Anthropic asks its users what they want from AI
Anthropic Interviewer
An AI system that conducts the study by asking participants questions.
representative sample
A group chosen to reflect the wider population.
de-identified findings
Results presented without details meant to identify a person.
What happened
Anthropic, the company that makes Claude, launched a new study on September 29, 2026. It asks people about their experiences with AI and what they want AI companies to do. The conversation is led by Anthropic Interviewer, an AI that asks questions. The study runs through October 6. It is open to Free, Pro, and Max users of Claude and Claude Code. Their accounts must be at least two weeks old. Anthropic estimates that each interview takes about 15 minutes.
The questions cover meaningful positive and negative experiences with AI. They also ask whether participants want AI to change parts of work, school, health care, or government. A final question asks what participants want from companies developing AI, including Anthropic.
Public sharing is optional
Participants can read their full interview before deciding whether to publish it. If they agree, Anthropic says it will publish the complete interview and the country linked to the conversation. It will not publish Claude account information, such as a name or email address. Anthropic says it does not plan to edit interviews. A final review may remove interviews containing confidential material or other content it considers irresponsible to publish.
The warning is important. A person may reveal an age, job, city, or past project without giving a name. Other people may combine those details and identify the speaker. Anthropic can remove its own copy after a request. It cannot remove copies that other people already saved. If someone declines publication, Anthropic may still analyze the interview and publish aggregated or de-identified findings. Participants can end early. A partial interview is then permanently deleted and not analyzed or published.
Why collect these answers?
Anthropic says AI is becoming more capable, useful, and dangerous at the same time. It argues that decisions about benefits and risks should not be left to AI companies alone. Public answers could give Anthropic, other labs, and policymakers evidence about how people experience AI.
The company says a December study gathered hopes and worries from 81,000 people. That work shaped the Anthropic Institute’s agenda, reached leaders at the World Economic Forum, and continues to inform its societal-impacts and economics research.
This new project updates the questions. It includes more Claude products. It offers public release for the first time. Anthropic also says it wants the answers to become a public record that researchers, policymakers, journalists, and future historians can examine.
What is confirmed, and what is limited?
The source clearly describes the invitation, the consent choices, the privacy warnings, and the study’s limits. Everyone in the study uses Claude. Published interviews are an even smaller opt-in group. This is not a representative sample of the public. The questions themselves also shape what participants discuss. The eventual collection can show what some Claude users chose to say. It cannot describe everyone’s view of AI.
On Hacker News, the post received 28 points and 64 comments. That records community attention. Those numbers do not prove that Anthropic’s claims are correct, and this article does not treat the comments as evidence about the study.
What to watch next
The study page does not yet say how many people will participate, how many interviews will be published, or what the results will show. Anthropic says interviews will appear online after the study and its initial analysis are complete. The useful tests will be whether the published group differs from the wider user base, whether the findings explain their limits clearly, and whether Anthropic can point to changes in its models, products, or research agenda.
Anthropic asks Claude users what they want from AI
📰 Full story: Anthropic asks its users what they want from AI
Anthropic is collecting hopes and worries about AI. People may publish their answers, but privacy risks remain.
Anthropic Interviewer
An AI system that asks people questions for the study.
representative sample
A group that fairly reflects the larger group.
de-identified results
Results with identifying details removed.
💡 The gist
- Anthropic asks Claude users what they want from AI.
- An AI interviewer asks about helpful and harmful experiences.
- Public answers can help others learn, but privacy risks remain.
Anthropic, the company behind Claude, launched a study on September 29, 2026. It runs until October 6. It is open to Free, Pro, and Max users. Their Claude or Claude Code accounts must be at least two weeks old. Each interview takes about 15 minutes.
Anthropic Interviewer asks the questions. Participants describe important AI experiences. They can discuss good moments and bad moments. They can suggest changes for work, school, health care, or government. They can also say what they want from AI companies.
Sharing is optional. Before sharing, people can read the whole interview. A public version includes the full interview and the country connected to it. It does not include the Claude account name or email. Anthropic says it does not plan to edit interviews. A final review may remove sensitive or irresponsible interviews.
