ask Perspective directly or contact the team from the same page.

Sign up with your company email and Perspective now shows you the workspaces your teammates already use, so you join the team where work is already happening instead of unknowingly starting a duplicate. Admins stay in control from the new Domain access section in workspace account settings: approve each request individually (the default) or let anyone with a matching email join automatically. Owners and admins get an email when someone asks to join, and requesters hear back the moment they're approved.

• Reading the blog is easier on the eyes. Each article now opens with a clear title and byline instead of a cropped image, body text is larger and airier, data tables render as clean report tables, the "In this article" panel sits in a compact two-column card, and every post ends with visual suggestions for further reading.
• The category sidebar in the templates gallery now stays in view as you scroll and opens to the category you're browsing.

• Three templates (Restaurant Feedback, Hotel Guest Feedback, and Competitor Analysis Interview) belonged to no category in the templates gallery, so browsing never surfaced them. They now appear under Customer feedback and Marketing & brand research.
• Typing quickly in the template search no longer loses keystrokes: the field could snap back to an earlier value while results were loading.
• Workspace membership checks now match member emails regardless of capitalization, so a teammate whose email was saved with capital letters no longer gets missed by access checks.

The templates gallery is a revamped experience. Each template's demo conversation, title, and usage now sit together in one card, colored by category so similar templates are easy to spot at a glance, and you can try a template live right from the gallery instead of opening its page first.

• Voice conversations now follow your brand. The tap-to-talk button and the animated background pick up your custom theme colors instead of the default purple, so voice matches the rest of your branded experience.
• The controls in the conversation header are more neatly spaced: buttons group cleanly and the logo no longer crowds them on smaller screens.

• Asking your analysis to break conversations down by voice vs text now returns the right numbers. Conversations held over voice were sometimes counted as text; the breakdown is accurate again.
• After our recent upgrade to Claude Sonnet 5, the short summary of the assistant's reasoning in research and analysis chat went blank. It shows again, so you can follow how it works through your request.
• In voice conversations on desktop, the "by Perspective AI" badge overlapped the appearance control in the bottom-left corner. It now sits on its own in the bottom-right, so the screen looks clean.

New agents now start with a background scene by default instead of a blank backdrop, so your participants' conversations look polished from the very first draft. Different creators get different default scenes, and you can change or clear the scene anytime in your agent's appearance settings.

Designing your Perspective Agent and analyzing its results now run on Claude Sonnet 5, Anthropic's latest Sonnet model, for higher-quality reasoning as you build an agent, dig into your results, and generate summaries and charts. Prefer the previous model? Claude Sonnet 4.6 is still there in the model picker.

Uploading a logo or avatar with a transparent background no longer puts a black box behind it. Your branding images now keep their original format, so a transparent image stays transparent.

Behind-the-scenes improvements to keep your workspace running smoothly.

• You can now turn off the "Tell us why?" follow-up link that appears after participants rate their conversation experience. The rating options always stay visible, so you keep the quick pulse without sending participants off to a separate feedback page.
• The control for changing a conversation's background scene is easier to spot, with a clearer icon and a matching "Scene" label across desktop and mobile.

Voice conversations run on a dark screen, but popup alerts like "microphone blocked" were still showing in light mode, flashing up as bright boxes that looked broken. They now match the dark theme, so the screen stays consistent.

• Fixed a brief error message that could flash at the very start of an analysis session, even when nothing had actually gone wrong.
• Fixed a reconnection that could quietly fail for an integration that needs a setup detail up front, like a Confluence site address. If we didn't already have that detail saved, the reconnect was rejected with no explanation. Now you're asked for it again so the connection goes through.

Updated underlying dependencies to keep your workspace secure.

Behind-the-scenes improvements to keep your workspace running smoothly.

Export your analysis highlights and pages, or a single chart or table, to Google Docs, Notion, or Confluence. Connect a destination right from the export step or your workspace settings, and your findings land as a formatted document in the tools your team already works in. No copy-paste, no reformatting.

When you export an analysis highlight or page to Google Docs and later revise it, the Google Doc updates in place, so a link you have already shared always shows your latest version.

• Reports built for you, right in analysis. When you analyze your conversations, the agent now assembles a complete report: written findings, tables, and charts in a single document, so your analysis comes together in one place instead of scattered chart by chart.
• Edit and refine alongside the agent. Comment on any passage or block, ask the agent to revise a section in place, and step through a full version history to compare changes or roll back, so you can shape the report until it says exactly what you mean.
• Publish it as a shareable page. When a report is ready, publish it as a polished page you can hand off, taking you from raw conversations to a shareable insight without exporting anything.

• Switching between workspaces is now noticeably faster, so your new workspace opens right away.
• Setting up Slack notifications through an AI assistant over MCP is more reliable: it can now look up your channels by name and confirms the channel before saving, so notifications reach the right place from the first one.
• Our embed documentation now explains how the inline interview widget looks and how to control it: it renders as a centered card by default, and you can restyle it or have it fill its container.
• The default background behind an interview now uses a warmer tone.

• Fixed a slowdown where your credit balance could intermittently take several seconds (occasionally much longer) to load.
• Fixed an overlap in embedded slider and popup interviews, where the close button collided with the participant's profile icon in the top corner. It's now a single, cleanly placed button.

In interview conversations, messages now stack from the top of the screen instead of stretching apart to fill the height. A short conversation looks like a conversation again, not a couple of messages spread across empty space.

• The welcome card participants see on embedded conversations now stays clean even with a long company name. The name shortens gracefully when space is tight, so the Perspective attribution stays readable and nothing runs off the edge on any screen.
• Opening the design agent no longer flashes a duplicated header and placeholder message bubbles before loading. The real header stays in place while your conversation streams in.

• The appearance controls in embedded interviews now behave correctly. The light and dark toggle works and a participant's choice sticks for their session, and in voice conversations (which are always dark) the toggle is hidden instead of showing a control that does nothing.
• Voice conversations now keep the custom background color set on an embed instead of forcing the background to black, so the conversation matches the page it lives in.

Listing your automations through the Perspective MCP server no longer returns an error when an automation’s trust-score trigger sets a minimum threshold with no upper bound.

Updated underlying dependencies to keep your workspace secure.

Working with the agents that help you design and analyze your Perspectives now feels like one continuous conversation. The messages you send share the same clean, bordered card as the box you typed them in and the agent's replies, with refined spacing that reads beautifully in both light and dark mode. A small detail, the kind we care about, that makes the whole exchange feel more cohesive.

• The Outline got a major redesign. The Outline is the brief that defines what your agent does, and it now reads like a clean, document-style page you can take in at a glance. Everything is grouped into clear tabs, so what your agent covers, how it behaves, and what it captures each have their place.
• Watch it come together live. As Perspective writes your Outline, the section being written lights up, scrolls into view, and the page follows along, so you see your agent take shape in real time. The moment it's ready, you can test your agent right there.
• Refine with confidence. A version history shows what changed, when, and by whom, so you can iterate with the agent and roll back anytime, and always know exactly what your agent will do before it goes live.

The "Checking credits" label in the workspace switcher now shimmers gently while it loads, matching the loading cues used across the rest of the app instead of sitting still.

• The "Try it out" preview you use while designing a Perspective now sizes the conversation to the preview panel's width instead of the full browser window, so it renders its compact layout cleanly in the narrow panel instead of cramming in a desktop-width layout that looked broken.
• When you're designing a Perspective, the control for the image behind your conversation is now labeled "Scene," matching the scene picker and the rest of your design settings.

• The Related Articles section on template pages now shows its images instead of broken thumbnails.
• New Perspectives now get clean, all-lowercase share links by default, consistent with the custom links you can set yourself.

Behind-the-scenes improvements to keep your workspace running smoothly.

Behind-the-scenes improvements to keep your workspace running smoothly.

The area behind your conversation is now a clean, neutral canvas by default instead of a preset backdrop, keeping the focus on the conversation. When you embed a conversation on your site, your background brand color now paints that canvas so it blends into your page, and embed branding options now list only the colors that actually take effect.

Participants who answer by tapping a suggested option are no longer mistakenly flagged as bots. The trust assessment now reads tap-to-answer responses as the genuine answers they are, so real participants keep their scores and you don't lose good responses to false flags.

The profile row at the bottom of your dashboard sidebar now lines up cleanly with the navigation above it, for a more balanced, consistent look.

• When you're designing a Perspective, the "Try it out" preview now works past the first response: the chat input stays visible instead of being pushed off-screen, and the preview now matches what your participants actually see.
• When a participant types their own answer instead of tapping one of the suggested options, their response is now saved and shown in the transcript instead of being marked as skipped, so you never lose what someone actually said.
• When you review a conversation, clicking a citation now highlights the exact message it refers to.

Behind-the-scenes improvements to keep your workspace running smoothly.

Try AI Focus Groups, a fast way to pressure-test an idea before you recruit anyone. Describe a target audience and the question you want to explore, and get three distinct synthetic respondents in seconds, deliberately varied from an enthusiast to a skeptic to a pragmatist, so you hear the full range of reactions instead of one. It is free to try with no credit card, and when you need ground truth the same research runs as AI-moderated interviews with real people at scale.

Conversations have a refreshed, more immersive look. Participants now answer on a clean, focused card, with theme and appearance controls that keep the whole experience on-brand. We are also starting to roll out background scenes you can set behind your conversations. A conversation that feels considered, not like a form, is one more people stay in to finish.

• On the interactive question panel in your conversations, rating and scale questions now fit the screen instead of scrolling sideways. On smaller displays the options wrap onto multiple rows, so every choice stays easy to read and tap.
• While the AI designs your Outline, its progress card now shows a live timer so you can see it is actively working, matching every other tool.
• The profile menu in your dashboard sidebar is easier to spot, with a small indicator showing it opens a menu, and Settings now sits in a consistent place above your profile.

• The Ask Perspective assistant in our docs is more dependable: it answers your question directly instead of pointing you elsewhere, no longer stalls on a slow lookup, and never surfaces raw internal error text. Off-topic questions get a brief, friendly redirect instead of a long apology.
• Workspaces created from names with accents or non-Latin characters now get clean, readable URLs instead of strings of dashes.
• The sign-up dialog on the concierge demo now reads consistently: the title and every button say "Sign up" instead of mixing in "Sign in".

Updated underlying dependencies to patch known vulnerabilities and keep your workspace secure.

Updated underlying dependencies to keep your workspace secure.

When you review a completed conversation, you now see the same closing message your participant saw at the end, including the next step you guided them to. Reviewing a conversation now reflects how it actually wrapped up for each participant.

On the analysis page, the suggested prompts now sit just below the chat input and tuck away the moment you start typing, so the page reads cleanly from the input down to your session history without jumping.

• When your closing instructions ask a conversation to end on a specific final line, the agent now follows them more faithfully. In rare cases it could add its own wrap-up or a second goodbye after your closing line; conversations like these now close just as you wrote them.
• Once a conversation is complete, reopening it can no longer slip in a stray message or trigger another response from the agent.
• When Perspective is removed from a Slack channel, that channel no longer lingers as a dead option in the channel picker, so the channels you can route conversation outcomes to are always ones that still work.

Analysis now works on preview conversations, not just live ones. Since previews are saved alongside the rest of your conversations, every conversation you collect, including the ones you ran while testing, can feed your analysis instead of being left out.

