Estimate AI research use in your market

Which audience and decision are you studying?

Define the population you need to understand: recent customers, qualified prospects or another clearly described group. Choose a purchase context and recall period. Someone researching a personal trip is not automatically evidence about how an enterprise buyer selects a software provider.

Record how participants will be recruited and which buyers are likely to be missing. A survey of customers who completed a purchase can reveal their experience, but it cannot by itself represent every prospect who considered the category and chose someone else. Present the findings as evidence from the observed group. AAPOR’s survey guidance provides a useful reference for the sampling and questionnaire review.

What should you ask?

Separate stages of the journey. Ask where the person first discovered the supplier, which channels they used to investigate alternatives and which evidence mattered to the eventual decision. Allow multiple channels for evaluation and include an open response so your categories do not force an inaccurate answer.

Use concrete, neutral wording. For example: “Thinking about your most recent purchase of [category], which sources did you use to compare options?” Then distinguish AI assistants, ordinary web search, social or video platforms, publisher articles, peer referrals and direct supplier material where useful. Avoid suggesting that using AI is expected or more sophisticated.

Ask an optional follow-up about the kind of task an assistant helped with, such as finding suppliers or checking a feature. Collect only the information needed for the research; buyers do not need to disclose private conversation histories.

How should the percentages be reported?

Show the number invited, the number responding and the number answering each question. Keep “do not remember” and skipped responses visible. For a multiple-choice question allowing several selections, percentages can sum above 100%; label that clearly.

An illustrative result is that 18 of 60 respondents reported using an AI assistant during evaluation. That is 30% of these respondents, not 30% of all buyers in the market. It does not reveal whether AI was their first touch or the decisive influence unless those questions were asked separately.

What can analytics add?

Use GA4 Traffic acquisition to inspect observed sessions with a consistent date window and scope. Its current default channel definitions include AI Assistant for recognized assistant referrers and Organic Search for Google AI Overviews and AI Mode. The Traffic acquisition documentation distinguishes session acquisition from first-user acquisition.

Compare these observations with CRM notes and survey responses without adding them together as if they were mutually exclusive people. One buyer may use several channels, return repeatedly or report AI influence without an identifiable assistant referral.

What should the final assessment say?

State the group studied, the questions asked, recruitment limitations and the observed pattern. An AI answer’s source list measures sources exposed by the system, not human channel usage. Generated personas are constructed hypotheses, as the public synthetic-persona research makes clear from its construction method.

End with a decision proportionate to the evidence: improve the survey, run a limited channel test or change a specific information resource. Avoid translating an unrelated student survey or zero-click statistic into a confident budget share for your market.