WTF Are Synthetic Audiences?
There is a particular relief that comes with not having to talk to people. Any practitioner who has sat through a difficult focus group, who has watched a beautifully reasoned brief crumble against the actual opinions of actual humans, knows this feeling. Real research is inconvenient. Real people are inconsistent. They contradict themselves, misremember what they wanted, change their minds between Tuesday and Thursday, and occasionally tell you things you would genuinely rather not hear.
So when the tools arrived that promised to remove all of that — to simulate human behaviour at scale, without the friction, without the awkward silences, without the need to drive to a rented room in a suburb you don't know — it is not hard to understand why the industry reached for them.
Synthetic audiences are, in the plainest terms, AI-generated stand-ins for real people. Fed on vast datasets of behavioural patterns, they can be conjured in their thousands to predict what a market segment would think, feel, or do in response to a piece of creative, a product launch, a price point. Where once you might have recruited two hundred participants over three weeks, you can now generate twenty thousand virtual surrogates in minutes. The numbers are convincing. The speed is genuinely impressive. And somewhere in all of that efficiency, something quiet gets lost.
We watch this from the sideline…as a studio that still believes, perhaps unfashionably, in the irreducible value of real encounter. We haven't deployed synthetic audiences in our work. But we've watched what happens in the wider culture of brand practice when research loses its friction, and we think it's worth naming what we see.
“The synthetic audience isn’t the problem. It’s the symptom.”
What the enthusiasm for these tools reveals is a way of thinking about people that has been gathering momentum for much longer than the tools themselves. The turn toward optimisation…toward cleaner data, faster cycles, more predictable outcomes…didn't begin with AI. It began when brands started treating human beings primarily as data-generating entities rather than meaning-making ones. The synthetic audience is just the logical conclusion of that drift: a world in which the customer has been so thoroughly abstracted that we can replace them entirely with a convincing simulation and feel, honestly, like we've improved the process.
This is where the danger lives. Not in the technology itself, which is genuinely capable and genuinely useful in certain contexts. The danger is in what the relief signals. When a team is grateful to be freed from the inconvenience of real customers, that gratitude is worth examining. Because the inconvenience was never incidental…it was the point. The contradiction, the unexpectedness, the thing the customer said that didn't fit the brief: that is where brands find the edges of their own understanding. That is where the real work begins.
What synthetic audiences are constitutionally unable to provide is surprise. By definition, they reflect patterns that already exist in the data they were built from. They are, in a meaningful sense, the past — shaped, smoothed, and made legible. Real people are something else entirely:
They are perpetually ahead of what you think you know about them
They contradict the patterns you've built your strategy around
They reveal the gap between what a brand understands and what it still needs to learn
They surprise you in ways that no simulation, however sophisticated, is designed to do
We think about this in terms of what we'd call the texture of real encounter…the quality of information that only comes from genuine contact. The slight hesitation before someone answers. The way a person's face changes when they're being polite rather than honest. The offhand remark at the end of a conversation that turns out to be the most important thing anyone said all day. None of that survives the abstraction into data. None of it can be synthesised. And yet it is often exactly that quality of information that separates a brand which truly understands its people from one that merely knows its metrics.
“The inconvenience was never incidental — it was the point.”
There is a version of synthetic audience use that is thoughtful and bounded…as a rapid pressure-test before real research, as a way of stress-testing hypotheses before you bring them to actual humans. We don't think that's where most of the enthusiasm is heading. Most of the enthusiasm is heading toward replacement: faster, cheaper, scalable, no awkward silences. Which is understandable. And which is, we think, a slow erosion of something a brand cannot afford to lose.
The brands that hold their ground here — that maintain the discipline of genuine human encounter even as the tools make it easier not to…won't be doing it because they're romantic about methodology. They'll be doing it because they understand that the knowledge which matters most, the knowledge that actually compounds into brand distinctiveness over time, comes precisely from the places that resist optimisation. From the friction. From the inconvenience. From the very real, very messy, very unscalable fact of other people.
What are synthetic audiences?
Synthetic audiences are AI-generated groups of virtual people, built from large datasets of real behavioural patterns. Rather than conducting research with actual participants, brands can use these tools to simulate how a market segment might respond to creative, messaging, pricing, or product decisions…at scale and near-instant speed.
They are not real people. They are statistically convincing reflections of how real people have behaved in the past.
What they can do
Simulate predicted responses, map hypothetical behaviours, pressure-test creative concepts quickly
What they can’t do
Surprise you, contradict your assumptions, or reveal what your brand still needs to learn
The real risk
Not the technology itself… but using it as a reason to stop talking to actual people
