Market Research and Insights in the AI Age: From Producing Answers to Vouching for Them
Can You Trust AI's Answers in Market Research?
Only after you check them. AI output arrives fast and polished, and a wrong answer from AI can look more convincing than a right answer from real people.
Convincing answers go wrong in four ways, each with a question to ask:
Too eager. Deanna Zawadiwsky and Jessica Benoit, quantitative researchers on Calvin Klein's global consumer insights team at PVH, found that AI summaries will build a theme from two mentions. Ask: how many people actually said this?
Too attentive. When Ananta Engineer's team at McCain Foods tested ads on consumers and synthetic personas, the personas noticed "every little button" of branding. Real consumers "are looking at four screens at a time." Ask: would a distracted shopper notice?
Too plausible. In a September 2026 Pew Research Center study, AI stand-ins estimated that 97% of Hispanic adults would follow the World Cup. The real figure was 43%. Ask: would you believe this result if it surprised you?
Too narrow. AI reflects only what its data contains, so it can't get "a good read" on something truly disruptive, says Engineer. The same goes for unmet Jobs to be Done, the progress customers want to make but haven't voiced. Ask: what can't the model see?
Leaders use AI where its mistakes are cheap. Amanda Michelangeli, who heads the luxury and beauty group at research agency Reach3, sees a role for synthetic data in narrowing, say, 30 concepts to the five a budget allows. The final round uses people. A PyMC Labs test with Colgate-Palmolive researchers backs that split: across 57 surveys, AI purchase-intent ratings tracked human ratings 90% as closely as a repeat human survey.
See how we find needs customers haven't voiced โ Jobs to be Done
Read the full article in Forbes: Insights Teams' Hardest New Job Is Doubting Plausible Answers [FORBES LINK 1]
Are Your Survey Respondents Real?
Not always, and AI has made fakes much harder to spot. Fake answers reach market research surveys from bots, from people who let AI answer for them and from suppliers who cut corners.
A senior market researcher in the tech industry caught a bot farm only because an odd share of her sample seemed to live in North Carolina. Fakes have improved since. Dartmouth political scientist Sean Westwood built an AI agent that passed 99.8% of survey attention checks, at about five cents per response versus the $1.50 a human typically earns. In August 2026, Pew Research Center found that trap questions flagged 18% of an opt-in online sample as likely bogus. And in 2025, federal prosecutors charged eight people over a scheme in which two U.S. panel companies billed clients more than $10 million for fabricated data.
What leaders do depends on the source:
Against bots, look across the whole dataset. Michelangeli trained Claude to clean a 12-country tracking study, flagging AI-written answers and comparing each wave by demographic group and sample vendor.
Against AI-assisted respondents, know who is answering. Clients increasingly ask for communities of verified members.
Against bad suppliers, audit what you used to assume. "Suddenly we now have to check our vendors on things that we used to assume they were doing right," says the tech-industry researcher.
Quality now costs more, but Engineer reminds leaders that $50,000 spent on research can save $2 million to $4 million later.
Read the full article in Forbes: Insights Teams Can No Longer Assume Their Respondents Are Real [FORBES LINK 2]
The economics of fake answers
An AI agent can complete a survey for about 5 cents. A real respondent typically earns $1.50. At 30 times cheaper, fraud scales by itself. The problem isnโt only bots: in a 2024 study of about 800 participants on the research platform Prolific, 34% said they had used AI to answer open-ended questions.
Who Checks the Answers When Anyone Can Ask AI?
The insights team, working upstream. When colleagues can self-serve answers built on its research, the team's job shifts from answering each question to governing how questions get answered, like an editor.
At one major consumer goods company, people consult an AI-enabled research library about 30 times as often as they once called researchers. They trust those answers because they draw on the insights team's work, though no researcher checked them. Leaders guard against three failures:
Answers that please the asker. The tech-industry researcher's team lets its persona bots answer only from observable data. Beyond that, a bot must flag its reply as a hypothesis. "You cannot cite this as research," she says.
A partial view of the data. At a major pharmaceutical company, commercial, medical and R&D teams each see data the others can't, so one person's AI tool may give "a version of the truth, but maybe not the complete version," says an insights executive there.
Conflicts smoothed over. McCain chose an AI assistant built for research that calls out when sources disagree.
Good editors also test before they publish. Engineer's team quizzed a new assistant on questions it already knew the answers to, and sent it back to the vendor when replies didn't match. At Just Eat Takeaway, research leader Naoil Sbai's team labels findings by confidence and reviews the questions colleagues ask. "Research has never been about finding answers," she says. "It's always been about reducing uncertainty."
Read the full article in Forbes: Insights on Tap Still Needs an Editor [FORBES LINK 3]
An Octopus Organization in miniature
How Will Junior Researchers Learn Judgment?
On purpose. AI now does much of the grunt work that once taught researchers their craft, so leading teams are building judgment through a deliberate new apprenticeship.
