A hundred AI personas are not a hundred customer interviews.

Use imagined reactions to develop questions, then label them as generated hypotheses. Do not report them as people who used your product.

Our view: AI role-play can be a source of questions to investigate, but the output should keep its origin attached. A plausible complaint is not evidence that a customer actually experienced it.

Separate a hypothesis from a response

Suppose you ask a model to imagine why a new user might abandon a form. A possible friction point can guide a usability check. It does not establish how many real users left, why they left or what a particular customer said.

What the professional code distinguishes

AAPOR’s ethics code, revised June 2026, defines research participants and respondents as human beings. It says AI-generated cases must be identified as such and that terms such as poll and survey should not describe AI-created data. This is a professional code, not a universal law.

Keep generated material out of testimonial counts

If a planning note contains generated reactions, label the section “Generated hypotheses for investigation” and record how it was produced. Keep it separate from interviews and observed behavior. Do not attach an invented customer name, purchase history or satisfaction percentage.

Test the question with real evidence

Decide which observation would support or challenge the hypothesis, then collect appropriate evidence with consent and a clear method. More generated variants can broaden brainstorming; their number alone does not establish representativeness.

Published October 14, 2026. Editorial analysis; the persona scenario is hypothetical. We generated no respondent dataset and conducted no customer interviews or survey.

Sources: AAPOR · Ethics code.