Not showing your name does not guarantee privacy. Someone might mention an age, job, city, or past project. Another person could combine those clues. That person might guess who spoke. Anthropic can remove its copy after a request. It cannot remove copies saved elsewhere. Public sharing should therefore be treated as permanent.
People who keep answers private still contribute to research. Anthropic may publish combined or de-identified results. A participant can stop early. The unfinished interview is then deleted permanently. It is not analyzed or published.
Why does Anthropic want these answers? The company says AI is becoming more useful and more dangerous. It wants public voices included in decisions. It says a December study collected hopes and worries from 81,000 people. That study influenced the Anthropic Institute’s agenda. This new study updates the questions and covers more Claude products.
The study has an important limit. Every participant uses Claude. Public participants choose themselves. So it does not form a representative sample of everyone. It cannot describe everyone’s opinion about AI. The questions also shape the answers.
On Hacker News, the post got 28 points and 64 comments. The post shows attention from that community. It does not prove the study is correct.
We still do not know the final number of participants. We do not know how many interviews will become public. Anthropic says results will appear after the study and first analysis. Watch what gets published. Then ask what changed because of the answers.
💬 An easy summary of the HN discussion
Most of the debate is about why Anthropic is asking, and what happens to users' time and data, rather than about whether feedback can ever help. Some people found the interview useful; others did not trust it.
- Many commenters do not want to give Anthropic free ideas or time. This thread is not a representative survey of everyone.
- Some suspect the interview also advertises Anthropic's AI interview tool, gets new sign-ups, or gathers data. These are suspicions, not proven facts.
- Access was uneven: some commenters said a login was required and one left, while another said no login was required. This is self-reported.
- The experience was mixed: one person reported stopping at question four because it felt like a form; another reported finishing and finding the AI's follow-up useful.
- People want smaller, local, open-source LLMs and question huge compute, data centers, and calling today's LLMs AGI. Some worry an AI interviewer misses human tone and context; others say asking users is sensible, though paying people could change who answers.
initial digest at 64 comments (revision 1). We fetched 64 comments and sampled 64 across the thread. These are HN users’ reports, not independently verified facts.
Anthropic asks, “What should AI do for you?”
📰 Full story: Anthropic asks its users what they want from AI
Anthropic asks AI users what they liked, disliked, and want changed.
Anthropic Interviewer
The AI that asks the questions.
public
Shown so everyone can read it.
Hacker News
A place where people talk about technology news.
Anthropic, the company behind Claude, is asking people questions.
It asks, “What should AI do for you?” Anthropic Interviewer is the AI asking questions. People can share happy AI moments. They can share sad AI moments too. The talk takes about 15 minutes. People can keep answers private. They can also make answers public. Public means everyone can read them. A public answer shows the whole talk. It also shows a country. It does not show a name or email. A town, job, or age can still give clues. Someone may guess who spoke. Someone may save the answer. Anthropic cannot erase every saved copy. Private answers may still help Anthropic learn. An earlier study heard from 81,000 people. Claude users are not everyone. Their answers cannot speak for everyone. Hacker News, a place for tech news, noticed it too. The post showed 28 points and 64 comments. That shows attention. It does not show truth. We do not know the final number of people. We do not know what will change.
💬 Anthropic's AI question, for a five-year-old
People are unsure whether Anthropic really wants to listen or is also trying to advertise and collect data.
- Some people stopped because they did not like logging in. Another person said no login was needed.
- One person self-reported stopping at question four because it felt like a form. Another self-reported finishing and liking the AI's follow-up.
- Some people want small, open AI and worry that a computer may miss human feelings. Others say asking users is good, but paying people may change their answers.
initial digest at 64 comments (revision 1). We fetched 64 comments and sampled 64 across the thread. These are HN users’ reports, not independently verified facts.
💬 HN reactions to Anthropic asking for thoughts on AI
The main dispute is less about whether feedback can be useful than about transparency, access requirements, and whether users are being asked to provide time, ideas, and data for free. Some participants reported a worthwhile experience, while others defended asking users directly.
initial digest at 64 comments (revision 1). We fetched 64 comments and sampled 64 across the thread. These are HN users’ reports, not independently verified facts.