Anthropic Interviewer Asks If You Want to Go Public

A 15-minute chat with an AI about your hopes for AI

Anthropic has launched a new round of its Interviewer study, and this time there is a twist. The study runs from September 29 through October 6, 2026, and for the first time participants can choose to make their entire interview transcript public. The sessions are run by Anthropic Interviewer, an AI system built on Claude that conducts open-ended, adaptive conversations.

The study is open to Free, Pro, and Max users across Claude and Claude Code whose accounts are at least two weeks old. Each session lasts about 15 minutes. According to explainx.ai's detailed FAQ on the study, the interviews feed Anthropic's Societal Impacts research program.

The questions probe how people actually use AI, what they wish it could do for them, whether it has already moved them toward that vision, and what directions of AI development would run against their values. The interviewer follows up on each answer, pressing for the experiences and values behind the words.

The public opt-in comes with a re-identification warning

Publishing a transcript is optional. Participants who skip it still have their interview analyzed for research, and Anthropic may publish aggregated or de-identified findings. But participants who opt in should treat the choice as permanent: Anthropic can remove its own copy on request, but it cannot retract copies other people have saved.

That permanence is the crux of the privacy question. Anthropic says it cannot guarantee participants will not be re-identified. The company does not attach names or email addresses to published transcripts, and it does not edit what people said. But researchers have already demonstrated that interview transcripts can be re-identified by combining small details such as an employer or a past project. A public interview is, effectively, a public document.

Anthropic also reserves the right to drop interviews before publication after a final review. Transcripts that look irresponsible to publish — ones containing confidential information or violations of usage policy — can be withheld. Ending an interview early deletes the partial record entirely, so it is neither analyzed nor published.

The December 2025 round set the scale record

The new round builds on a much-discussed study from last winter. In December 2025, Anthropic ran the same instrument for a week and collected about 81,000 completed interviews, describing the effort as the largest and most multilingual qualitative study ever conducted.

The numbers behind that claim are striking. Respondents came from 159 countries and wrote in 70 languages. Anthropic used Claude-powered classifiers to tag each conversation across the participants' goals, their sense of whether AI had delivered, their fears, and their overall sentiment.

The findings, reported by Dataconomy, challenged the idea that AI is mostly a productivity tool. The most desired outcomes clustered around professional excellence, personal transformation, and life management. Many respondents who started by talking about work eventually described a more personal goal: gaining time, flexibility, and a better life outside work.

Optimism dominated but was not uncomplicated. About 81 percent of respondents said AI had already moved them at least somewhat toward their vision. Concerns were more fragmented, with respondents naming an average of 2.3 distinct worries each. Unreliability — hallucinations, inaccuracies, and the need for constant verification — topped the list, followed by economic and job-related fears and worries about the loss of human agency.

AI interviewers produce data, not meaning, researchers caution

Not everyone is convinced that an AI can do this job well. In a critique published via TechXplore, qualitative researchers who study digital technologies argued that AI interviewers cannot connect with people the way human researchers can. They can produce only data, not meaning, the researchers contend.

The argument is that qualitative research is designed to explore tensions, ambiguities, and paradoxes — the texture of lived experience. An AI without lived experience or the capacity for self-reflection can run the protocol, but it may miss the human connection that makes deep interviewing work. The researchers do not dismiss AI tools for social science, but they caution that ignoring their limitations risks undermining what makes this kind of research valuable.

That debate gives the current study a second layer of significance. The September round is not only collecting fresh data on how attitudes are shifting; it is also testing whether the public will volunteer their conversations with a machine to the world. If large numbers of people choose to publish, the result will be an unprecedented public corpus of humans telling an AI what they want from it — and what scares them.

For readers following the broader shift from chatbots to autonomous agents, see the AI News desk, and for the research and policy side of the story, the Science desk.