How the study works
Anthropic is using an AI interviewer to ask Claude users about meaningful positive and negative experiences, changes they want AI to support and what they expect from AI companies. The study runs from 29 September to 6 October 2026 and is open to eligible Free, Pro and Max users on Claude and Claude Code whose accounts are at least two weeks old. Anthropic estimates an interview takes about 15 minutes.
After participating, a person can choose whether to make the complete interview public. Anthropic says a public record will include the country associated with the conversation but not Claude account details such as a name or email address. Participants can end early and have a partial interview deleted.
What the findings can and cannot show
Large numbers do not automatically make a sample representative. Participants are Claude users who volunteer for an AI-led interview. Public interviews are a further self-selected subset. People willing to publish their views may have different experiences or risk tolerances from people who participate privately or do not participate at all.
The questions and follow-up prompts can also shape what people discuss. Anthropic acknowledges these limitations. The results can provide rich qualitative examples and reveal themes worth testing, but they should not be presented as the views of the general public or of Australian businesses.
Why the approach is still useful
Open-ended interviews can expose issues that a fixed survey misses. People can describe where AI saved time, created anxiety, failed in context or changed how they approached work. Publishing interviews, where participants consent, could also allow outside researchers to test Anthropic’s interpretation against the underlying material.
For businesses, the useful lesson is methodological: product telemetry shows what people clicked, while interviews can explore why they acted, what happened next and what they expected. The two forms of evidence answer different questions.
How an Australian business could apply the idea
- Interview a small cross-section of staff and customers, including people who stopped using the tool.
- Ask for concrete recent examples rather than broad opinions about AI.
- Separate interviews about usefulness from performance assessment of the system.
- Explain how notes, recordings and quotations will be stored and used, and obtain meaningful consent.
- Look for repeated themes, then test them with usage data or a larger survey.
- Report who participated and who did not so readers understand the limits of the findings.
The practical takeaway
Anthropic’s study may produce valuable stories about how Claude users experience AI, but it will not be a population poll. Australian decision-makers should use the findings to generate questions and hypotheses, then validate those ideas with their own staff, customers, risks and operating data.