Stanford economists found that employment of 22- to 25-year-olds in the jobs most exposed to AI is about 19% below where it would be had it kept pace with less-exposed peers, an early indicator rather than proof of cause. Young workers lost ground in jobs built on textbook knowledge, while experienced workers in jobs built on tacit knowledge gained. Judgment is tacit knowledge. Michelangeli fears that newcomers who learn only with AI won't develop "that judgment of, yes, I'm going to take this, but I'm not going to take this other thing that AI is telling me to do."
What leaders do:
Do the first pass by hand. Michelangeli drafts her own conclusions before turning to AI. Fetzer, who also teaches market research at Rutgers, makes students run qualitative interviews themselves.
Make the checks explicit. Benoit built a custom GPT that produces a chart book first. Researchers edit it before the final report, so each step forces a judgment.
Let the teaching run both ways. "There's no experts yet," says Benoit. Juniors show seniors what AI can do; seniors teach juniors when to trust it.
Read the full article in Forbes: Junior Researchers Need a New Apprenticeship [FORBES LINK 4]
Why the grunt work mattered
Amanda Michelangeli learned research by doing every step herself over 15 years. That repetition built the instinct to know which AI suggestion to take and which to leave. Newcomers now skip many of those steps, so teams have to build the same instinct on purpose.
How Is AI Changing UX Design?
It is moving the experience off the screen. As customers hand tasks to AI agents, UX design is shifting from arranging screens to shaping how answers and agents behave.
In a 2026 Gartner survey of more than 3,500 customers, people were about three times more likely to use a general AI tool like ChatGPT than a company's own chatbot in their latest service interaction. Yuhgo Yamaguchi, who leads UX at Fidelity, suspects that "the days of doing information architecture design on a website, that may already be behind us."
What UX teams design now:
Behavior, including restraint. When should an agent ask, when should it act and when should it hold back?
Experiences for a second kind of user. In a small test highlighted by usability pioneer Jakob Nielsen, AI shopping agents completed 89% of tasks on a store with structured product data, versus 49% on one built only for people. Nielsen argues user testing now needs to include agents.
What stays human. Yamaguchi points to deep customer understanding and taste, which he compares to "picking an emerging stock." AI trained on history struggles to predict what will be popular next.
Read the full article in Forbes: UX Design Is Moving Off the Screen [FORBES LINK 5]
How Is AI Changing Customer Experience Teams?

What Ties the Six Shifts Together?
AI makes data, answers, and designs easier to produce, which raises the value of judgment and of acting on what you learn. Three lessons hold across all six shifts.
Judgment is the new bottleneck. We found the same pattern when we interviewed product managers about AI โ How AI Is Transforming Product Management: 3 Major Shifts. When production gets cheap, deciding what to trust becomes the scarce work.
AI sees what has already happened. It struggles with disruptive concepts, emerging taste and needs customers haven't voiced. Those unmet Jobs often point to new markets, and finding them still takes conversations with real people.
Test before you trust. McCain quizzed its assistant before launch, and Bank al Etihad makes chatbot vendors compete. Side-by-side tests with real people show where AI holds up.
The most important findings may look wrong at first. Fetzer's looked wrong twice before it proved right.
Judgment is the new bottleneck
AI sees what has already happened
Test before you trust
Ask Yourself Eight Questions Before You Trust an AI-Generated Insight
How many people actually said this?
Would a distracted customer notice what the AI noticed?
Would you believe this result if it surprised you?
What can't the model see, such as a disruptive idea or an unvoiced need?
Are the respondents verified people, and did any let AI answer for them?
Which sources did the answer draw on, and do they disagree?
Is it labeled as research, or as a hypothesis to test?
Has the tool been tested on questions whose answers you already know?
About the Research
In 2026, New Markets Advisors interviewed [NUMBER] insights, UX and CX leaders in pharmaceuticals, fashion, food, financial services, technology and research agencies. Steve Wunker, Managing Director of New Markets Advisors, wrote the six-part Forbes series:
Insights Teams' Hardest New Job Is Doubting Plausible Answers [FORBES LINK 1]
Insights Teams Can No Longer Assume Their Respondents Are Real [FORBES LINK 2]
Insights on Tap Still Needs an Editor [FORBES LINK 3]
Junior Researchers Need a New Apprenticeship [FORBES LINK 4]
UX Design Is Moving Off the Screen [FORBES LINK 5]
AI Is Turning CX From Listening Into Fixing [FORBES LINK 6]
Resources
our Industries
Trusted by Global Leaders for AI Strategy
Our growth strategy clients range from international leaders to promising startups across our four key industries:

Frequently Asked Questions
What is the future of market research with AI?
Answers get cheaper and judgment gets more valuable. Leading insights teams will spend less time producing data and more time vouching for it, governing self-serve tools and turning findings into fixes.
































