What's the Real Risk in Using Synthetic Data for Market Research?

Here's a question worth asking before your next planning meeting: if an AI can generate a hundred synthetic respondents in the time it takes to book a conference room, what exactly are you paying your research team to do?

It's not a trick question. And it's not meant to be alarming, either. But it's the question a lot of insights leaders are quietly avoiding, because the honest answer requires admitting that the job is changing shape.

Dale Evernden, EVP of Innovation and Design at Rival Technologies, takes this on directly in his piece for Research World, Human context in an age of synthetic research. He skips the "will AI replace research" debate almost entirely. Instead, he gets specific about where the real risk actually lives: not in whether synthetic data is good or bad, but in what happens when nobody's maintaining the human understanding it depends on.

Why the old constraint on research doesn't exist anymore

Think back to how innovation used to work. Building a prototype took months. Testing a new message meant weeks of production before you even got to the research phase. That slowness was annoying, sure, but it also acted as a natural filter. Only a handful of ideas ever made it far enough to need testing.

That filter is gone now. Dale shares a line that stuck with him: "It costs less to actually build a prototype and show it to customers than it does to hold a meeting about whether you should build it in the first place."

Is that literally true everywhere? Probably not. But it captures something real. Teams can spin up dozens of concepts before lunch. So the bottleneck isn't creation anymore. It's judgment. Somebody still has to decide which of those ideas are worth a real human's time, and that's not a task you can fully hand off to a model.

So what are synthetic personas actually good for?

Good question, and the answer might surprise you: not for replacing your participants. For pressure-testing your thinking before your participants even show up.

At Rival Technologies, Dale's team builds synthetic respondents from segment personas specifically to surface edge cases early. Where might this concept fall flat? What objections haven't we considered yet? Where are we about to waste real research time asking a question we could've answered ourselves?

Used this way, synthetic tools don't shrink the role of research. They sharpen it. That's a very different outcome than quietly automating the parts of research that need a human eye, and it's the difference between using AI well and using it lazily.\

The actual risk with synthetic data: it sounds right long before it is right

This is the part of the article worth sitting with, honestly.

Synthetic models are good at sounding sure of themselves. That's kind of their whole design. Which is exactly what makes them useful for early-stage pressure-testing and risky for anything with real stakes attached. A synthetic respondent can hand you a smooth, confident answer that has nothing to do with how an actual customer would react.

Dale calls the slower version of this problem "context rot." It happens when teams feed research findings, support tickets, and community conversations into a system, then stop paying attention. Nobody updates it. Small inaccuracies pile up quietly. Before long, the AI is drawing from a version of your customer that no longer exists, and it still sounds completely convincing while doing it.

Not every decision needs the same amount of proof

Here's a helpful gut check from the piece: match your evidence to your stakes. Using a synthetic persona to poke holes in an early concept? Low risk, go for it. Using synthetic output to justify a seven-figure investment? That's a different conversation entirely, and it needs real human signal in the mix.

This is where a connected system really pays off. Picture an insight community, a well-maintained knowledge base, and a synthetic layer, all feeding each other instead of working in isolation. Synthetic tools explore and narrow the field. The community validates what's actually true. That feedback updates the knowledge base. And the next round starts smarter than the last.

The insights job isn't going away. It's splitting into two new ones

So where does that leave researchers? AI already changed the job description of insight professionals — and synthetic data will only accelerate the evolution. 

Dale sees two responsibilities becoming central to the role, and neither one existed in quite this form a few years ago.

The first is stewardship: someone has to own the quality and freshness of the human context these AI systems rely on. The second is translation: AI can summarize a pattern in seconds, but someone still has to explain what that pattern actually means and what to do about it.

As Dale puts it: "Maintaining that human context is becoming one of the core responsibilities of modern insights teams."

That's not a lesser job. If anything, it asks more of researchers than before, because it combines the methodological rigor the role always required with a brand-new discipline: keeping AI's inputs honest over time.

So back to that original question. What's the risk of using synthetic data in market research? It's not that the technology is bad. It's what happens when the human context behind it goes stale and nobody notices until it's too late. The teams getting this right aren't chasing faster synthetic output. They're treating their customer knowledge as something that needs constant upkeep, because everything else gets built on top of it.

As one of the top market research trends in 2026, synthetic data will continue to shape the discussion in our industry. For business leaders, now is the time to consider how this new way of generating insights affects research and how and when we engage real humans. 

Want to learn more about synthetic data and how it will change the future of market research? Check out our Q&A with Dale here.