By Mark Amin, Co-Founder and Lead Researcher at xplorit.io with 30 years experience in market and user research in Asia-Pacific
Here’s an uncomfortable truth about the application of AI in market and user research: most of it was built by people who have never designed a research study in their lives. Not a criticism. Just an observation — especially if you’re about to make a marketing decision based on what one of those tools tells you.
Speed is not the same as rigour
The pitch is seductive. Instant insights. Automated surveys. AI-generated reports. All in minutes. And for certain tasks — transcription, sentiment tagging, thematic coding at scale — AI can deliver them too. It’s fast, it’s tireless, and it doesn’t need a coffee break between tasks.
But automating research tasks is not the same as applying research thinking.
The difference? Tasks are repeatable. Thinking is contextual. A task says: code these 500 responses into themes. Thinking asks: are we even asking the right question in the first place? One is a workflow problem. The other is a judgement call — and judgement calls cannot be automated away.
The most important work in any research project happens before the data collection begins. It’s in the framing of the hypothesis, the design of the questions, the selection of the method. Get those right, and even modest data can produce breakthrough insights. Get them wrong, and no amount of AI processing will save you.
Why commercially motivated tools are incentivised to oversimplify
Most AI research platforms were built with a clear go-to-market proposition: make research faster and cheaper. That’s a legitimate ambition. But it creates a structural incentive to oversimplify.
Nuance is expensive. Ambiguity takes time to resolve. Exploratory research is hard to package into a clean UI. So the tools get designed around the things that look like research — dashboards, auto-generated reports, “confidence scores” — rather than the things that make research reliable.
The result is a generation of tools that produce outputs that look authoritative but aren’t. Plausible-sounding answers to badly framed questions. Clean charts built on shaky methodological foundations. And critically, no flag to tell you when the framing was the problem.
This is not a hypothetical risk. AI research tools have been found to generate correlations between variables that are wrong approximately half the time when compared to gold-standard survey data. That’s not a rounding error. That’s a coin flip framed as an insight.
What “researcher-led AI” actually means
When we say xplorit.io is researcher-led, we mean something specific.
It means the platform was designed by people who have sat across the table from a sceptical CMO and had to defend a methodology. People who have redesigned a study at 11pm because the original brief was pointing at the wrong question. People who have read enough bad research reports to know exactly which shortcuts produce which failure modes.
That experience shapes everything:
Question design. We know the difference between a leading question and a projective one. Between a scale that measures the right thing and one that produces satisfying but misleading data.
Hypothesis framing. Good research starts with a hypothesis worth testing — not a confirmation of what the client already believes. This requires intellectual rigour and experience.
Interpretation of ambiguity. Data is almost always ambiguous. The skill is in knowing which ambiguity matters and why. AI excels at finding patterns. Researchers decide which patterns are meaningful.
Researcher-led AI doesn’t slow the process down. It makes the speed worthwhile.
The marketing consequence: bad research is expensive
Let’s talk about what’s actually at stake.
Research is cheap compared to the decisions it informs. A brand campaign. A product launch. A positioning shift. A new market entry. These decisions routinely involve six-figure (and often seven-figure) investments. The research that informs them might cost a fraction of that.
But if the research is poorly framed — if it asks the wrong questions, recruits the wrong audience, or interprets findings through the wrong lens — the cost isn’t the research budget. The cost is the decision made on the back of it.
Coca-Cola’s 1985 “New Coke” disaster is the canonical example: taste tests confirmed a sweeter formula preference, but the research failed to account for the emotional attachment consumers had to the original. The brand spent millions on a product the research “validated” — and was forced to reverse course within months.
The lesson isn’t don’t do research. It’s the quality of the question determines the quality of the answer. And while you can automate questionnaire design, without first knowing why a question should be asked, the response can be misleading, invalid or unusable.
In 2026, 83% of enterprise leaders still require human validation before acting on AI-generated insights. That’s not technophobia. That’s the market correctly identifying the gap between what AI tools produce and what decision-making actually requires.
The 17% who act on AI outputs without validation? We genuinely hope they’re asking the right questions first.
What to look for when evaluating an AI research platform
Not all AI research tools are equal, and the difference isn’t always visible in a demo. Here are the questions worth asking:
1. Who built the methodology? Engineers build for speed and scale. Researchers build for validity. Ask which team made the foundational decisions about how the tool frames questions and interprets data.
2. How does the tool handle ambiguity? If it always produces a clean, confident output, be suspicious. Good research surfaces uncertainty. A tool that never says “we’re not sure” is either very good or hiding something.
3. What does the output actually tell you to do? Data without decision-direction is decoration. The mark of researcher-led thinking is a finding that connects clearly to a decision.
4. Is the research design customisable? One-size-fits-all templates produce one-size-fits-all insights. Real research is designed around the specific question, not the nearest default.
5. Can you see the reasoning, not just the result? Interpretive transparency is a hallmark of quality research. If the tool can’t show its work, it can’t be trusted.
The bottom line
AI in market research is not the problem. AI without research expertise is.
The platforms that are genuinely transforming how brands understand their customers aren’t the ones that move fastest. They’re the ones that bring genuine methodological thinking to every stage of the process — and then use AI to do it at scale.
That’s what xplorit.io was built to do. By researchers. For the decisions that matter.
Ready to see what researcher-led AI looks like in practice?
Footnotes & References
xplorit.io is an AI-powered market and user research platform built by experienced researchers — not technologists.
We exist at the intersection of research rigour and AI efficiency, because your decisions deserve both.
1. https://www.linkedin.com/posts/jcmecke_airesearch-saas-marketresearch-activity-7425164666940661760-_4vs
2. https://researchworld.com/articles/beyond-the-hype-what-ai-can-and-can-t-do-for-market-researchers
3. https://www.mordorintelligence.com/signal/insights/ai-only-market-research-risks
4. https://www.linkedin.com/pulse/market-research-mistakes-cost-brands-millions-how-smart-fblye