Connecting Slack on a large workspace now reliably finds every channel. The picker used to show a channel count that changed between refreshes and leave out channels you had already invited Perspective into. It now loads your full channel list in the background, keeps it current as channels are added or archived, and lets you add a channel by its ID when search comes up empty, so conversation outcomes route to the Slack channel you actually chose.

• Voice conversations now reliably end the way a call should. In rare cases a call could linger after the agent's closing line, repeating its goodbye instead of wrapping up; now it closes promptly every time, with your transcript and recording saved before the page moves on.
• Voice conversation transcripts now reliably read in the order things were said, with timestamps that match when each person actually spoke. In rare cases a participant's answer could land out of order, after the reply it prompted; reading back a call now always follows the real flow of the conversation.
• Demo conversations you generate now finish with a proper closing message instead of cutting off on the participant's last reply, so they read like a real conversation from start to finish.

Ask the design agent to review your conversation outline and it now evaluates it across multiple dimensions: whether it captures what you set out to learn, whether it follows solid conversation-design practices, and whether it's cleanly structured. You get a score on each, so you can see exactly where the outline is strong and where it needs work, plus specific numbered suggestions you can apply all at once or pick from, like "apply 1 and 3."

Automations that send participant emails through a connected Gmail account are now recorded correctly. They were delivering fine but getting logged as failed runs, which made working automations look broken in your history.

Conversations now open the way a person would: a warm greeting and the first question, instead of an out-of-place reaction before you've said a word. Participants feel welcomed from the very first line.

Embedding a voice conversation in a page builder like Wix or Squarespace? New troubleshooting guidance walks you through fixing the "microphone access blocked" error these wrappers can cause.

When you connect an AI assistant to Perspective, like Claude, ChatGPT, Cursor, or another MCP client, you can now choose which workspaces it can access rather than granting every workspace you belong to. Pick the workspaces when you approve the assistant, and see exactly where each connected app has access from your settings.

Gemini 3.5 Flash is now featured as a recommended model in the research AI chat, so the faster, newer option is front and center when you pick a model.

The entire conversation experience has been overhauled. Powered by the latest models and a warmer, friendlier persona, conversations across text, voice, and phone now feel genuinely human — natural, attentive, and easy to talk to from the first question to the last. The result is a smoother experience that gets participants talking, keeps them engaged, and draws out richer, more honest answers.

Previewing a conversation no longer shows a mismatched header color. Previews look exactly like the live view, with the preview label as the only difference.

Behind-the-scenes improvements to keep your workspace running smoothly.

Behind-the-scenes improvements to keep your workspace running smoothly.

The Dashboard is now called My agents: the place where you keep tabs on all of your agents' activity.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Perspective now reasons through your goal and plans the structure before writing your Outline, so it captures your intent more accurately and follows best practices for asking good questions. You can watch that thinking unfold live as it works.
• Generated Outlines need less cleanup: each section now stays distinct instead of repeating or blending into others the way it occasionally did, so what you get is closer to ready to use.

Your Outline's versions now stay accurate: rebuilding from scratch continues the version count instead of resetting to Version 1, and an edit that changes nothing no longer adds an empty version.

A new AI Focus Groups template runs the same probing, concept-testing focus-group conversation with every participant individually, at scale, with a live demo you can try right on the page.

• Template pages now show what's actually inside each one before you use it: the fields it captures, example questions, and how it adapts and routes, with the most common questions answered inline.
• Security details across template and product pages now reflect our full posture, including SOC 2 Type II and ISO 27001:2022, with a link to the Trust Center.

Template page titles no longer repeat the brand name, so browser tabs and search results read cleanly.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Analysis sessions can now chart your conversation data as line, pie, stacked bar, scatter, and heatmap visualizations, not just bar charts and tables, so trends, proportions, and outliers are easy to spot at a glance.
• You can now create multiple MCP keys and scope each one to specific workspaces, replacing the single all-access token. Connect Claude, Claude Code, or Cursor to only the workspaces a key allows, and revoke keys individually without disrupting your other connections.

When you make a copy of an agent, you can now keep its original owner instead of taking ownership yourself. You're added to the copy as a member, so you keep access either way.

The primary button in your research sidebar now reads New agent instead of New conversation, making it clearer that you're spinning up a new AI agent.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Have a specific Perspective question? Every docs page now has an Ask Perspective button. Click it to ask the Perspective assistant directly and get a guided answer drawn from our docs, so you don't have to read through them to find what you need.
• The Perspective SDK now supports a "push" mode for the slider embed. Instead of overlaying the page, the slider takes its own column and lets the rest of the page stay fully interactive.

Perspective's marketing site, docs, and blog are now properly discoverable to AI assistants and crawlers. Every page advertises pointers to our machine-readable resources, and blog and docs pages expose a clean text version of their content so agents do not have to parse HTML.

Signing up no longer breaks if the welcome interview captures a name with unusual characters. Your name is cleaned up automatically and you land in your workspace instead of getting bounced back to sign-in.

Your workspace view is now a clear list of all your agents (Interviewer, Concierge, Evaluator, and Advocate), each showing its type and whether it's a draft, ready, or active, with a new filter to narrow the list by agent type.

• Two new models for your text interviews: pick GPT-5.5 or Gemini 3.5 Flash from your workspace's text interviewer model setting.
• Text interviews running on Gemini reply faster, with thinking depth tuned to keep reasoning quality high.
• The Concierge designer now pulls brand assets and a website overview as two clear cards when you give it your URL. Brand assets refresh on demand, and the website overview falls back to a live homepage scan when web search comes up empty.

The Slack channel picker recovers from connection errors. After reconnecting Slack and clicking Retry, your channels load and the picker becomes interactive, instead of staying stuck on the error.

Settings has been redesigned with a dedicated sidebar, clearly grouped sections, a one-click way back to your research, and a consistent layout on every page.

• Slack channels deep in the list are reachable again. Channel search now works across your entire workspace, even on workspaces with thousands of channels, so you don't have to scroll or click "Load more" to find one near the bottom.
• Designing a conversation no longer triggers stray "connection timeout" popups, and the hidden Try-it-out preview won't silently start a voice call in the background.
• Text conversations no longer flash a blank frame between the welcome screen and the conversation itself. Your greeting and first reply stay visible the whole way through.

• The embed setup guide now walks you through the tabbed embed panel step by step, so the instructions match what you see when you set up an embed.
• New step-by-step guidance on refining your outline from past conversations, with copyable examples for pointing the design agent at a transcript.

Stop manually writing outline updates. The conversational agent designer can now read your past conversations and propose outline changes based on what actually happened. Ask it to look at a previous conversation, and it reads the full transcript and writes the updates for you.

• Docs answers in the designer chat now start streaming within a second, instead of waiting 10 to 20 seconds for the answer to appear.
• The designer now spends more reasoning effort on the hard parts of your question and moves quickly through the simple parts.
• Embed setup is easier to navigate, with tabs for preview, code, coding agent, and customization. The new coding agent tab ships a copyable prompt that points Cursor or Claude Code at the Perspective MCP server for your real embed snippet.
• You can now send files and screenshots in chat without typing a message, and the dictate and send buttons no longer compete for the same slot.

The knowledge section of generated outlines stays focused. It used to pick up behavioral instructions, form fields, or participant criteria; now it only carries the product, business, or policy facts you provided.

Perspective now supports Gemini 3.5 Flash. Pick it from the model menu in the chat input. Stronger reasoning, a 1M-token context window, and faster output mean the design AI can think harder over longer sessions without making you wait.

Perspective conversations (Interviewer, Concierge, Advocate, Evaluator) sound less like a chatbot in voice and text. They stop narrating what they're doing internally, stop parroting your words back, and stop re-asking for things you've already told them.

• Phone calls flow better at topic transitions. The AI now jumps in with a quick acknowledgment while it puts together a full reply, which should reduce long silences.
• The AI now handles phone calls and text chats differently. On phone, it skips questions that don't translate well to voice, like asking you to spell your email, and uses your phone number from the call itself.
• Speakerphone calls are smoother. The AI no longer mistakes its own voice for you trying to interrupt, so it stops cutting itself off mid-sentence.

• Your dashboard now shows what actually happened on each phone call. You can tell whether the person hung up after wrapping up, went to voicemail, didn't answer, or hit a busy signal, instead of every call looking the same.
• Teammates whose email contained capital letters could be locked out of perspectives they should have access to. That's fixed. Owners of older perspectives can also reliably access their own work.

Your whole team can now edit a perspective together, not just the owner. Teammates and workspace members can change settings, manage automations and webhooks, chat with the design AI, invite more teammates, and mark interviews as test or rejected. Deleting a perspective and changing ownership are still owner-only.

When a workspace is out of credits, teammates who aren't workspace members now see "ask a workspace member to add more" instead of an "Add Credits" button they couldn't use.

The Form Data tab on an interview now shows "No form data extracted from this interview." when no fields were filled in, instead of going blank.

Interview summaries now skip empty form fields, whether you're reading them in-app or in a member summary email. The focus stays on the values participants actually provided.

Shared interview and changelog links now show their preview card again on Slack, X, LinkedIn, and other social platforms.

Buying credits works again on first navigation into the app. The Select Payment Method screen now loads the card form right away instead of showing a blank dialog with a dead Continue button.

• Your Concierge, Interviewer, Advocate, or Evaluator no longer paraphrases the welcome message you wrote in its first message. Participants see your welcome and then get fresh dialogue that builds on it instead of echoing it. On Slack, WhatsApp, and phone conversations, where the welcome message isn't displayed separately, your agent includes the welcome's intent in its first message.
• Behind-the-scenes improvements to voice conversation startup.

Behind-the-scenes security updates to keep your workspace safe.

Spend less time waiting after you hit the Talk button. Voice conversations now connect about 15x faster than before.

Perspective conversations embedded on your site now load about 3x faster, so visitors spend less time staring at a loading state and more time actually talking.

Behind-the-scenes security updates to keep your workspace safe.

• Curious what a Perspective conversation sounds like over the phone? Give our demo line a ring at +1 (559) 785-2453 and have a live voice conversation with our concierge. You'll be able to give your own conversations a phone number soon.
• Speak instead of type. The dictation mic now shows up on every chat input across Perspective, so you can dictate your prompts anywhere you'd normally type one.

Voice conversations sound more natural. The AI no longer slips in awkward filler before it answers or acts, and it stays in the same language for the whole conversation.

Slack messages sent by your automations now display their full content in the conversation's activity: text, attached blocks, button labels, and emoji all show correctly. Before, the activity often said "No message content".

Voice conversations now use a smarter realtime model with a context window four times larger for longer, more coherent back-and-forth. Live transcripts run on a new streaming-tuned model for sharper, lower-latency captions.

Voice interviews on iOS no longer get stuck in a non-scrollable state on first load, and the animated background now stays pinned to the screen instead of drifting with scroll.

Behind-the-scenes security updates to keep your workspace safe.

When the design chat agent picks a research methodology, the response now wraps cleanly on narrow screens instead of breaking into squeezed columns.

