Verasight study finds AI interviews drive deeper answers but weaker representation

Verasight says a randomized study of 3,160 U.S. adults found AI-moderated interviews produced far richer responses than written survey questions, but with far lower completion and more skewed results. The findings suggest researchers may get more depth from AI interviews, but risk losing representativeness when asking people what they think about AI.

Why it matters: - Verasight’s findings point to a tradeoff for researchers using AI to collect qualitative survey responses. - AI-moderated interviews can surface more detail, but they may also filter out respondents who are less engaged with AI or less likely to finish. - That creates a risk of measuring opinions that are not representative of the broader population.

What happened: - Verasight released a randomized study of 3,160 U.S. adults on Sept. 2, 2026. - Participants were assigned either to traditional written open-ended questions or to an AI-moderated video or audio interview. - The study tested whether AI interviewing could deliver deeper responses and whether that depth came with costs in participation and representativeness.

The details: - AI-moderated interviews generated 4.8 times more respondent words than traditional written open-ended questions. - AI interviews produced 1.8 times as much usable information per assigned respondent. - Roughly 80% of the depth advantage came from adaptive follow-up questions asked by the AI interviewer. - Only 40.5% of participants assigned to an AI interview completed it. - Completion was 99.4% for participants assigned to written questions. - Most dropoff happened when respondents were sent to the external interview platform. - AI-interview completers were more likely to use AI regularly, trust AI companies with their privacy and write longer responses. - Completers were 7.4 percentage points more optimistic about AI than the full sample assigned to the AI condition. - Agreement that AI will improve daily life was 12.5 percentage points higher among AI-interview completers than among non-completers. - Demographic and political differences remained even after weighting. - Among people who completed the AI interviews, 86% rated the experience four or five out of five. - Another 91% said the interview was easy to complete. - Thirty-four percent said they believed they were more honest with a machine interviewer than in a standard survey. - The study was titled “Depth at a Cost: A Randomized Comparison of AI-Moderated Interviews and Written Survey Open-Ends.” - Verasight said the research was conducted from May 22–31, 2026. - The full report is available here.

Between the lines: - The core issue is selection bias. - If people who finish an AI interview are systematically more comfortable with AI, then the results can overstate enthusiasm for AI. - A researcher reading only the completed AI transcripts could conclude public opinion is more favorable toward AI than it really is. - The study suggests AI-moderated interviews may work best for discovery, nuance and explanation rather than population-level measurement. - Traditional written surveys or mixed-mode approaches appear better suited for broad participation and representative estimates.

What's next: - Verasight expects researchers to use AI-moderated interviews selectively, especially when depth matters more than breadth. - Broader survey work will likely keep using written questions or mixed methods when representativeness is the priority. - The study arrives as AI-moderated interviewing gains traction for faster, larger-scale qualitative research.

The bottom line: - AI can help researchers hear more from respondents, but in this case it also changed who stayed in the sample and what the results looked like.