Attest's AI Interviews Changed a Researcher's View of Shoppers

Attest research lead Nicholas White changed a conclusion after hearing how shoppers made decisions in 60 AI-moderated interviews. Their October 26 account gives marketers a reason to check a buyer’s unanswered product question before assuming the problem is awareness.

The Head of Strategic Research was trying to explain findings from Attest’s survey of 4,000 U.S. and U.K. consumers about first-time purchases. The survey already showed people choosing familiar brands. The interviews helped White understand what they still needed to know.

SOAM has covered a separate Semrush survey on AI talking buyers into and out of purchases. Attest’s new account examines the question that brought a buyer to the assistant and how that question changes the explanation of the purchase.

The missing question was already in the buyer’s mind

Attest ran 60 audio interviews through its Explore product, split between the U.K. and U.S. These weren’t the same people as the survey respondents. Attest says both groups had the same audience criteria and screening question.

White describes a shopper who already wanted a drinks machine and used Gemini to resolve an uncertainty before buying. The conversation gave the researcher an explanation that the survey hadn’t supplied.

That’s a different marketing problem from low awareness. If a customer already knows your name, another impression may not move the sale. The unanswered question might be missing from your product page or sales copy. For a lean B2B team, confusing discovery with doubt can send limited budget toward the wrong fix.

The interviews also changed a claim about product rejection. White had started to describe AI as deciding what shoppers ruled out. After reviewing the conversations, they wrote:

“The interviews didn’t support it.”

White says many interviewees had formed the objection before opening the assistant. The account now describes the assistant as a place where a rejection is finalized. That’s a narrower claim than saying the system created the objection.

For marketers, the sequence determines which question to investigate. An incorrect product description calls for a different response than an accurate answer confirming a concern the buyer already had. The examples in White’s account illustrate that distinction; they don’t establish how often either situation occurs.

The useful test is what changed the buyer’s mind

A survey can show a pattern without explaining it. Interviews can expose the question behind a number, but a small sample can’t tell marketers how common that explanation is. Use the conversation to improve the next question. Don’t treat one vivid answer as proof of market-wide behavior.

There is relevant independent evidence about the method. Nielsen Norman Group’s January 30 study of AI-moderated interviews used 10 research professionals and two products. AI interviewers could collect structured feedback, but struggled to change direction when an unexpected answer mattered. That’s a small study too. NN/G said the products it tested didn’t sponsor or compensate the work. Its finding doesn’t replicate Attest’s shopping results. Marketers should inspect the method before trusting an interpretation.

Maria Rosala, NN/G’s Director of Research, explains the human work in a separate July 17 essay. The parts worth protecting, they write, are:

“the ones where the learning lives”

Rosala includes observing sessions and interpreting what people mean. That’s the work a marketing team still has to make time for after receiving a research summary. NN/G also sells research consulting and training; the essay is a practitioner’s argument.

A separate May 29 qualitative-methods paper compared approaches to analyzing discussions among teachers. AI-assisted analysis reproduced predictable themes but missed some context-dependent meaning. It didn’t test AI moderators or Attest’s product. It supplies another reason to examine an interpretation before using it.

Attest sells research software and says the interviews used its Explore platform. The full consumer report landing page gates the report behind registration, and the account doesn’t expose the survey instrument, weighting, or transcripts. The 60 interviewees were different from the 4,000 survey respondents, so the conversations suggest a possible mechanism but don’t show how often it occurs.

Fix the doubt before buying more attention

Ask what the customer thought before opening the assistant and what changed afterward. That sequence helps separate a new preference from a question answered about a product they already wanted. This account doesn’t compare AI interviewers with human interviewers.

That separates this story from AI shopping recommendations and creator reviews, which examined what evidence shoppers want in an answer. It also differs from synthetic research, about AI-generated respondents rather than real shoppers interviewed by AI.

Before you fund another campaign around “AI discovery,” ask one customer what they needed to know before opening the assistant. If it’s a product fact you’ve published, make it easier to verify. If you can’t find it on your site, you may have found work that matters more than another impression.