Anthropic Interviewer is turning up on far more Claude screens this week. On 4 October 2026, TestingCatalog posted screenshots of an in-app invitation headed “Use your voice to help guide AI” and reported that Anthropic had “started prompting more users to participate into a 15 minutes voice interview”. The card opens with a question meant to put people at ease: “Excited, worried, or somewhere in between?”
The invitation belongs to a study Anthropic opened on 29 September under the title “What do you want from AI?”. It runs until 6 October 2026. It is open to Free, Pro and Max users of Claude and Claude Code, and for the first time each participant can choose to publish their whole Anthropic Interviewer transcript for anyone to read.
That last part is the real news. Anthropic’s own FAQ warns that a public transcript can be traced back to the person who gave it, and that the decision should be treated as permanent. Independent researchers have already shown that AI agents can re-identify people from the Anthropic Interviewer transcripts released in 2025, and those were partly redacted. This time, Anthropic says it will not edit them at all.
Below we set out what the invitation says, whether it really is a voice interview, who can take part, what Anthropic Interviewer asks, how this study builds on the 81,000 interviews Anthropic ran last December, what happens to your answers, and how to decide whether to go public.
Table of contents
- What the Anthropic Interviewer Invitation Says
- Is the Anthropic Interviewer Study Really a Voice Interview?
- Who Can Take the Anthropic Interviewer Study and When
- What Anthropic Interviewer Asks You
- How the Anthropic Interviewer Study Builds on 81,000 Interviews
- Making Your Anthropic Interviewer Transcript Public
- The Re-Identification Risk Anthropic Is Warning About
- What Anthropic Does With a Private Anthropic Interviewer Session
- Should You Take Part, and Should You Go Public?
- What the Anthropic Interviewer Study Means for Businesses
- Questions Anthropic Has Not Answered Yet
- Anthropic Interviewer FAQ
- References and Further Reading
What the Anthropic Interviewer Invitation Says
The invitation is short and friendly, and it is labelled as an announcement from Anthropic rather than a chat message from Claude. TestingCatalog’s two screenshots, taken in the Claude iPhone app, show the flow up to the start of the Anthropic Interviewer session.
The pop-up card
The first screen shows a corkboard illustration with sticky notes, a small tag reading “Announcement from Anthropic” and the headline “Use your voice to help guide AI”. The body text reads: “Excited, worried, or somewhere in between? We’re interested to hear how AI fits into your life and your hopes for what comes next.”
Two short lines follow. One says “Spend 15 minutes chatting about your experiences with AI.” The other says “Afterward, choose whether to make your interview publicly available.” Two buttons close the card: Begin the interview, and Not now.
The screen before the first question
Tapping the first button opens a page headed Anthropic Interviewer, which greets the user by name. A panel titled “About the interview” explains: “This 15-minute interview is conducted by Anthropic Interviewer, an AI that asks you questions.” Underneath sit three statements worth reading slowly:
- “Your interview will be shared with Anthropic once completed.” You can end it at any time before you finish.
- “Additionally, you can choose to make your interview public.” That choice comes after the interview.
- “Even if you choose not to make your interview public, Anthropic may still read and use it for research and improvements.”
A Keep reading button scrolls on to the rest of the terms, so the screenshots stop before the full consent text.
Why more people are seeing it now
Anthropic published the study page on 29 September. Within two days, Taiwanese news site ABMedia had photographed the Anthropic Interviewer invitation in its own Claude app, and Claude users in Japan were writing about taking the interview. TestingCatalog’s post five days later suggests the prompt is now reaching a wider slice of users. It added that the study “looks like it will be expanded to a broader audience soon”. Anthropic has not said how many people it plans to interview this time, or whether the 6 October closing date will move.
| Screen | Wording | What it tells you |
|---|---|---|
| Card tag | “Announcement from Anthropic” | It comes from the company, not from a chat |
| Card headline | “Use your voice to help guide AI” | The pitch: your opinion counts |
| Card, first line | “Spend 15 minutes chatting about your experiences with AI” | The time it takes, and the format |
| Card, second line | “Afterward, choose whether to make your interview publicly available” | The public option comes last |
| About page | “Conducted by Anthropic Interviewer, an AI that asks you questions” | There is no human interviewer |
| About page | “Anthropic may still read and use it for research and improvements” | Private does not mean unused |
Is the Anthropic Interviewer Study Really a Voice Interview?