• Long research questions are faster to say than to type, and you can ramble through them without compressing your thought into something easier to type. The new mic in the design chat input (or press Cmd/Ctrl+Shift+D) transcribes you in real time and splices the words in at your cursor, so a long prompt takes the time it takes to speak it.
• Ask "how does X work in Perspective?" inside the design chat instead of opening another tab to dig through the docs. The assistant now searches the public docs and FAQ when you ask product questions and cites its sources inline, so you can verify or jump to the full page in one click. Docs citations link to the source page; FAQ citations expand into a popover with the question and answer.

Refreshed the docs across collect, design, and settings so guides match the current product.

Improved request reliability so the app no longer occasionally stalls on the first request after a period of low traffic.

Behind-the-scenes security updates to keep your workspace safe.

Grok 4.3, xAI's latest reasoning and vision model, is now available across the AI chat and analysis flows, and as the default model for data extraction. A faster low-reasoning variant is available for text interview deployments where response latency matters more than depth. Existing sessions on the previous Grok 4 generation transparently route to Grok 4.3.

The research analyst chat now keeps time-based horizontal bar charts in chronological order. When the AI charts data across weeks, months, or sequential stages, the bars stay in order instead of being re-sorted by value.

Behind-the-scenes security updates to keep your workspace safe.

Perspective MCP responses now include each perspective's internal name alongside its public title, so AI agents can identify perspectives by the label you recognize in the app.

Monospace text across the app now renders in DM Mono, giving code blocks, keyboard hints, and similar elements a consistent typeface that pairs with the existing DM Sans body font.

• HubSpot form automations now correctly link each submission back to the participant's anonymous browsing session, restoring page-view and source attribution on the resulting HubSpot contact.
• Perspective MCP outlines that include navigation steps with templated redirect or submit URLs now load reliably in strict clients like Claude Desktop, instead of failing with a tool execution error.

• Redesigned AI tool calls to look more consistent. Generating outline, classifying research, generating a greeting, evaluating an outline, saving an automation, picking a Slack channel: every action now flows through one consistent card with crisp loading, success, error, and stopped states, expandable details, and inline actions that keep the next step one click away.
• Results and Interviews dashboards load noticeably faster, especially on busy workspaces.
• Workspace credit status is now visible to everyone with access to the workspace, not just admins. The plan label and credit chip sit side by side in the workspace switcher, with clear "Checking credits" and "Credits unavailable" fallbacks so the chip never silently disappears mid-load.
• Refreshed the webhooks guide and automation recipes to match the product as it actually works today: the real event names, the real payload shape, the custom auth header model, idempotency keys, retry behavior, and the trust-score filter gotcha that silently excludes abandoned interviews.

Entry redirects on the Try Concierge page now preserve every query parameter, even when they contain encoded characters.

Sign-in popups in embedded conversations now load reliably on Safari.

• Perspective MCP tools now ship with complete behavior hints (read-only, open world, destructive), so MCP clients can show clearer confirmations before running each tool.
• Behind-the-scenes improvements to keep your workspace running smoothly.

Behind-the-scenes improvements to keep your workspace running smoothly.

• A polished new reading experience for blog posts. Long-form articles now feature an "In this article" table of contents that anchors directly to each section, one or two context-relevant product suggestions woven into the reading, and a curated list of related reads in the same category at the end. Section headings click-to-anchor cleanly under the top navigation.
• The blog home is now Field notes on conversational AI. The index reads like an editorial publication, with the same category bar and post grid you already use.

Shared interview link previews on social platforms now reliably show the host's avatar.

Preview interviews now unlock automatically when you add credits to your workspace. Previously, previews you started before purchasing a credit pack could stay locked even though credits were available.

Perspective is now agent-ready. AI assistants and search crawlers can discover what's available on Perspective's site (including the Perspective MCP server), and every blog, docs, and changelog entry is now also available as a plain markdown version that AI tools can fetch and cite directly.

• You can now add a description to each workspace in workspace settings (or when creating a new one). AI assistants connected through Perspective MCP use what you write to pick the right workspace for the task, so be specific about what each workspace is for ("growth research" vs. "support feedback") instead of relying on the name alone.
• The Slack channel picker on the invite page now loads instantly even on large workspaces.
• Thirty-six new posts on the blog covering AI customer engagement, research, insurance, intake and lead qualification, onboarding, customer success and churn, voice of customer, product discovery, and surveys.

• Slack integrations are now more reliable: invites and automations consistently use the active workspace connection, including right after reconnecting a Slack workspace.
• The trusted-companies logo strip on the homepage and the comparison pages now lines up with the rest of the page content on wide screens.

• Each use-case page now reads like a guide to solving a specific business problem with Perspective: leading with the value, walking through real features alongside photos and integrations, and pointing to a clear next step.
• You can now add Perspective MCP to Claude.ai on the web as a custom connector. The setup guide has the deep link and the exact fields to fill in.
• Twenty new posts on the blog:
  • Customer success and churn:
    • AI for Customer Success Is Stuck on Dashboards. The Real Unlock Is Conversations.
    • Customer Success Automation in 2026: The 4-Layer Stack Every CS Org Needs
    • How to Identify At-Risk Customers Before They Churn (A 2026 Playbook)
    • How to Reduce Customer Churn in 2026: A Modern SaaS Playbook
    • Churn Prevention Software in 2026: A Buyer's Guide to the Right Stack
    • Customer Churn Prediction with AI: Why Models Alone Aren't Enough in 2026
  • AI conversations:
    • AI Feedback Collection: From Static Surveys to Conversations That Actually Tell You Something
    • AI vs Surveys: Why Conversations Win for Real Customer Research
    • AI Qualitative Research: A Practical Guide for Modern Research Teams
    • Replacing Forms with AI Chat: When, Why, and How to Make the Switch
  • AI-native customer engagement:
    • What "AI-Native Customer Engagement" Actually Means (And Why Most Vendors Get It Wrong)
    • AI Customer Engagement Tools in 2026: A Buyer's Roundup by Use Case
  • Intelligent intake and vertical playbooks:
    • AI-Native Onboarding Software: What to Look For in 2026
    • Best AI Tools for Insurance Brokers in 2026: A Practical Roundup by Workflow
    • AI for Insurance Agencies in 2026: From Lead Capture to Renewals
    • AI for Real Estate Agents in 2026: A Practical Playbook for Top Producers
    • AI Client Intake for Law Firms: How to Replace PDF Intake Forms with AI Conversations
  • Product discovery and UX research:
    • UX Research at Scale: How AI Interviews Break the Researcher Bottleneck
    • User Interview Software in 2026: A Comparison Guide for Modern Research Teams
    • AI Product Feedback Tools in 2026: A Buyer's Guide for Product Teams

Working with Perspective MCP from Claude, ChatGPT, Cursor, and other clients is more predictable now: conversation totals match between tools, statuses come back in the same casing everywhere, invalid date filters surface a clear error instead of being silently ignored, and looking up a missing invite returns one predictable message. AI clients also pick the right tool for the right agent more reliably, since tool descriptions now reflect the four agent types: Interviewer, Concierge, Evaluator, and Advocate.

You can now connect Claude, ChatGPT, Cursor, VS Code, and other MCP clients to your Perspective workspace through standard OAuth, with no more pasting access tokens by hand. When an app asks for access, you grant or decline on a clean consent screen with the requested permissions spelled out in plain language. A new Connected Apps section in your workspace settings shows every authorized app and lets you disconnect any of them in one click. See the MCP setup guide.

Polished how Perspective MCP exposes automations to AI agents. Managing automation rules now uses dedicated list, create, update, delete, and test tools instead of a single catch-all, and tool titles and descriptions are sharper across the surface, so agents calling them behave more predictably.

• The context window indicator in research and analysis chat no longer flickers off the moment you send a new message. It now keeps the last known size visible until the next reply lands.
• Smoothed several inconsistencies across Perspective MCP tools: searching perspectives by title finds matches even when a custom display name is set, conversation totals match between the get and stats tools, the integrations tool recognizes every connected provider including Gmail, and listing conversations now signals whether more results exist alongside the cursor.

• Welcome chat now surfaces the interactive quick-reply panel and a Talk button. Participants can answer single choice, multi choice, and scale questions right in the welcome flow, or jump straight to a voice conversation without typing.
• When you create a new agent by uploading a brief, form, survey, or playbook, the system now reads the file alongside your text prompt to pick the right agent type for you.

• Interactive questions now support more options. Single and multi choice questions can show up to 10 choices, and scale questions can span up to 20 points, so the AI can present richer panels when the question calls for it.
• Copy-paste embed snippets now include a Perspective AI attribution comment above the script tag, so AI crawlers that can't execute JavaScript still see the attribution.

• HubSpot per-interview submissions are reaching HubSpot again after a brief regression, and the activity timeline now accurately shows failed submissions instead of masking them as successful.
• Brand logos in the interview header are now sized to be legible on both mobile and desktop. Configurations with a single uploaded logo keep a subtle backdrop for contrast; multi-logo and themed configurations drop the invisible frame so the logo has room to breathe.

Perspective MCP is now more reliable and handles long, multi-step perspective creation sessions without hitting tool timeouts in Claude Desktop and other MCP clients.

• The mobile navigation menu now adapts to your sign-in state. Signed-in users go straight to their dashboard, and signed-out users land in the sign-up flow.
• Polished the interactive question panel in text conversations. Hover and keyboard focus now share one selection ring so mouse and keyboard navigation stay aligned, scale endpoints read as labeled captions beneath the row, and the panel border highlights when focus moves inside.

Internal links in blog and docs posts now open in the same tab; external links continue to open in a new tab.

The homepage, blog, docs, and marketing pages are now blazingly fast. Pages load instantly from cache and stay smooth as you scroll and interact. You'll feel the difference the moment you land.

Behind-the-scenes improvements to keep your workspace running smoothly.

Interactive question panels: AI conversations can now present participants with single choice, multi choice, and scale/NPS questions as native UI panels instead of plain text. Includes keyboard navigation and "something else" free-text option. Messaging channels (Slack, WhatsApp) fall back to quick-reply chips.

Behind-the-scenes improvements to keep your workspace running smoothly.

New blog post: "The Glasswing Principle: Why Your Customer Feedback Tools Have the Same Blind Spot": a fresh take on the limits of today's AI feedback tools.

• Restored the Use Cases page to show only the original 25 curated items instead of all 181 templates.
• Lighter breadcrumbs for snappier page loads.
• Faster post-login redirects with a leaner workspace lookup.

• Long-running analysis chats no longer break. Sessions are now durable: you can close your browser while analysis is running, come back to finished results, and resume an in-progress analysis from any device.
• Faster navigation across results, design, and conversation pages.
• SDK 1.7.0: auto-injected attribution signals so AI crawlers (GPTBot, ClaudeBot, PerplexityBot) can discover Perspective AI on your site. Enabled by default with a one-line opt-out.

• Faster results page navigation.
• Performance tuning across conversations, automations, and analysis sessions.
• Better citation chip contrast in light mode; preview conversations are now included in redirect count checks.
• Now serving llms.txt to follow the emerging convention for AI-discoverable content.

Routine security and dependency updates to keep your data safe.

• Template gallery expanded to 181 pages with strong search visibility and a clean category system.
• New conversation-based template cards with dynamic previews and avatars.
• Templates now appear in the global navigation (header, mobile, footer).
• Slack profile context is now included in AI prompts for richer participant context.

