Anthropic built an AI interviewer to survey 1,250 workers

User research usually hits an annoying ceiling: surveys give you rigid numbers without nuance, while deep qualitative interviews take too much human time to scale past a few dozen people.

Anthropic just found a clever way around that bottleneck by launching Anthropic Interviewer, a Claude-powered system designed to conduct adaptive, one-on-one qualitative interviews at unprecedented scale.

To put the tool through its paces, Anthropic deployed it to interview 1,250 working professionals on Claude.ai. The cohort included 1,000 general workforce employees—mostly from education, computing, and creative media—alongside two targeted groups of 125 creatives (predominantly writers and visual artists) and 125 scientists across over 50 fields. The anonymized dataset has now been released publicly for researchers.

How an AI conducts a 15-minute qualitative study

Anthropic structured the interviewer pipeline into three distinct phases: planning, interviewing, and analysis.

First, researchers use system prompts to establish interview rubrics and methodology guidelines. Claude drafts the conversational flow and specific questions, which human researchers then review and approve.

Next, the model conducts real-time, 10 to 15 minute text interviews directly inside Claude.ai. Instead of presenting a rigid questionnaire, the agent follows the master topic guide while dynamically digging deeper when participants bring up unexpected insights.

Finally, the transcripts are processed by human researchers working alongside an automated AI thematic analysis tool, which clusters emerging themes and quantifies how often specific viewpoints appear across hundreds of conversations.

Secret workflows and peer stigma

The findings paint a vivid, complicated picture of how people actually navigate AI at work today.

While 86% of general workers reported that AI saves them time, workplace culture has not caught up. About 69% noted social stigma around using AI on the job. One fact-checker admitted staying completely silent when colleagues ranted about hating AI, keeping their own workflows entirely hidden.

Career anxiety is also acute. While 41% of workers feel secure in their human skills, 55% expressed worry about the future. One interpreter was already studying for a career switch, anticipating broad job displacement.

In specialist fields, the friction looks different. Creatives are actively using AI to boost output while navigating harsh peer judgment and existential dread. Scientists, meanwhile, universally want AI partners to help draft hypotheses and design experiments—but because they still do not trust the models with core research, they currently restrict AI to manuscript writing and code debugging.

Why it matters

Anthropic just handed product and social researchers a blueprint for scaling qualitative research without losing conversational nuance. Traditional polling tells you what people clicked, but adaptive AI interviews capture the awkward, messy reasons why they did it. For organizations trying to understand fast-moving customer sentiment, this approach cuts months of manual interview labor down to days.

More broadly, the study reveals that the biggest bottleneck for workplace AI is no longer capability, but trust and stigma. When nearly seven in ten professionals feel compelled to hide their AI usage from coworkers, productivity gains are happening in the shadows. Until leaders address workplace culture and scientists can verify core outputs, AI integration will remain fraught with anxiety and quiet resistance.