The headline on the card says “Use your voice”, and TestingCatalog described the session as a voice interview. The rest of the evidence points to a typed conversation, with “voice” used in the sense of having a say.
What the wording says
The same card that says “Use your voice” describes the session as “15 minutes chatting”. Anthropic’s study page uses the word in the same way: “Your voice can help guide us, other AI labs, and policymakers as we navigate this uncertain future together.” Neither the card nor the study page mentions a microphone, audio, recordings or speech.
What Anthropic has said about the format before
When Anthropic launched the tool in December 2025, its research post listed the format as a limitation: “As Anthropic Interview is text-only and can’t read tone of voice, facial expressions, or body language, it might miss emotional cues”. ZDNET’s Sabrina Ortiz, who took that first public interview, called Anthropic Interviewer “a chatbot experience based on Claude”.
The privacy notice for sessions completed after 29 September 2026 lists what Anthropic collects as “the responses you provide”, survey or study metadata, account data, and participation metadata such as timestamps. It does not mention audio.
Why the difference matters
If Anthropic Interviewer has gained a spoken mode for some users, Anthropic has not documented it, and a recording would carry far more than the words in it. That is the subject of a separate request Anthropic made this week, when Claude users began seeing a prompt asking for Anthropic voice data to help train its speech models.
The two requests are easy to confuse. The voice data toggle is about the sound of your voice. The Anthropic Interviewer invitation, on everything published so far, is about your opinions, typed into a chat. If you dictate your answers with your phone keyboard’s microphone, the interview still receives text, but the dictation itself is handled by your phone’s speech service under that company’s terms rather than Anthropic’s.
Who Can Take the Anthropic Interviewer Study and When
Eligibility is narrower than “every Claude user”, and the window is short.
Dates, plans and products
Anthropic’s FAQ says the study “will run from September 29 to October 6, 2026, and is open to Free, Pro, and Max users on Claude and Claude Code whose accounts are at least two weeks old.” Claude Code appears to be the new addition: the December 2025 Anthropic Interviewer study invited people with a Claude.ai account.
Who is left out
The Anthropic privacy centre says Anthropic Interviewer “is not currently available for our commercial products such as Claude for Work and the Anthropic API”. Anyone who only uses Claude through a Team or Enterprise seat therefore cannot take part from that account, and neither can a developer whose only access is through the API. Accounts opened in the last fortnight are also excluded, which keeps out people who sign up just to take the survey.
How long it really takes
Anthropic estimates roughly 15 minutes, and participants’ experiences vary. Ortiz finished the December 2025 interview in six minutes, about half the estimate. A Japanese participant in the current study described it as a dense 15 minutes spent taking stock of how they use AI. Anthropic Interviewer adapts to your answers, so a short reply will usually be followed by a request for more detail.
| Question | Answer |
|---|---|
| Study window | 29 September to 6 October 2026 |
| Plans | Free, Pro, Max |
| Products | Claude apps and Claude Code |
| Account age | At least two weeks |
| Claude for Work (Team, Enterprise) | Not available |
| Anthropic API | Not available |
| Expected length | About 15 minutes |
| Public release | Optional, chosen after the interview |
What Anthropic Interviewer Asks You
Anthropic says it does not want to “give too much away” before you start. It has, however, published the three questions the study is built around, and participants have described the flow.
The three big questions
The study page lists them:
- What are your most meaningful experiences with AI, both positive and negative?
- Is there anything about how the world works (like work, school, healthcare, or government) that you’d like AI to help change?
- What do you want from the companies developing AI?
The FAQ adds that the third question covers Anthropic itself. “There are no right answers,” it says. “We want to hear about your experiences in your own words.”
How the conversation adapts
Anthropic Interviewer is a version of Claude prompted to run a research interview. In December 2025 Anthropic described three stages. A planning stage sets a consistent framework of questions. An interviewing stage follows that plan while adapting to each answer. In an analysis stage, human researchers work with Claude to find themes across the transcripts.
In practice that means follow-up questions. A participant who wrote about the current study on note.com said that after answering that they used AI for a wide range of things, Anthropic Interviewer pressed them to be more specific.
What participants say it felt like
Two Japanese users who took the current study published accounts on 1 October. One, who writes as KAO, said a conversation that began with recent uses of AI ended up covering parenting, sleep and work, including the fact that ChatGPT now tells them when to go to bed. KAO was also asked what AI should protect, not just what it should change, and chose to make the interview public. The other was asked whether there was anything they had decided never to use AI for, and found they had almost no such limits.