• Redesigned templates gallery with sidebar categories and grouped sections.
• Template detail pages and embed previews use the live SDK widget for a richer experience.
• Simplified citation chips in analysis sessions.

• Cleaner whitespace in participant nudge emails.
• Faster template search and general UI polish across cards, sidebar, and embeds.
• Only one embed is active at a time when the preview modal is open.

Analysis now works across hundreds of interviews

• The assistant runs qualitative theme discovery and quantitative pattern analysis in the same conversation.
• Search: Semantic search across all interview transcripts. Ask natural language questions and get answers with citations linking back to specific participants.
• Data analysis: Quantitative queries across your structured output fields (counts, distributions, trends, cross-tabulations).
• Deep exploration: Full-text search across all transcripts to surface patterns, specific quotes, or emerging themes that structured fields miss.

• Citations now show participant initials and first names as clickable chips.
• The assistant refreshes its knowledge of your interviews on every turn, staying current on participant counts, date ranges, and metadata.

• Trusted per-invite participant context: invite links can now carry verified context (URL parameters, invite metadata, Slack profile) that's stored securely and merged into the participant's session.
• Increased URL parameter value limit from 20 to 50 characters.
• The invite tool now validates context and prompts you when something is missing.

• Legacy metadata entries and empty values are now handled safely.
• Empty metadata wrappers are no longer emitted when there are no valid entries.

Routine security and dependency updates to keep your data safe.

• Faster interview research page loads thanks to leaner data fetching.
• SDK 1.6.1: fixed avatar padding on the float launcher bubble.

Automation test sends (digest, per-conversation, invite) now correctly deliver to the test recipient instead of the configured recipients.

Routine security and dependency updates to keep your data safe.

New text conversation models: GPT-5.4 (mini, nano) and Claude Haiku 4.5 for faster, cheaper AI conversations.

Routine security and dependency updates to keep your data safe.

Analysis session visibility: admins can hide sessions from customers with an admin-only / everyone toggle. Admin-created sessions default to hidden, with instant UI feedback and direct-URL access enforcement.

• Three-tier conversation completion handling: conversations are now classified as complete, partial, or abandoned, with the right depth of AI processing for each tier. Slack notifications get colored sidebars and emails adapt to status.
• Completion source tracking: manual, participant, and no-activity completions each get distinct messaging. Background processing now prevents duplicate runs.

Routine security and dependency updates to keep your data safe.

• Voice instructions: researchers can customize voice accent and speaking style under Settings > Voice & Tone or through design chat. Applied in voice mode only.
• Persisted embed settings with a redesigned accordion UI: appearance, launcher, and auto-trigger options now save to your workspace and can be updated through design chat with an explicit publish step.
• SDK 1.6.0: applies your saved embed settings automatically (taking precedence over inline embed code) and replaces the loading spinner with a polished skeleton UI.

Routine security and dependency updates to keep your data safe.

Voice conversations no longer end prematurely when a participant asks for time. The AI now treats pauses as natural moments to wait, not signals to end the call.

Routine dependency updates.

• More reliable markdown link rendering across the marketing site.
• Aligned footer styling with top navigation for visual consistency.
• Logo marquee now scales more responsively across screen sizes.

• Restored links that were incorrectly blocked by overly strict link validation.
• Fixed image aspect ratio warnings in the logo marquee.
• Restored the missing footer top border.
• Silenced a smooth-scroll console warning during page transitions.

• Blog and docs now feature syntax-highlighted code blocks, copy buttons, and improved heading anchors aligned with the table of contents.
• React SDK documentation added to the Embed API reference.
• 17 new long-form blog posts covering Intelligent Intake, AI customer research, AI conversations at scale, and product discovery and UX research.

• Consistent agent styling: unified colors and variants across dropdowns, chips, and buttons, including per-agent colors in suggestion chips.
• Stronger search coverage with internal linking across the template gallery and cross-linking between blog posts.
• Better accessibility: aria labels and focus states for heading anchors.

• Agent colors now render correctly in dropdown menus.
• Fixed mobile overflow issues in docs pages and enabled horizontal scrolling for code blocks on mobile.
• Resolved duplicate anchor and override issues in heading links.
• Cleaned up blog content artifacts: unicode escaping, truncated descriptions, and incorrect template mappings.

SDK 1.5.1: float button now expands to fullscreen on mobile.

• Voice conversations no longer time out on long sessions, thanks to a leaner conversation prompt.
• Conversations can now resume when a participant switches back to text after ending voice mode.
• Fixed an iOS welcome chat scroll overlap on small screens.

• Rewrote the Typeform alternatives blog post for stronger search visibility, with refreshed metadata, internal links, and updated pricing and positioning copy.
• Perspective AI now ranks #1 in the Typeform alternatives list with a unified comparison view.

New global error page preserves URL query parameters on reload, so tracking and redirect context aren't lost after recovery.

• Docs reorganized around what you're trying to do: Overview, Design, Collect, Analyze, Settings, with a new "How Perspective Works" intro, agent-based framing, and core guides covering all four agent types.
• AI-native doc actions: every docs page has a "Copy page" dropdown to export as Markdown or open it in ChatGPT, Claude, or Gemini with a prefilled prompt.
• Expanded developer docs: API overview, webhooks, automation recipes, and Embed API reference.

• Redesigned docs homepage with a quickstart CTA and a two-column layout; sidebar now uses shorter navigation labels.
• Refined CTA button styling: consistent outline design with improved spacing, hover states, and sizing.
• Stronger search visibility: structured data for articles, how-tos, and videos, plus accurate sitemap priorities and last-modified dates.

• Docs correctness: fixed broken links, typos, API values, and webhook auth docs; updated Outline docs to match the actual UI.
• AI copy/export: prevented duplicate page titles and corrected title handling.
• Date and timezone fixes for search-engine and sitemap output.

• Outline evaluator: new tool reviews research outlines with full context and streams its reasoning and evaluation as it works.
• Selected tags now show on conversation cards and the details page, backed by a centralized tag vocabulary and reusable badges.

• Claude Sonnet 4.6 context window increased from 200k to 1M tokens, supporting much larger inputs.
• Faster, smoother voice-to-text handoff: mode switching is now instant with no blocking revalidation between modes.

Voice dark theme is now scoped to the voice experience, preventing premature theme changes elsewhere.

• HubSpot automation submissions no longer silently fail: the portal is now resolved automatically from your connected account, integration errors are detected properly, and bot-protection warnings now surface to users.
• Partial-conversation polish: stale processing records no longer crash, cleanup runs in the background for lower latency, and the progress bar no longer shows 100% for partial conversations.

Routine security and dependency updates to keep your data safe.

AI conversations now use correct grammatical gender for self-reference in gendered languages like Hebrew, with voice mode picking gender automatically from the selected voice.

AI streaming now reports errors consistently so failures surface instead of silently dropping.

Switched to a more stable streaming transport to prevent crashes from malformed responses.

Voice conversations now use dark mode from the very first screen, including welcome and sign-in, for a consistent, polished feel across previews and embedded dialogs.

Contact and newsletter forms now use the standard embed widget so theming and host handling stay consistent.

• iOS conversations no longer auto-scroll unexpectedly when the address bar or viewport shifts; user scroll position is preserved.
• Custom slugs with hyphens (e.g. /contact-us) now load correctly instead of failing to authorize.

Claim flow now supports multiple workspaces via an allowlist instead of a single hardcoded workspace.

Blog and docs now render from local content for faster page loads and reliable static generation.

Claim flow now shows a clear error state when no workspaces are available instead of hitting a dead end.

• Idle interviews now resolve automatically into Complete or Partial after a configurable timeout per research, so abandoned conversations stop hanging in progress.
• Participants can resume or restart abandoned and ended interviews across web, Slack, and WhatsApp with a clear Continue or Start Over choice.
• New per-research setting to remove Perspective branding across the interview UI, emails, embed sign-in, and previews.

• Refined messaging, timing, and researcher insights for partial interviews (previously called "abandoned") across email and Slack.
• AI tool error states now use a calmer neutral style instead of alarming red.

Branded email previews now reflect the remove-branding setting immediately instead of showing stale content.

• New participant invite tool generates magic-link invites and can email them directly.
• Float embeds now support launcher customization: pick an icon (or branded avatar), size, and shape, with live preview and updated embed snippets.
• Centralized FAQ system with structured data across homepage, pricing, compare, and agent pages for better search visibility.

• Redesigned embed Visual Options panel for clearer configuration.
• Embed config now exposes a branded avatar URL for launcher icons.

• Float embed preview bubble is now fully visible thanks to a taller preview frame.
• FAQ links no longer show a stray arrow icon, and CTA tracking now uses referral parameters.

A new claim link lets users claim a pre-built research perspective and copy it into their workspace through the standard sign-in flow.

• The claim flow now restricts cloning to workspaces you have access to.
• The interview footer no longer mismatches the surrounding theme.
• Pasted or uploaded attachments now stay visible even when the input is empty.
• Long attachments no longer cause horizontal page scroll in the interview view.
• Social link preview images now render correctly for image URLs without a file extension.

Routine security and dependency updates to keep your data safe.

New scrolling logo strip on the homepage, compare pages, and concierge trial page: 26 trusted company logos in a slow continuous marquee, with dark mode and reduced-motion support.

• Voice assistant no longer responds to its own audio output during conversations.
• Reduced transcription errors in noisy environments by tuning the transcription model and handling empty transcripts more gracefully.
• Fixed a small visual jump when the logo marquee loops.

Voice connection timeout extended from 10 to 30 seconds so slower networks don't drop calls during setup.

Routine security and dependency updates to keep your data safe.

• Voice calls stay connected through quiet moments — keep-alive pings now happen every 30 seconds instead of every 2.5 minutes to prevent idle timeouts on the upstream voice provider.
• Final voice events are no longer lost during call shutdown.

Redesigned interview status pages: polished cards replace the old emoji-based error screens, and invalid URLs now redirect to a dedicated marketing page instead of a dead end.

The top nav, sidebar, and footer now show the full set of roles and industries from the central use cases directory instead of a hardcoded subset.

"My perspectives" filter now correctly returns research for users with multiple emails, and deleted items disappear from the list immediately.

• New URL input on the comparison pages with validation and a responsive layout.
• Submitting a URL on a compare page now flows directly into the concierge experience, preserving attribution context through signup.
• Added a Compare link to the footer for better discoverability.

• Compare pages now match the design system: shared typography, theme colors instead of one-off values, and a simpler hero with headline, subline, and URL input.
• Refined comparison tables and competitor messaging across all compare pages, with stronger positioning of AI conversations vs forms.

• Interview chat no longer auto-scrolls to assistant messages before they have visible text, eliminating jumpy scroll behavior.
• The sticky footer now has an opaque background so message text doesn't bleed through underneath.
• Signed-in users who land on the sign-in or signup page without a redirect target now go to the research dashboard instead of the homepage.

Routine security and dependency updates to keep your data safe.

• Send participant email automations directly from your connected Gmail account, with built-in connection, validation, and permission setup.
• New sign-in banner above the chat input for authenticated conversations, translated across 18 languages.
• "Sent via Perspective" inline attribution appears when the email footer is hidden in branded Gmail mode.

