Most brands are asking the wrong questions about AI research

AI adoption within market research is moving fast. For years, the central question for anyone commissioning research was a familiar one: qual or quant? Human-to-human interviews and focus groups, or surveys at scale? Or some combination of both?

Anyone who works in research will appreciate that even this has never been a straightforward conversation. There are always trade-offs around budget, depth and speed.

And now, two entirely new approaches have arrived to complicate that decision further.

First, "qual at scale", which replaces the human moderator with an AI one, enabling asynchronous interviews with real humans but closer to the speed and scale of quant. Second, synthetic research, where human participants are replaced entirely by AI-generated participants.

Naturally, when disruption arrives, we reach for familiar comparisons: is it faster? Is it as accurate? Is it really qual? etc.

But I think these are the wrong questions because they assume we're comparing like for like, which is not the case at all.

A big part of this problem is that many of the new wave of suppliers are presenting these tools as essentially the same as what came before, just faster and cheaper (i.e. ‘qual at scale’). A more helpful approach might be to start by understanding how these new methods actually work along with their benefits and their limitations>

This shifts the question to - when should I use this, and when should I prioritise real humans and skilled researchers?

To answer that honestly, we need to start with something the industry still under-appreciates: traditional research with real humans isn't perfect either.

A quick reality check on traditional methods

This isn't a takedown of conventional research. It's a necessary starting point.

Decades of behavioural science have established that we are not the rational, self-aware beings we like to think we are. The same applies when we sit people down as research participants. They post-rationalise decisions. They say what sounds acceptable. They predict their future selves poorly. They're shaped by question framing, group dynamics, and context in ways they're completely unaware of.

This doesn't make traditional research worthless, far from it. But it does mean that "we asked real humans" is not, by itself, a guarantee of validity and emphasises the point that the quality of insight depends enormously on how the research is designed and executed.

That context matters when we evaluate what AI-powered methods can and can't offer.

Qual at scale: valuable, but not what it says on the tin

Qual at scale is where the human moderator is replaced with an AI one and real participants take part asynchronously. The AI asks questions, probes responses, and synthesises findings across hundreds or thousands of interviews simultaneously.

The appeal is obvious: the openness of qualitative conversation, at the speed and scale of a survey.

But there's a fundamental tension here that doesn't get talked about enough. If you understand qualitative research properly, you'll know it doesn't use small sample sizes because of time or budget constraints. It does so because, beyond a certain point, there is very little value in adding more participants. Once the key themes have emerged and been validated across a well-defined sample, the returns diminish rapidly.

Human-led qual is deep, iterative, and human. It lives in the pauses, the contradictions, the thing someone almost said, and what their words, body language, and reactions reveal about how they'll actually behave. That's the difference between collecting quotes and genuinely decoding behaviour.

As far as I can see, we're still a long way from an AI moderator running 500 asynchronous interviews overnight being able to replicate that.

The core limitation: reactive, not instinctive

A skilled human moderator doesn't just follow a discussion guide. They notice when someone hesitates, contradicts themselves, or deflects — and they follow that thread. That in-the-moment judgment, built on experience and human intuition, is what generates unexpected insight. Current AI moderators can probe, but can we trust them to reliably recognise what's worth probing?

The self-report problem — potentially amplified

Much of the best qualitative research comes not from taking what people say at face value, but from working out what they're implying — reading body language, noticing pauses, sensing what's being avoided. Real qual is often the opposite of a straightforward Q&A.

This problem could be compounded by AI. A model trained to interpret and report responses confidently might consistently miss the hidden driver — the thing that required a different line of probing, or a particular context, to surface. We also lose body language entirely. A moderator without eyes, without human context — that's a significant amount of signal gone.

So what is it, right now?

A fast, scalable tool for directional qualitative exploration. Genuinely valuable for early-stage discovery, hypothesis generation, and identifying broad themes across large, diverse samples. Not a replacement for skilled human moderation where depth, emotional nuance, or high-stakes decisions are on the line.

It's also worth saying: this space is genuinely new. We're still working out where AI-moderated interviews produce insights that lead to better decisions — and where they produce plausible-sounding noise. The honest answer is that we don't fully know yet.

Synthetic research: modelling human reactions

While qual at scale still involves real human participants, synthetic research removes them entirely.

Think of it as modelling. Synthetic participants are built from a combination of data sources (large-scale surveys, social and behavioural data, demographic and psychographic profiles etc.) and trained to simulate how different types of people respond to different types of questions.

Again, the potential upside is obvious. No recruitment costs, no incentive payments, always on, instant feedback.

A compelling proposition. And in my opinion, for certain jobs, it’s a genuinely useful one. Particularly when you consider that most teams still do very little research at all due to time and budget constraints.

The most powerful use case right now as far as I see it, is early-stage screening: quickly stress-testing ideas, concepts, messaging, creative etc. before committing real time and budget. Used alongside human expertise and experience, it can rapidly surface early objections, new angles, and opportunities to refine.

But like weather forecasting, models aren't always right. And like the weather, modelling human behaviour is no mean feat, real decisions are shaped by emotion, context, and subconscious drivers that are hard to capture in training data. The gap between what a synthetic participant predicts and what a real person does in a real moment may narrow over time, but it's worth understanding it exists.

Used well, in the right context, and alongside human judgment, synthetic research opens up feedback loops that simply weren't previously possible for most teams. The key, as with any tool, is understanding the limitations and working within them.

The question isn't "AI or traditional research?"

The mistake is assuming every new method has to replace an old one.

Every research approach is, ultimately, a way of reducing uncertainty. Each comes with its own strengths, blind spots, costs, and assumptions. Human-led qualitative research excels at uncovering the hidden drivers of behaviour. Quantitative research helps us estimate how widespread those behaviours are. AI-moderated interviews make exploratory research faster and more accessible. Synthetic research has the potential to make testing accessible to many more businesses.

For brands, the opportunity isn't choosing between traditional and AI-powered research. It's understanding when each method is most valuable.

As a practical rule of thumb, today's AI-powered approaches are well suited to stress-testing early ideas, identifying promising directions, uncovering blind spots, and informing lower-stakes decisions—particularly where the alternative might have been doing no research at all.

As the stakes increase—whether it's a brand repositioning, packaging redesign, or major product launch—the case for involving real participants and skilled researchers becomes much stronger.

The brands that get the most value from these new tools won't be the ones that replace everything with AI. They'll be the ones that understand where AI creates an advantage—and where human insight remains irreplaceable.