Earlier participants were positive too. Ortiz wrote that, compared with any other survey Ortiz had taken, the process “felt much more comprehensive and, as a result, enjoyable”.
How the Anthropic Interviewer Study Builds on 81,000 Interviews
This is the third outing for Anthropic Interviewer. The earlier rounds explain both why Anthropic is repeating the exercise and why it now wants to publish transcripts.
The 1,250-professional pilot
Anthropic launched the tool on 4 December 2025 with a test of 1,250 interviews: 1,000 people from the general workforce, 125 scientists and 125 creatives. It released the transcripts on Hugging Face with participants’ consent, and reported that 86% of professionals said AI saves them time, while 55% expressed anxiety about AI’s effect on their future.
Participants liked the format. 97.6% rated their satisfaction at 5 or higher on a seven-point scale, and 99.12% said they would recommend the format to others.
The 81,000-person study
The same month, Anthropic invited everyone with a Claude.ai account to sit an Anthropic Interviewer session over one week. It reported the results on 18 March 2026: 80,508 people in 159 countries and 70 languages took part, which Anthropic believes makes it “the largest and most multilingual qualitative study ever conducted”. Globally, 67% of interviewees expressed net positive sentiment toward AI, and the average respondent raised 2.3 distinct concerns.
What people hoped for and feared
Anthropic asked Claude to categorise what each person most wanted from AI and what worried them. The top three hopes and worries are below.
Top hopes and worries in the 81,000-person study, % of interviewees (Anthropic, March 2026)
Each bar is scaled against the largest figure, 26.7%, so 22.3% draws at 84% width and 13.5% at 51%. Hopes and worries are separate classifications, so the six figures do not add up to 100%.
What changed for this round
Anthropic lists three differences: “We updated the questions, the study is open on more Claude products, and for the first time you can choose to make your interview public.” The reason for the last change is on the study page. “For the previous study, only the high-level results and a small number of quotes were made public,” Anthropic writes. “This time, we want your thoughts to be made widely available so that anyone can learn from them.”
Making Your Anthropic Interviewer Transcript Public
The public option is what sets this round apart, and Anthropic has written an unusually frank FAQ about it.
What gets published
If you opt in, Anthropic will publish “your complete interview and the country associated with your conversation”. It will not include your name, email address or other Claude account information, and it does not plan to edit Anthropic Interviewer transcripts. After the study closes, it will do a final review and may remove interviews it does not think it would be responsible to publish, such as those containing confidential information or breaking its usage rules. The transcripts will go online once Anthropic has finished its initial analysis. It has not given a date.
Why Anthropic wants it public
Anthropic’s argument is that its own staff cannot see everything in the data. Researchers and policymakers elsewhere “can read interviews in their own languages, read responses shared from different regions, and ask questions we didn’t think to ask.” Anyone can also “check whether our published findings actually reflect what participants said.”
A public record is also harder to ignore. “If many participants say companies like Anthropic should do something differently,” the FAQ says, “that language stays public where anyone can reference it.”
What you cannot undo
The FAQ is blunt about the downside. “Once your interview is public, anyone can read, save, and use it. We can remove our copy of your interview on request, but we can’t delete what people have already saved. Treat the decision as permanent.”
The Anthropic Interviewer privacy notice repeats the point. A deletion request removes the transcript from Anthropic’s internal stores and its public repository, but Anthropic is “not able to guarantee that this deletes all copies”.
| What happens | Private interview | Public interview |
|---|---|---|
| Anthropic reads and analyses it | Yes | Yes |
| Used for research and improvements | Yes | Yes |
| Full text published online | No | Yes, unedited, with your country |
| Name and email attached | No | No |
| Seen by outside researchers or journalists | Possibly, de-identified, in line with your choices | Yes, anyone can read it |
| Deletion request removes every copy | Anthropic’s copies, subject to your local rights | No, saved copies may survive |
The Re-Identification Risk Anthropic Is Warning About
Anthropic does not just say re-identification is possible in theory. Its FAQ links to research showing it has already happened with Anthropic Interviewer data.
What researchers did with the 2025 transcripts
In January 2026, Tianshi Li published a paper showing that widely available AI agents with web search could link six of 24 scientist interviews from the December 2025 dataset to specific scientific papers, “recovering associated authors and, in some cases, uniquely identifying the interviewees.” Li found that existing safeguards could be bypassed by breaking the job into harmless-looking steps, and notified Anthropic.