• Sign-in-first welcome flow now requires authentication before collecting user info, with refreshed calls-to-action and translations.
• Participant email templates now render list fields cleanly as checklists, bullets, or comma-separated values based on context.

• The welcome flow now clears the sign-in prompt after authentication completes.
• Restored reply-to addresses on participant emails delivered by Perspective.
• Fixed missing accent marks in Spanish, Portuguese, Polish, and Turkish translations.

Routine security and dependency updates to keep your data safe.

• Docs now support deeper category nesting, with breadcrumbs, sidebar, and prev/next navigation that traverse the full tree.
• Docs categories without their own page now show a list of their child pages automatically.
• The marketing homepage now shows for everyone instead of redirecting signed-in users to the research dashboard.
• Simplified the mobile menu by removing the non-interactive Channels and Automations sections and giving "Get Started" a more prominent button style.

• Fixed invisible UI elements across marketing pages — badges, gradients, workflow diagrams, feature cards, and story-player effects — caused by bad theme color values.
• Fixed step indicators and connector lines on use-case detail pages so they render and stay aligned in their column.

Concierge email buttons no longer contain a malformed URL segment, so links now route correctly across all three concierge email types: created, no form found, and error.

• Redesigned homepage with a new narrative emphasizing the shift from forms to AI conversations, plus refreshed Problem, Solution, Benefits, Use Cases, and Automation sections.
• New Studies and Tutorials entries in the top navigation.
• New visualization elements for product demos: leaderboards, tag clouds, insight bubbles, and insight cards.

• Docs sidebar now uses a collapsible category layout with icons, faster preloading, and smarter highlighting of the current page.
• Standardized marketing calls-to-action and FAQ sections across landing pages for visual consistency.
• Faster page rendering thanks to parallel data fetching and removal of duplicate blog API calls.

• Fixed unreadable button text in dark mode and several interaction bugs in the docs sidebar.
• Fixed a brief flicker on some pages and a layout bug in the calendar component.

• Native Slack interview thinking and streaming, so participants see responses appear in real time.
• Channel @mention support: interviews now work in channels with the same active-conversation handling as direct messages, plus a branded call-to-action when no conversation is active.
• Slack file-share replies with text are now accepted (the attachments themselves are ignored).

• Slack mentions and links are preserved through automation and interview replies, including the final streamed message.
• Restarting a Slack interview now reuses the same interview but starts a fresh direct-message thread, cleaning up the old one.

Routine security and dependency updates to keep your data safe.

Embedded anonymous interviews now restore session state correctly when the embed reloads, so participants skip the welcome screen even when browsers block third-party cookies.

Routine security and dependency updates to keep your data safe.

• GPT-5.4 model support Added OpenAI's GPT-5.4 with a much larger context window, plus reasoning and vision. Now the recommended model in the model selector.
• Context usage indicator in chat A circular progress indicator in the chat input shows how much of the model's context window your conversation is using, with a tooltip for details.
• Voice interview controls Added mute/unmute and end-conversation buttons, a new "Left Early" status for participants who hang up, and clearer visual states for voice input.
• Stronger email verification Sign-in code throttling is now per-email and more reliable, with clearer feedback when codes are rate-limited and more secure code generation.

Preview interview sign-in When an existing session conflicts with preview interview sign-in, you now get a warning before the session is cleared.

Voice reconnect stability Voice connections no longer get stuck during startup or reconnects, with bounded retries, clearer error messages, and reliable recovery from connection failures.

• Interviewer now asks "anything else?" before the summary confirmation, so participants get a chance to add anything they missed.
• Tightened the closing flow so the interview only ends after all closing steps are complete, and the final message is always a statement rather than a question participants can't answer.

Prevented interviews from ending prematurely. Previously the closing message could contain a question participants had no way to respond to.

Voice interviews now wait for the assistant's closing message to finish playing before completing. Previously the connection could drop mid-goodbye, cutting off the final words to participants.

• Sign-in required interviews in embeds Authenticated interviews now work inside embedded iframes even when browsers block third-party cookies, using a sign-in popup with a new-tab fallback if popups are blocked.
• Auto-captured participant emails When participants share their email during a conversation, it's now captured automatically so thank-you automations work without extra setup.
• Compare hub page A new index page linking all competitor comparison pages, with a hero, stats, and category-grouped logo cards.

Concierge landing page polish: cleaner hero state, better typing indicator, and consistent text sizing in chat messages.

• Quick reply chips are now readable in light mode.
• Email button links no longer include a doubled domain.

Routine security and dependency updates to keep your data safe.

• Scheduled invite automations Send recurring invitations to participate in research conversations via email or Slack — perfect for daily standups, weekly check-ins, pulse surveys, sprint retros, and ongoing feedback loops.
• AI-composed invite messages Each invite is generated with the share URL and tailored copy, plus an interactive Slack channel picker in the assistant chat.
• Auto multi-conversation mode Invite automations automatically allow multiple conversations so each invitee can participate.

• Test mode for participant email automations now redirects to your override recipient instead of silently failing.
• The research assistant now offers to test automations in the first setup step, matching the later steps.

• Comparison pages New pages comparing Perspective AI to Typeform, Formstack, Qualtrics, SurveyMonkey, Google Forms, Jotform, and Medallia — each with a URL converter, feature table, FAQ, testimonials, and pricing details.
• Embedded newsletter signup Newsletter signup is now an embedded Perspective AI conversation instead of a static multi-step form.

• Outline generation restored Fixed errors that had been silently breaking outline generation for around a week. Outlines now generate reliably across all models.
• Improved dark mode contrast and accessibility on comparison page tables.

• No more duplicate interviews Background sign-in refreshes no longer restart interviews, so participants won't get duplicate sessions or restarted voice conversations.
• Outbound voice calls wait to greet The AI no longer speaks immediately on outbound calls — it waits for the recipient to speak first, which feels more natural for voice conversations.

No more corrupted chat messages Added a guard against rapid double-clicks and overlapping send-or-regenerate actions across all chat surfaces, preventing message corruption.

• Slack link previews restored Perspective AI URLs now show proper previews when shared in Slack — including blog posts, pricing, and the homepage — instead of appearing as plain links.
• Preview deployment URLs now generate link previews correctly too.

Routine security and dependency updates to keep your data safe.

• SEO overhaul across marketing pages Added social sharing metadata, structured data, and canonical URLs across the homepage, pricing, about, use cases, products, roles, and landing pages so they surface better in search and link previews.
• Cleaner page titles and descriptions Removed duplicate brand names, rewrote use-case titles to be keyword-targeted, and trimmed long titles and descriptions to fit search result limits.
• About page heading hierarchy Cleaned up the page heading structure for better readability and SEO.
• Renamed the blog from "Your AI Customer Compass" to "AI Customer Research Blog" for better discoverability.

Automation errors now display readable messages instead of empty objects.

Embeds now load your allowed channels and custom welcome message from the server, so they render channel-specific UI and show your welcome text out of the box.

Smarter auto-redirect Auto-redirects after interview completion now use the same URL handling as manual clicks for consistency, including search parameters and environment swaps.

• Welcome wizard timing The welcome screen now waits for participant info to be ready before processing, so fields no longer go missing on completion.
• Broader compatibility for participant data: common alternative field names for first name, last name, company, company size, and email consent are now recognized automatically.

• Richer blog link previews Blog posts now include structured metadata, social share images, and improved Twitter card details so they render beautifully when shared.
• Site overview for AI crawlers A new structured site overview, dynamically generated from published blog content, helps AI search tools index our content accurately.

• SOC 2 and ISO 27001 badges Added to the footer to surface our security certifications.
• Cleaner version list in the dropdown with proper scrolling and no viewport overflow.

• New landing pages for Intelligent Intake and CX Teams, each with hero, features, how-it-works, impact, use cases, FAQ, and CTA sections.
• Added interactive demos for modal previews, animated conversation playback, embed mockups, and routing, insights, and intent analysis visualizations.

Conversation redirects now consistently carry the conversation reference, so context is preserved whether the redirect happens early or at the end of the flow.

Fixed early navigation redirects dropping the conversation reference, which previously caused inconsistent landing experiences.

• Auto-trigger embed popups Popups can now fire automatically based on a delay or exit intent instead of requiring a button click, with frequency controls (always, per session, or per visitor) and code generation for Script, NPM, and React.
• Workspace invite roles When inviting teammates, you can now choose between Admin and Member with role capability descriptions inline.
• Faster voice model Upgraded to a new realtime voice model: better instruction following, +10% transcription accuracy, and improved tool calling at the same price for voice conversations.

• Reliable interview goodbyes Interviews now consistently deliver the closing message instead of skipping the farewell or showing raw tool output.
• Lower latency for AI responses Streaming responses arrive faster across interviews, research, and analysis, with reduced time-to-first-byte in multi-step tool flows.

Routine security and dependency updates to keep your data safe.

• Two-step conversation ending Conversations now end in a clean two-step sequence: the model marks the session complete, then delivers its goodbye to the participant, so sessions no longer get stuck open and farewells are reliably shown.
• Restored an additional model option for conversations now that the new ending flow handles its constraints.
• Right-to-left languages now flow correctly through entire action card headers.

Prevented sessions from getting stuck when models output internal commands as raw text. Participants now consistently see a proper goodbye at the end.

Routine security and dependency updates to keep your data safe.

• Per-agent prompt configuration Each agent now owns its own role, objectives, engagement rules, and methodologies. Cleaner separation makes agent behavior more consistent and easier to evolve.
• More reliable session endings Recovered cases where some models occasionally failed to end sessions properly, so conversations now close cleanly across more model variants.
• Clearer session lifecycle Models now reliably distinguish between wrapping up a conversation and ending the session, so sessions end when they should.

Embed integrations now correctly fire navigation callbacks instead of silently dropping them.

Standardized color usage across the app for cleaner visuals.

• Restored missing animations and theme accents that were lost in the recent design system update.
• Attachments now hide when voice mode is on and the input is empty.
• Mobile attachment button no longer renders when there isn't enough space.

• Run conversations over WhatsApp as a second messaging channel alongside Slack, with secure delivery, per-recipient status, post-completion feedback buttons, and clear error messages for participants.
• Added support for a new high-context reasoning model with a 1M token context window and configurable thinking depth, now the recommended choice.

• More responsive thinking state across AI chat and conversations, with the logo morphing into bouncing dots during reasoning.
• Refreshed styling across research, outline, admin, and marketing surfaces with a faster, more consistent design system.

• Fixed reasoning steps not rendering for the new high-context model.
• Fixed WhatsApp invite ownership and feedback confirmation consistency.

• Slack DM conversations Participants can now complete conversations directly in Slack DM threads, with branded DMs, channel broadcast announcements, and owner mentions in the invite.
• Slack channel option in invite wizard A flatter invite wizard with a new Slack channel option, permission checks, and a reconnect flow when access needs to be refreshed.
• Smarter Slack experience Permalinks open the Slack thread panel, hints mention participants, completed conversations get a thread reply, and every invite has a DM button.

• Automatic context trimming Long conversations now have history pruned automatically as they grow, preserving the last successful outline.
• Google One Tap is now hidden when a conversation doesn't require sign-in.

Conversations in Advocate now wrap up smoothly instead of cutting off abruptly.

Fixed a small bug in model selection.

