Claude interviewed almost 81,000 people in 70 languages

Almost 81,000 people across 159 countries sat down to talk about how AI has changed their lives, and the interviewer in every one of those conversations was Claude. The tool is called Anthropic Interviewer, and Anthropic describes it as a way to run detailed conversational interviews at scale and pass the results back to researchers for analysis.

At a glance

• Anthropic has built a research tool on Claude that plans interview studies with researchers, runs the conversations itself and then sorts the answers into themes for the team to check.
• Its global research pilot gathered accounts from almost 81,000 people in 70 languages about how AI has affected them and what they want from the technology.
• The announcement describes the method and the scale but shares no findings, completion rates or validation details, and the results themselves sit in separate survey write-ups.

If you haven't been following, Anthropic's earlier research looked at what people use Claude for, and the company says it learned a lot there. It learned far less about how Claude's outputs affect people, how they feel about AI and what role they imagine it playing in their futures. Anthropic concluded that it had to ask people directly, and Interviewer is the tool it built to do that.

Almost 81,000 people in 159 countries answered in 70 languages

Anthropic says it has now used Interviewer in multiple studies. One was a focused test-study on how workers, scientists and creatives bring AI into their professional lives. Another was a global research pilot, in which almost 81,000 people across 159 countries and 70 languages described in their own words how AI has affected them and what they want from it.

Before agreeing to an interview, each participant is told how Anthropic plans to use their responses, and a separate privacy policy covers the tool. According to Anthropic, the insights feed its Societal Impacts research and help shape how Claude is built. The company also says that, in line with its policy commitments, it shares them with policymakers, independent researchers and the public.

Researchers set the goals, and Claude drafts the interview guide

The tool works in three stages. In planning, researchers define the study's questions and goals. Claude then drafts an interview guide, which Anthropic describes as a strategy and list of questions grounded in best practices, and the research team refines and finalizes it.

An interview guide is the script a qualitative researcher brings into a conversation: the topics to cover, roughly in what order, and the probes to use when an answer comes back thin. It works like a recipe that a cook can adjust at the stove but can't throw away. In this case the model writes the first draft, and researchers approve the final version.

Claude adjusts its follow-up questions, and researchers check the themes

In the interviewing stage, Claude holds a real-time conversation with each participant. It adjusts its follow-up questions to what that person has just shared, and it runs the whole exchange in the participant's own language.

Follow-up questions are what separate an interview from a survey. A form asks everyone the same fixed questions, while an interviewer can ask why, ask for an example or chase a detail the participant raised unprompted.

In the analysis stage, the tool synthesizes the responses against the original research plan. It identifies emergent themes, meaning patterns nobody listed in advance that keep recurring across conversations, and surfaces illustrative quotations. Anthropic says researchers collaborate throughout to validate and interpret the findings.

The page doesn't show how well any of this works. Anthropic gives no figures on interview length, completion rates or how often researchers rejected the themes the tool proposed. In our view, the validation step is the weakest-documented part: with almost 81,000 conversations, readers have to take the human check on trust.

What the next Interviewer write-up could answer

Anthropic gives no schedule for further studies and no date for a next round. It also doesn't say whether outside researchers will be able to run their own interviews with the tool. The open question for any future survey write-up is whether it explains how researchers checked the tool's themes, and how many of them held up after that check.