A month later, a team from ETH Zurich, advised by Anthropic researcher Nicholas Carlini, went further. Their agent correctly identified 9 of the 33 scientists who had discussed their past research, at 82% precision, without breaking the task into steps. They estimate that at least 9 of the 125 scientists in the Anthropic Interviewer dataset can be re-identified. In a separate test linking Hacker News accounts to LinkedIn profiles, the same approach identified 226 of 338 people.
Why unedited transcripts raise the stakes
The 2025 scientist transcripts “were partially redacted for participant privacy”, according to the ETH Zurich paper, and they were still traced. The 2026 Anthropic Interviewer transcripts will not be redacted at all. Anthropic’s answer is to put the job of protection on the participant: “The best protection is to not share information that could identify you or others.” During the interview it shows examples of details to avoid, and afterwards you can reread everything before deciding.
Share of people correctly re-identified in published tests (arXiv, January and February 2026)
Each percentage is the count divided by the group size (226 ÷ 338, 9 ÷ 33, 6 ÷ 24 and 9 ÷ 125), rounded to a whole number. Bars are scaled against the largest, 67%. The last figure is the researchers’ lower bound for the whole scientist set, not a measured rate.
The small details that give you away
Anthropic’s own list makes a useful checklist. It names “age, job, city, or where you studied” as details “that seem harmless on their own” but make you easier to find, and warns that “someone could use AI to combine small details, like employers or past projects”. It notes that Anthropic Interviewer does not ask for those details. The risk comes from what you volunteer while telling a story, which is exactly what a good interviewer encourages you to do.
What Anthropic Does With a Private Anthropic Interviewer Session
Choosing not to go public keeps your transcript off the open web. It does not keep it out of Anthropic’s hands, and the privacy notice for this round is broader than the study page suggests.
Research, products and models
The study FAQ says Anthropic will “analyze interviews as part of our Societal Impacts research, share our findings publicly, and use the results to inform how we develop our models and products.” The supplemental Anthropic Interviewer privacy policy lists research “including model training” among its purposes, and gives consent, scientific research and legitimate interests as the legal bases. Anthropic has not said whether it will train models on the 2026 transcripts.
Linked to your Claude usage, and possibly your Memory
The privacy notice for sessions completed after 29 September 2026 says Anthropic “may combine your interview responses with aggregated, privacy-preserving summaries of your Claude usage data”. It may also analyse your answers across more than one Anthropic Interviewer study “to understand how perspectives change over time”.
If you have Claude’s Memory feature turned on, your interview responses “may be added to your Memory or used to personalize future conversations with Claude, consistent with the choices you make during a session.” Anthropic may also contact you about your transcript.
Who can read it
Anthropic staff “with a need to know” can access your answers. As with any Claude conversation, a session flagged under the usage policy may be reviewed by the trust and safety team. Anthropic also reserves the right to share de-identified responses with “select third parties such as external researchers or journalists”, again in line with the choices you make during the session.
Ending early
There is one clean exit. Click “End interview” before you finish and, Anthropic says, “your partial interview will be permanently deleted, and won’t be analyzed or published, including by Anthropic.” Once you complete the Anthropic Interviewer session, it is shared with Anthropic.
| Use | Applies to a private interview? | Source |
|---|---|---|
| Societal Impacts analysis and published findings | Yes | Study FAQ |
| Informs model and product development | Yes | Study FAQ |
| Combined with summaries of your Claude usage | May be | Privacy notice, 29 September 2026 |
| Compared across studies you take | May be | Privacy notice, 29 September 2026 |
| Added to Claude Memory | May be, if Memory is on | Privacy notice, 29 September 2026 |
| Shared de-identified with researchers or journalists | May be, in line with your choices | Privacy notice and supplemental policy |
| Deleted if you end early | Yes | Study FAQ |
Should You Take Part, and Should You Go Public?
These are two separate decisions, and it helps to make them one at a time.
Reasons to take part
Fifteen minutes is a modest cost for a direct line to a company that says the last round shaped its research agenda and was presented at the World Economic Forum. Anthropic’s chief executive, Dario Amodei, has argued for pacing the frontier as AI grows more capable, and the study page cites that essay when it describes the stakes. If you have strong views on how AI should be built, Anthropic Interviewer is one of the few places where a lab asks for them at scale.