• Global admins can now view, initiate, cancel, and send reminders for workspace ownership transfers.
• New call-to-action closing flow for Advocate conversations: a single CTA response instead of a summarize-and-confirm step.

• End-of-conversation text (thank-you, loading, CTA, redirect) now renders in the conversation language across 17 languages (English, Spanish, French, German, Hebrew, Arabic, Chinese, Japanese, Portuguese, Russian, Italian, Dutch, Korean, Polish, Turkish, Swedish, Hindi).
• Redirect destinations are now visually highlighted to make the target site clearer.
• Switched the default and recommended model to a faster, more capable option, with older model selections automatically mapped forward.

Routine security and dependency updates to keep your data safe.

• Outline settings UI refresh: renamed "Brief" tab to "Overview", tighter typography and spacing, scrollable sections, and a single unified "Add" button for empty sections.
• Outline versioning simplified: version dropdown now shows version numbers and "Latest" without per-version metadata clutter.
• Hide outline "Add" buttons while the chat is loading or streaming to prevent conflicting actions.

Research settings now correctly accept values that exactly meet the minimum length (e.g., 3 or 10 characters) for Title, Welcome Message, Brand Name, and other fields.

• New Advocate research type with dedicated prompts, methodology guidance, and a tailored experience for persuasion, objection handling, and message playbooks.
• Workspace-level model setting for text conversations, with owner and admin controls in Settings.
• Outlines now have a Knowledge field for structured factual context required during conversations, and can read attached PDFs and images.
• Outline change tracking shows who changed what and when, with version metadata and key differences in the editor.
• Refreshed model lineup with updated context windows and pricing, including transcription and text-to-speech.

• Updated conversation and research assistant prompts for advocate support and stricter knowledge boundaries.
• Research classifier now appears only when multiple research types are available for the chosen agent.

• Payment provider sync now uses the workspace owner's email instead of an admin's.
• Custom-domain surveys now fall back through a chain of detection methods, with AI as a last resort.
• Added support for multi-step survey navigation, including SurveyMonkey.
• Broken brand favicons are now cleared instead of retried indefinitely.

• Participant emails now show your brand logo, colors, and signature when brand mode is enabled, replacing the purple top bar with a clean, white branded design.
• Brand favicons now save reliably so they always render in admin previews and emails.

• Up to 85% faster first response after the opening message, thanks to smarter caching.
• Faster conversation startup: research data now loads in parallel with session lookups.
• Branded CTA buttons now pick light or dark text automatically for better contrast on any background color.

• Switching between voice and text no longer drops in-flight messages, with smoother transitions.
• Fixed stuck "Preparing your summary…" state when a reflection finishes without content.
• Smoother thank-you to CTA transition at the end of conversations.

Routine security and dependency updates to keep your data safe.

• New chat-first conversation experience: a voice-only mode with a centered microphone, and an inline "Talk" button that swaps with submit when typing begins.
• Localized placeholders and hints across 16 languages.

• Smoother end-of-conversation transitions, with a progressive loading indicator that eliminates layout shifts as reflections resolve.
• Tuned the welcome greeting animation timing for a snappier feel.

Voice-only actions like ending a conversation and extracting names now run reliably again.

Replaced the static contact page with an embedded Perspective interview that routes sales and support requests to the right destination and notifies Slack on completion.

Removed the dimming overlay from the mic and keyboard selection sheet for a cleaner experience.

Restored the external tool integration endpoint after a third-party update broke the connection.

Behind-the-scenes improvements to keep your workspace running smoothly.

• Stop button now cancels in-flight AI generation and tool runs, including intake form inspection and website analysis.
• Analysis sessions and the research assistant also support cancellation via the Stop button.

• Tool interfaces now clearly show streaming, interrupted, and error states.
• Activity tab now renders for conversations in any state.
• Smoother avatar uploads with built-in cropping.

• Switching between voice and text no longer drops assistant turns mid-conversation.
• Fixed a duplicate timestamp issue in avatar uploads.

• Renamed "Post conversation flow" to Automations and replaced collapsible sections with a tabbed layout (Per Conversation / Scheduled) for clearer configuration.
• Embedded conversations now forward all URL parameters from your page, not just UTM tags, so context flows through cleanly.

• Moved "Generate conversation summary" and "Generate trust assessment" toggles into Automations settings.
• The Add button tooltip is now context-aware based on the active tab.

Solution pages no longer load duplicate demo widgets: consolidated to a single responsive section.

Routine security and dependency updates to keep your data safe.

• New option to hide the fullscreen button in embedded conversations for a more contained experience.
• Added a "Fullscreen Button" toggle to embed appearance options so generated snippets reflect your choice.
• The Research Assistant can now answer questions about webhook setup, authentication, and event payloads.

Greeting generation now uses the configured display name for more accurate personalization.

• Complete redesign of the Solutions section with a Jobs-to-be-Done content model, filterable card grid, and shareable links.
• New Solutions detail pages with a two-column layout, sticky embedded demo conversations, and a 4-step "How it works" flow.
• New end-of-conversation action messages with completion CTAs, auto-redirects, countdowns, and thank-you states, including a reflection card that streams a participant-specific summary.
• Quick reply chips for text conversations, generated contextually per question.
• Calendly automation type added with its own icon and dedicated configuration.

• Solutions pages now load faster and rank better in search.
• Lighter-weight completion confetti for a snappier finish.

• Skipped automations no longer report as successful in test runs.
• More reliable streaming of reflections at the end of a conversation.
• Better text direction handling for multilingual conversations.

Improved voice conversation reliability so transcripts and audio always save.

More reliable voice connection startup: eliminated stale state issues that could cause the start guard to misfire.

Brand avatar now shows correctly in preview interviews when brand mode is enabled.

• New Agent mode for research creation: generates a complete research outline in one pass without clarifying questions.
• Research created via the API and external tool integrations now defaults to Agent mode; in-product creation remains interactive.

Social share image generation now handles a wider range of avatar formats by automatically converting them.

Routine security and dependency updates to keep your data safe.

• Brand mode for conversation identity: toggle between personal owner info and discovered brand identity, with automatic brand selection when website analysis finds brand data. Email templates respect the selected display identity.
• New Activity timeline tab on completed conversations shows post-conversation workflow steps (extraction, trust, automations, integrations) in real time.
• New animated processing indicator with logo and shimmer replaces the static "Processing…" text across conversation screens.
• The right agent type is now auto-detected and saved for older research projects.

• Workspace billing access restored: all members can view billing, credits, and subscription details again.
• Faster realtime voice responses thanks to quicker turn detection.
• Smoother avatar and logo uploads with built-in cropping and instant preview refresh.

• Fixed stuck loading states and stale data issues in brand mode.
• Streaming reliability improved with automatic retries and graceful backoff.

Routine security and dependency updates to keep your data safe.

• Fullscreen interview button in the header and floating controls: available for authenticated users and for anonymous users in embed mode.
• Tracking and context URL parameters are now preserved when opening an interview in fullscreen from an embed.
• Independent branding controls: custom logos and theme colors can now be enabled separately.

• Smarter logo and color extraction from analyzed websites, with better SVG handling, CAPTCHA detection, and favicon fallback.
• More concise responses from interview chats.

• Hebrew (right-to-left) titles now align correctly in URL previews.
• Anonymous sessions are now generated correctly when skipping welcome steps.
• Site favicon now loads reliably across the app.

• Automation failure tracking: in-app banners, inline error indicators in settings, and email alerts for persistent failures.
• Customizable call-to-action copy via a new actionText option on the research input (defaults to "Start research").
• Form inspector now extracts fields from custom web components, including Salesforce forms.

Errors are now classified as temporary versus permanent to drive smarter retry and notification behavior.

• Slack automations now survive channel renames by sending via channel ID instead of name.
• Underlying tool errors (e.g. "channel not found") now surface clearly instead of generic failure messages.
• More reliable file uploads: automatic client retries, fixed stuck spinners after failure, and multi-step embedded HubSpot forms now extract correctly.

Routine security and dependency updates to keep your data safe.

• New token-authenticated API endpoint to create conversations: POST /api/v1/perspective/create with token-based authentication.
• Auto-generated conversation titles and sensible defaults when creating via your AI assistant.

• Better first-message personalization: greeting generation now uses the user's message context.
• Shared default greeting and signature across the app and AI assistant tooling for consistent behavior.

Restored blog and docs content loading after a recent infrastructure change.

• Embedded conversation flow controls: Skip the welcome screen and start directly in voice or text, and immediately redirect on completion when a return URL is provided.
• Embed SDK migration: Embed now ships via published @perspective-ai/sdk and @perspective-ai/sdk-react packages with global CDN delivery for faster loads.
• UI theming: Custom theme support for chat input buttons and user message bubbles.

Simplified connection messages, tooltips, and prompts in the voice input for clarity.

Conversation header now always shows a logo, defaulting to the Perspective logo when no custom logo is set.

Multiple conversations per participant: Participants can complete a conversation more than once when enabled, supporting longitudinal studies, recurring feedback, and follow-ups. Each completed conversation creates a fresh attempt.

• Standardized toast notifications: Migrated to a consistent toast system across the app for clearer, more uniform feedback.
• Smoother credit-locked chat: Out-of-credits accounts now see a persistent warning banner above the input, with the input UI remaining visually consistent.

• Accurate trust labels in Slack: Trust labels in Slack messages now match the in-app score thresholds, preventing misleading labels for low scores.
• Conversation restart edge cases: Fixed priority and edge conditions around restarting conversations vs. starting new ones.
• No more horizontal scroll: Conversation link previews no longer scroll sideways on small screens.

Routine security and dependency updates to keep your data safe.

• All workspace members can manage billing: Buying credits, changing plans, and viewing invoices is no longer restricted to workspace owners.
• Clear "No Permissions" fallback screens appear in billing dialogs when access is missing.
• More reliable Stripe payment redirects and clearer error states in plan selection flows.

• Embedded conversation onboarding: New welcome flow for first-time users and a join flow for invited workspace members, both powered by AI-native onboarding with completion handling.
• Experience landing pages: New themed campaign pages that embed conversations directly (e.g., AI experience, product insights, TWIST).
• Embedded conversation widget: New widget with theme sync, lifecycle management, event callbacks, and brand customization support.
• Gemini 3 Flash (preview): Added across model selection.

Invited users are routed to the join flow and new users to welcome automatically.

Right-to-left languages now render correctly during streaming voice messages, spanning full width with proper alignment.

Routine security and dependency updates to keep your data safe.

• More reliable file uploads: Upload errors now return clear messages instead of failing silently, while keeping all existing safety checks.
• Cleaner conversation spacing: Adjusted message spacing so the thinking indicator no longer causes layout shifts.

• New participant-email automation sends AI-personalized follow-ups after conversation completion, with a preview dialog and reply-to support.
• Welcome screen now accepts a typed first response when voice is disabled, auto-submitting once the conversation starts.
• Embed SDK gains brand color customization and light/dark/system theme overrides, with branded loaders.

• Thinking indicator now stays visible for any assistant message with reasoning, while animated phrases play only during streaming.
• Email preview is more responsive on mobile, with default-template tooltips and cleaner setup.
• Automations keep running gracefully when individual steps fail or no conversations exist.

Fixed style bleed between light and dark modes in embed theming and corrected brand color formats.

