Anthropic is asking Claude users to tell an AI interviewer what they want from artificial intelligence, and this time participants can choose to put their complete interview online. The study launched on 29 September and asks about meaningful experiences with AI, changes people would like to see in work, school or government, and what they want from the companies developing the technology.
The unusual part is the publication option. Anthropic says a participant can choose to make the full interview and associated country public, without the person’s Claude account name or email address. It also warns that details inside an interview could still make someone identifiable and that copies saved by other people cannot be recalled later.
Claude users are being asked for experiences rather than a score
AI opinion surveys often reduce attitudes to a percentage: trust, fear, enthusiasm or adoption. Anthropic is using its Interviewer system to collect longer accounts in people’s own words. Each interview takes roughly 15 minutes and is open from 29 September to 6 October to eligible Free, Pro and Max users of Claude and Claude Code whose accounts are at least two weeks old.
The questions are broad by design. Anthropic says it wants to hear about positive and negative experiences, how people would like AI to change parts of everyday life and what demands they have of AI companies. There are no right answers, and people can take part without agreeing to publish the transcript.
That approach can capture contradictions that a simple rating misses. Someone may find AI extremely useful at work while also worrying about jobs, privacy or concentration. A long interview can preserve those tensions instead of forcing the person into a single category.
LiveAIWire has covered research on how AI responses can shape trust and influence. Asking users to describe their own experiences provides a different kind of evidence, but it also carries its own selection effects.
Publishing the interview creates a real privacy trade-off
Anthropic is unusually explicit about the risks of making a transcript public. It says a published interview will include the full conversation and the participant’s country. It will not attach the account name or email address, but the company cannot guarantee anonymity because people may reveal a job, city, project or other combination of details that points back to them.
The warning goes further. Anthropic says a public interview should be treated as effectively permanent. It can remove its own copy on request, but it cannot delete copies that somebody else has already saved. The company also notes that scammers or other actors could use public information in ways a participant did not intend.
Participants can reread the transcript before making a publication decision, and Anthropic says it will show examples of sensitive information that should not be shared. A person can also end the interview early, in which case the partial record is deleted and is not analysed or published.
Those controls are useful, but they do not eliminate the basic tension. The research becomes more valuable to outsiders if the raw material is open enough to inspect, while the participant becomes more exposed as the transcript becomes more detailed.
Anthropic has already tried this at much larger scale
The project follows an earlier Anthropic exercise in which about 81,000 people were interviewed about their hopes and worries around AI. Anthropic says that earlier work informed its research agenda and was presented to international decision-makers.
The new study differs because participants can make their full interviews public. That creates a record that researchers, journalists and policymakers can inspect independently rather than relying only on Anthropic’s summary of what people said.
It also means the study should not be mistaken for a representative poll. Anthropic says all participants are Claude users, which already makes them different from the public as a whole. People who opt to publish are another self-selected subset, and the questions themselves influence what people choose to discuss.
Those limitations matter if the interviews are later used to make claims about what society wants from AI. The material may be rich and revealing without being statistically representative.
AI companies are becoming researchers of their own users
There is a broader development underneath the study. Companies building general-purpose AI now have access to enormous populations using the technology in work, education and personal life. That gives them an unusual opportunity to study how the technology changes behaviour while they are simultaneously changing the technology itself.
Independent scrutiny is therefore important. LiveAIWire has looked at the growing use of conversation data to understand human behaviour. Making interview material available to outside researchers could improve transparency, provided participants understand what publication means.
The project may also reveal how quickly expectations are moving. The questions people asked of AI two years ago were largely about answers and generated content. As agents begin to act, remember and work in the background, users may have different hopes and different boundaries.
Anthropic’s study will not settle what people want from AI. It cannot, because its sample is too particular and its participants are choosing how much to disclose. What it can do is preserve a detailed snapshot of what a group of active AI users say matters to them at a moment when the technology is changing unusually fast.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.