AI chatbots are now better than human researchers at getting people to talk about what they actually need from a product.
A pre-registered study of 317 consumers, run with three industry partners, split participants into three groups: AI-moderated interviews, human-moderated interviews, and static surveys. The AI interviewer matched human moderators for depth of conversation and covered more topics, and for the same budget it uncovered significantly more customer needs than either the human interviews or the static surveys. People did report feeling more emotionally engaged talking to an actual person. Researchers then used the interview transcripts to build digital twins, AI models meant to predict how each participant would react to real marketing materials.
That last part is the catch. Twins built from AI-moderated interviews beat generic demographic personas at predicting reactions to six real marketing stimuli, but they did not beat twins built from cheaper static surveys. The extra conversational richness looks good in the transcript; it does not show up in the prediction accuracy.
Which is a useful reminder for anyone selling synthetic focus groups: a chattier AI interviewer is not automatically a smarter predictor, and the researchers traced the shortfall to two things, twins thinking differently than their human counterparts, and questions that strayed too far from what the original interview covered.