• Fresh research images after upload: Avatars and logos now refresh immediately after replacement instead of showing the old cached version.
• Corrected a duplicate Polish translation in the thinking indicator.

• Sign-in redirects preserve destination: Unauthenticated users sent to sign-in now return to their original page after logging in.
• Updated Hebrew thinking indicator translation for clearer phrasing.

• Greeting generator in research chat: A dedicated tool generates greetings and titles automatically on the first outline, without overwriting later edits.
• Embed theming everywhere: Theme config now applies to embed root and container elements, not just the embedded frame, for consistent appearance across embed modes.

• Faster research list: Centralized redirect logic and streamed loading for quicker perceived performance.
• Conversation summaries now render with proper markdown in the updates view.

• Conversation completion redirects work again for anonymous users.
• Voice conversations on iOS no longer have scrolling blocked by the embed scroll workaround.
• Automations correctly return their actual ID after creation, eliminating "Automation not found" errors.

Routine security and dependency updates to keep your data safe.

Footer copyright year is now dynamic instead of hardcoded.

• Custom HubSpot tools work reliably: Tool registration is now per workspace, so custom HubSpot tools are consistently available.
• AI assistant responses include explicit URLs: Tools now return both share and direct URLs in their responses.
• Resolved immediate "Automation not found" errors right after creating an automation.

iOS embed scrolling fix: Embedded conversations on iOS now keep content in view instead of loading off-screen.

• Reliable conversation endings: The closing message and end-conversation action now happen together in a single response, preventing missed completions.
• Correct embed redirects: Parent page navigation only happens in true embed mode, preventing misdirected redirects elsewhere.
• Updated footer copyright year to 2026.

Routine security and dependency updates to keep your data safe.

• Embed brand colors: Customize embedded conversations with light and dark brand colors, plus forced light/dark/system theme via simple data attributes or programmatic config.
• Fullpage embed mode: New embed mode renders the conversation in the full viewport for dedicated conversation pages.
• Dynamic system prompts: Research chat prompts are generated per turn with versioned tracking for better adaptability.
• Multilingual thinking indicator: Conversation thinking indicator now supports multiple languages.
• Resource scope badges: Resource cards show "Shared" or "Private" badges for visibility at a glance.

Better brand logo extraction: Logos now load correctly from externally hosted sources (e.g., WordPress) and are auto-categorized as light or dark based on luminance.

• Post-conversation automations: Complete automation system with one-click integrations, tag-based triggers, durable workflows, and scheduled digests.
• Embed redirect handling: Embedded conversations can navigate the parent page with security validation and cross-origin support.
• Thinking indicators in text conversations: Extended thinking now appears in text conversations for more thoughtful AI responses.
• HubSpot custom forms: Direct integration for HubSpot form submission within automation flows.
• Email tool: Send emails directly from automation workflows.
• Webhook configuration: Set up webhooks dynamically with auth headers and built-in test capability.

• Cross-workspace search from your AI assistant: Search by name or title across all accessible workspaces, with paginated results and disambiguation metadata.
• Default workspace on create: Creating a conversation no longer requires specifying a workspace; it defaults to your primary one. Responses now include workspace info.
• Workspace context in responses: Create, respond, and update responses consistently include workspace ID and name.

• Clearer search behavior: List endpoints handle browse and search, with detailed lookup reserved for fetching by ID.
• Embed options panel reordered: preview and embed code first, appearance options at the bottom.

Correct access control in search: Visibility checks now use the actual visibility list, ensuring users only see researches they're permitted to.

• Embed appearance options: Control visual elements of embedded conversations directly from the embed box.
• New embed URL parameters: Hide the greeting or branding via URL parameters for tighter integration.
• Floating resources button: When the header or branding is hidden, a minimal floating button still gives access to resources.
• Hardened embed security: Tighter validation on cross-frame messages.

The floating resources button now only appears when there are resources to show.

More reliable research URL detection across the app.

• Fixed broken focus ring behavior on sign-in buttons and aligned spacing between sign-in and sign-up pages.
• Focus rings no longer appear immediately when dialogs open, only on keyboard navigation.

• Infographic generation from insight cards: Generate and view shareable infographics directly from cards.
• Default workspace lookup from your AI assistant: New tool to fetch the default workspace, plus a browser-friendly endpoint.
• Workflow run tracking: Improved navigation flow inside workflows with run tracking.

• Upgraded file storage: More reliable uploads with size limits and improved avatar handling.
• Google One Tap is now hidden in embed mode for a cleaner embedded experience.

• SVG uploads disabled for logos: Prevents potential script injection via uploaded SVG files.
• Custom logo dimensions and display issues resolved across the conversation UI.
• Conversation completion CTA stays active as expected.

• More reliable embedded conversations: Consistent anonymous IDs across actions and resolved embed rendering mismatches.
• Auto-redirect on completion now defaults on: New returnUrl query parameter as a fallback completion redirect.
• Better code blocks: Line numbers, theme-aware syntax highlighting, and improved copy experience.
• Improved right-to-left language support: Direction-aware padding, list styles, and outline rendering.

Mobile layout fixes: prevented badge wrapping and kept action menus within card bounds.

• Full conversation lifecycle from your AI assistant: New tools to create, update, respond, and set actions on conversations end-to-end.
• Conversation retrieval: List, fetch, and batch-fetch conversations with filters and stats breakdowns.
• Stats with average duration: Conversation stats now include average duration in seconds.
• Embedding support tools: New tools to fetch preview links and embed options for embedding flows.
• Webhook testing: New capability to validate webhook setup.

AI assistant tokens scoped to user: Tokens migrated from workspace to user scope for a more secure permission model.

• Owner names now display correctly instead of showing the workspace name.
• Outline updates now respect user feedback instead of ignoring it.
• Custom slugs no longer get saved as the research ID on invite.

• Embed conversations directly into your own pages, with an updated sharing experience.
• New website analyzer tool plus expanded theme options in research settings.
• Choose agents and switch between voice and text per research project.

Smoother post-conversation research settings, now with the ability to trigger webhooks when conversations complete.

• New unified plan selection dialog replaces the separate buy and upgrade flows across the app.
• Rebuilt pricing page with tabs, dynamic starter prices, and adjusted Pro tiers.
• Pro subscriptions can now be started, updated, cancelled, and resumed from one place.

• Checkout now supports coupons, invoice previews, and a clearer success screen, with an enhanced Stripe payment method UI.
• Versioned credit policies, including zero-cost research and analysis, with smarter consumption logic.
• Workspace switcher now groups workspaces, prefetches prices, and shows plan badges.

• Sign in faster with Google One Tap and a new email verification code flow that replaces magic links.
• Send personalized interview invites with automatic reminders, plus a participant summary captured at completion.
• Pick from multiple OpenAI voices with sample playback, and preview interview resources in a new sidebar.
• Extract structured data from interviews at scale with a new AI-powered tool.
• Analysis answers now cite their sources with clickable links back to the original interview content.

• Redesigned sidebar navigation, refreshed empty states, and a simpler welcome wizard for new users.
• Better mobile layouts across the interview viewer, admin panel, and chat, plus stronger right-to-left language support.

Audio now plays reliably on iOS, transcription accuracy improved, and several scroll, dialog, and mobile layout glitches are resolved.

New research type system with clearer categories: Exploratory, Evaluative, Concept Testing, and Generative.

• Replaced auto-scroll with a manual scroll button so you stay in control of the conversation view.
• User messages now resize dynamically for easier reading, and the sidebar layout is simpler.
• Renamed "highlight" to "insight" across the app for consistent terminology.
• Public insight pages now show the analyst's avatar and name for proper attribution.
• Gift credits are consumed automatically when granted, removing an extra step.

Resolved a major bug in interview prompt handling, plus several mobile sidebar and scroll button issues.

• Sign in with LinkedIn for faster account creation.
• New trust assessment scores conversation credibility across multiple dimensions automatically when conversations complete.
• Mark conversations as test or rejected and filter by status, plus filter conversations by date range with fiscal quarter support.
• Public, shareable highlight pages with table of contents, hero images, and rich link previews.
• Edit highlights, titles, and descriptions inline, and use charts, tables, and a calculator directly in chat.

• Smarter research creation flow with an expert mode for advanced users.
• Voice conversations now save audio in MP4, auto-play next clips, and detect speech turns more accurately.
• Analysis model picker is grouped by family, and Claude Opus 4 is now available.

Fixed voice recording cutoffs, mobile scroll issues on results pages, trust assessment streams, and several workspace and outline glitches.

• Conversation progress bar now stays at least one-third full so progress always feels visible.
• Simulated conversations now lean toward positive sentiment and you can generate up to five per research project.

Resolved share copy button issues, fixed delete analysis session, and prevented overly long greeting messages.

Added timestamps to conversation transcripts and a visual indicator for natural pauses between messages.

• Simulate conversations with AI, complete with reaction tracking and summary emails.
• Marketing blog now supports tagging, categories, and cover images.

• Refreshed welcome prompt, multi-line template card descriptions, and clearer list headers.
• Smoother link sharing and improved anonymous interactions.

• Fixed scrolling in the highlight sidebar and the analyst assistant edit view, plus duplicate buttons in lists.
• Cover image and remote image rendering on post pages now works correctly.

• Anonymous access is now the default mode, so participants can jump in without signing up.
• Auto-generated welcome messages greet users at the start of every conversation.
• Share with teammates from the settings sidebar, plus easier public page sharing from list view.

• Major UI refresh for more consistent styling and better accessibility across buttons, inputs, dialogs, and dropdowns.
• Scroll mode is now the default in conversation chat, with smoother animations and a refreshed welcome screen.
• Renamed "Perspectives" to "Home" and simplified the homepage layout.
• Default credit allowance increased to 250 with an updated pricing page.

Resolved conversation assistant scroll positioning, mobile card layouts, confetti in scroll mode, and various hover and toggle button glitches.

• Toggle between "My Perspectives" and "All Perspectives" on the list page.
• Sort and filter your research list by status from new dropdowns.
• New workspace shared page for published conversations.
• Monthly credits allocation for workspaces.
• Share with teammates directly from the settings sidebar.

• Major speedup on list and search loading, tuned for snappier results.
• Refined outline generator experience and conversation preview, including hover tooltips and clearer cards.

• Redesigned Plan and Billing section with a cleaner credits counter and a manage button.
• Tier name now appears next to the Pro plan, and you can cancel Pro directly from the change plan dialog.
• Smoother mobile experience for conversation details and a simpler attachment upload indicator.
• Comprehensive email template redesign with stronger information hierarchy and a new gray-logo signature.
• Hebrew lists in the conversation guide now align correctly.

Fixed broken preview link in the research-ready email and ensured at least one credits package always appears.

• New lifecycle emails for plan downgrades, resumptions, low-credit alerts, purchase confirmations, and tier changes.
• Added Gemini 2.0 integration with an updated pricing model.

• Apply coupon codes during checkout.
• New in-app feedback dialog so you can share thoughts without leaving the product.
• Account lifecycle emails for signup, out-of-credits, purchase confirmations, plan upgrades, and tier changes.

• Stronger contact form validation, consistent request type design, and better dialog handling.
• Newsletter system now sends acknowledgment emails and internal notifications with cleaner wizard validation.
• Credits now track plan and expiration dates, and consumption logic prefers non-expired same-plan credits.
• Polished plan dialogs, centered credits counter, centered sign-in popup, and a more responsive about page.