Reasons to hold back
The sample is self-selected and limited to Claude users, which Anthropic itself flags as “not a representative sample of the public”. Your answers are read by Anthropic even if you keep them private, may be linked to summaries of your usage, and may feed into products and models. If you use Claude for sensitive personal matters, a conversation about your “most meaningful experiences with AI” will naturally draw on them.
A simple rule for the public option
Write your answers as if you were posting them under your own name, because a determined reader with an AI agent may be able to attach it. Then reread the full Anthropic Interviewer transcript at the end. If anything in it would embarrass you, your employer or a friend, keep it private. Anthropic will still analyse it.
- Decide before you begin whether you are likely to go public.
- Leave out your employer, job title, city, university and specific projects.
- Do not describe other people in identifiable detail.
- Use “End interview” if the conversation goes somewhere you do not want recorded.
- Reread the whole transcript before choosing.
- Treat a decision to publish as permanent.
What the Anthropic Interviewer Study Means for Businesses
The study is aimed at individuals, but it touches organisations in two ways.
Staff taking part on personal accounts
Because Anthropic Interviewer is not available on Claude for Work, any employee who takes part does so on a personal Free, Pro or Max account. Questions about meaningful experiences with AI will often lead to stories about work. A public transcript that describes a named client project, an internal tool or a recent incident is exactly the kind of detail the re-identification research exploited.
A one-line reminder in your AI use policy, and a short note to staff before the study closes, costs very little. If you do not yet have such a policy, our AI strategy team can help you write one alongside your wider plans.
A model for your own customer research
The bigger lesson is about method. Anthropic Interviewer shows that an AI interviewer can run tens of thousands of adaptive, open-ended interviews in a week, in dozens of languages, at a cost no human research team could match. Many firms will want to copy it.
If you do, Anthropic’s FAQ is a good template for the consent side: explain the risks before the interview, show examples of what not to share, let people reread before publishing, and confirm the decision once more at the end. Under UK GDPR, open-ended interview transcripts are personal data, and a free-text answer can easily stray into special category data such as health. Our data protection specialists can help you design that properly.
Questions Anthropic Has Not Answered Yet
Anthropic’s FAQ is more detailed than most, but several gaps remain.
- Format. Is any version spoken? TestingCatalog calls it a voice interview, while everything Anthropic has published describes a chat.
- Scale. How many users are being invited, and will the 6 October deadline move for the “broader audience” TestingCatalog expects?
- Publication. When will the Anthropic Interviewer transcripts appear, where, and in what format? The 2025 data went to Hugging Face.
- Model training. Will the 2026 transcripts be used to train Claude and, if so, will participants be asked separately?
- Memory. Is adding interview answers to Memory opt-in, and where is that choice shown?
- Review. Will Anthropic test public transcripts against the same AI re-identification methods that worked on the 2025 data before releasing them?
Anthropic Interviewer FAQ
Is Anthropic Interviewer a voice interview?
On the evidence so far, no. The invitation says “Use your voice to help guide AI”, but it also describes the session as “15 minutes chatting”, and Anthropic has described the tool as text-only. Anthropic has not announced a spoken mode.
How long is the Anthropic Interviewer study open?
From 29 September to 6 October 2026. Each interview takes about 15 minutes.
Do I have to make my interview public?
No. It is optional, and you decide after the interview. Anthropic will still analyse a private interview.
Can I delete my Anthropic Interviewer interview?
Ending the interview early deletes it permanently. After you finish, you can ask Anthropic to delete it, depending on your local privacy rights, but copies of a public transcript that others have already saved cannot be recalled.
Can I take part through Claude Team or Enterprise?
No. Anthropic Interviewer is not currently available on Claude for Work or the Anthropic API. It is open to Free, Pro and Max accounts that are at least two weeks old.
References and Further Reading
What do you want from AI? (Anthropic)
About Anthropic Interviewer (Anthropic)
What 81,000 people want from AI (Anthropic)
Anthropic Interviewer sessions completed after September 29, 2026 (Anthropic Privacy Center)
Supplemental Anthropic Interviewer Privacy Policy (Anthropic Privacy Center)
Anthropic Interviewer dataset (Hugging Face)
Large-scale online deanonymization with LLMs (arXiv)
I used Anthropic’s Interviewer tool to share my AI complaints (ZDNET)
An invitation from Claude: what Anthropic’s seven-day interview asked (note)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