Better error messaging for missing user accounts, fixed form layout glitches, and corrected upgrade subscription handling for edge cases.

Cleaner visual proportions on the about page and a more readable founder story.

Fixed the disabled tier selection dropdown and resolved stale content glitches in the blog.

Blog posts now support custom slugs, plus an option to bypass stale content when publishing.

• New tiered flow for Pro plan subscriptions.
• Contact form with submission handling, persistence, and notifications.
• Workspace auto-generation in the welcome wizard with unique naming, defaulting to your full name.
• Trusted company logos now appear on the signup page.

• Sidebar now reflects account type, surfaces quick actions, and shows unconsumed credits in a progress bar.
• Refreshed homepage copy, new about page design, and updated customer success and sales hero images.
• Featured use cases added to the website footer with a refreshed social proof section.

• Better handling of credit errors in analysis session creation and the credits counter now only counts open balances for the current plan.
• Restored chat input suggestion behavior when the input has a value, and tightened access control on the research route.

Homepage solutions section now showcases the first three featured templates.

Fixed a typo in the CTA section.

Updated the pricing page and removed an unused notifications dependency.

• Credit consumption and enforcement now apply to analysis sessions and research features.
• Locked input mode for analysis sessions with admin override.
• Restored the locked conversation behavior.

• Refreshed pricing page, solutions pages, and sales hero image, plus a solid-background website footer.
• Reordered the top menu, smaller trusted company logos, and a gradient added to the blog page.
• Suggestions moved into the chat input for quicker access.

Fixed home link display in header and footer, FAQ credit card typo, and ensured deleted conversations no longer count in perspectives list cards.

Routine security and dependency updates to keep your data safe.

Newsletter form with new user subscription support.

• Renamed "interview" to "conversation" and "messages" throughout the product, with consistent UI text and labels.
• Refined messaging in the sharing dialog and an enhanced single conversation view.

Restored simulator functionality.

• New announcements admin and a website banner so you can broadcast updates to visitors.
• Drag-and-drop sorting for templates makes organization easier.

• Updated pricing page layout with rounded card backgrounds, refreshed borders, and a layered hero image.
• Sticky navigation buttons, badge labels on solution cards, and overall navigation polish.
• New about page with initial structure and layout in place.

Fixed the header overlay issue with the solutions menu and a use case label typo.

• New workspace configuration step in the welcome wizard, with team size now optional in account settings.
• Email consent controls in your profile and welcome flow, plus an unsubscribe link in onboarding emails.
• Attachments support in research creation and new analysis sessions.
• Refreshed website with a new pricing page, centered solutions design, blog in the header menu, and gradient hero sections.

• Fade-in outline animation, redesigned highlight cards with repositioned tags, and template badges with dark mode support.
• Mobile menu now keeps the solutions list persistent, and the website footer has a cleaner layout.
• Research pages gained use case badges, a "Watch a Demo" button, and a responsive "Talk to Us" button.

Fixed website rendering issues and signup redirect problems.

• New workspace setup step in the welcome wizard for business users.
• Email consent controls in your profile and onboarding flow, plus an unsubscribe link in onboarding emails.
• Attach files when creating research or starting a new analysis session.
• Added Gemini 2.0 Flash as an experimental model.

• Outline templates now connect to the website solutions pages.
• Subtler outline fade-in animation and a cleaner AI model selector.

Profile settings now refresh session data automatically when the page loads.

• Read PDFs directly in chat with new PDF support in messages.
• Add attachments throughout the product, including research chat and analysis session chat.
• Auto-generated session titles, plus asset and highlight counts on session cards.

• Refreshed research outline view, AI model selector, and use-case icons in the new research page.
• Updated prompt system to allow questions, and removed the three-question limit from outlines.
• Shorter welcome CTA on mobile, default invite share message, and improved research and settings page titles.

Resolved errors that affected analytics settings.

Added unique page titles across documentation and blog for better search visibility.

• Attach files during conversations and view them in the conversation viewer.
• Paste files directly into chat with a refreshed input layout.

• Admin sidebar now matches the cleaner Research sidebar behavior.
• Better SEO with corrected canonical URLs and validated landing pages.

Resolved canonical link handling issues.

• New AI-generated landing pages with a unified view to manage them all.
• Customer success landing page variations now available.

• Refreshed landing page design and management views.
• Faster, more responsive conversation and preview experience.
• Outline version controls restored and improved.

• Experimental AI survey builder for creating forms.
• New analyst assistant sidebar with highlights and chat integration.
• Streaming charts and tables in analysis tools with version history.
• Redesigned outline creation experience and research settings.
• Automatic tags and descriptions generated for analysis tools.

• Outline now reliably delivered by the AI model.
• Welcome page redirects properly to the new creation flow.

• Preview emails directly in the admin interface.
• Email template management is better organized with icons for missing templates.

Major platform upgrades for improved performance and reliability.

Highlights "By Category" mode now displays correctly on mobile.

• Conversation preview now closes automatically after submitting feedback.
• New Day 7 email template with refined campaign linking.

Corrected typo in solution hero title.

• New math tool integrated into analysis chat with full calculator coverage.
• Edit highlights with an improved interface, including title and description editing in a sidebar.
• Highlight categories (tags) for better organization.
• Markdown content editing with a save experience for highlights.

• Highlight card width and table overflow issues resolved.
• Smoother category collapse transitions.

• Outline now lives in the sidebar with faster performance.
• AI tool generates outlines automatically.
• Anonymous users can set a name on the create page.

• Smoother outline streaming without flickering.
• Renamed "Engine" to "Model" throughout the app for clarity.

• Login flow no longer breaks when validating waitlist users.
• Selected AI model now carries over from the new analysis page.

New employee retention landing page.

Outline parse errors no longer break the list page.

• Upgraded to the latest Claude Sonnet model.
• Name validation now accepts unicode characters.
• Unauthenticated users land back on the welcome flow after signing in.

Friendlier subject line on the welcome email.

Embedded videos on the website no longer open at a tiny size.

The "new" button on the list page now opens the perspective creation flow directly.

• New welcome flow with a Day 1, Day 4, and Day 7 onboarding email sequence.
• Create perspectives directly from templates, with admin tools to manage them.
• Anonymous welcome and research design now available.

• Streamlined navigation with "New" as a primary menu item.
• Better mobile experience on the new page and home page scroll.
• Clipboard copy now plays nicely with Google Docs.

Deleting a conversation from the results page redirects properly.

• Fixed a broken support link.
• Deleting a conversation from the results page no longer redirects to the wrong place.

Upgraded AI models for better performance and stability.

• Add context URLs to perspectives with a new resources tool, including versioning and a header dropdown.
• Sign in with Google or other providers now opens in a popup for a smoother flow.
• Terms and conditions added to participant signup with a legal disclaimer on login.

• Smoother handling when sign-in popups are cancelled or hit errors.
• Resolved signup tracking and sign-up flow issues.
• Edit button in the list page now works correctly.

Improved primary and background colors in dark mode.

First-time users can now skip the initial setup wizard step.

• New welcome experience with refreshed visuals and an animated wizard.
• Profiles now support multiple job descriptions instead of one.
• Home page redesigned with a mobile-friendly hero section.

• Resolved conversation login and sign-in layout issues.
• Fixed assorted bugs in the welcome wizard.
• Various mobile look and feel improvements, including landing pages.

• Question outlines now handle malformed input gracefully.
• Resolved a bug in the perspectives list.

New perspectives list with a polished empty state.

Conversation summaries are now sharper and higher quality.

• Preview dialog when sharing links inside a conversation.
• Per-page video support, including an employee engagement video on the HR landing page.
• New testimonials section on the website.

Conversations now wrap up properly when they reach their end.

• Refreshed hero copy, FAQ content, and the "How it Works" section.
• New theme colors and updated partner logos (Berkeley, Stanford).

Launched the first landing page for employee experience teams, with three layout variants for testing.

• New "How It Works" section to explain the product end to end.
• Dedicated customer and employee experience use cases for more tailored analyses.
• Fullscreen video playback (note: autoplay isn't supported in fullscreen).

• Smoother scrolling to the active solution role button and to the results menu on mobile.
• Clearer empty states in results and a refined solutions content hierarchy.

• AI model labels now display in full on the analysis view.
• Highlights section is hidden when there are no highlights to show.
• Conversations can be marked complete with the right permissions.

• New marketing homepage and solutions page with responsive layouts and category scroll navigation.
• Redesigned results overview with tab-based filtering.
• Added Google Gemini 1.5 to handle longer conversations.
• Analytics now include AI-generated summaries for richer insights.

• Refreshed "Getting Started" experience and homepage layout.
• Invite emails now reach recipients even if they aren't registered yet.

• Fixed responsiveness on the banner and solutions hero.
• Fixed assorted typos and minor layout glitches across the marketing site.

• Workspace members: invite teammates, resend invites, and track last activity.
• Research now defaults to workspace visibility, with controls to update it.
• Buy credits and billing pages are now open to all users.
• Simulated conversations now consume credits, including empty simulations.

• Cleaner publishing header and a left menu with hover indication and links.
• Refined end-of-conversation experience with more prominent "done" and "tell us why" actions.

• First version of the blog with shared header and footer.
• New documentation site with "On This Page" navigation.
• Pro badge indicator for upgraded accounts.

• More reliable sign-in sessions.
• Refreshed branding and richer link previews when sharing Perspective URLs.

Better handling of payment errors during checkout.

• New credits purchase flow with simpler options and an inline payment form.
• Added an undo for changes to a workspace's unique URL.

• Renamed "Billing History" to "Invoices" for clarity.
• Reorganized the admin menu and refreshed the account view.

• Conversation URLs always show the latest content for participants.
• Hebrew text in link preview images now renders in the correct direction.

• Pre-fill participant name and email via conversation URL parameters.
• New anonymous mode for external conversations.

• Stronger authentication for external participants.
• Generated feedback image is now attached to the conversation summary email.

Fixed caching issues with feedback images.

• Generate a shareable feedback image for any conversation.
• Researchers can now override the default conversation language, and the AI can switch languages mid-conversation.

• Research list now sorts by most recent activity.
• Cleaner analysis prompt rendering and small icon refresh across the app.

Conversations now start in the participant's preferred browser language.

Emails now include a trackable homepage link.

• Fixed several bugs in email templates.
• Fixed the research list filter for legacy members and owners.

• Add a custom message when sharing conversations.
• Custom messages are now supported on shared highlights too.

• Skip option in conversations so participants can move past sections they don't want to answer.
• Add URLs to research context for richer source material.
• Soft delete for conversations, with the ability to recover them later.

• Tooltip showing participant names in the conversation summary.
• Sidebar menu toggle moved next to the logo for a more consistent layout.
• Smarter title generation.

• More resilient handling of bad outlines to prevent crashes.
• More accurate URL scraping when source pages contain partial lines.

• Share conversations via QR code.
• Completion emails now go to all Perspective members.

Removed autofocus from chat inputs on mobile for a better typing experience.

• Fixed a typo in the invite teammates flow.
• Fixed an endless redirect loop on the results page.

Behind-the-scenes improvements to keep your workspace running smoothly.