A focus group of synths just blew my mind
Opinion
Synthetic research can be a game changer for media strategists — so long as its limitations are taken into account.
If you’d told me five years ago I could have a focus group of my choosing, on standby and ready to respond within minutes, I’d have looked at you with a mix of incredulity and wild intrigue.
But that was before the generative AI and large-language model (LLM) revolution. Today, I can brief a virtual panel, generate personalised personas, and return with a set of interview transcripts and actionable insights in under ten minutes.
It’s called synthetic research and, despite feeling a sense of caution about its use, it holds clear potential for any strategist.
It’s also one that arrives just when we need it most. As any strategist can attest, our craft remains under pressure. Timelines can be short, budgets are often under scrutiny, but insight remains absolutely crucial. Traditionally, the answer would lie in getting up close and personal with a focus group, or ideally some ethnography, but they can be pricey and take time, which means they’re often reserved for the biggest briefs with ample lead time.
Synthetic research changes that. By simulating panels of real people and generating detailed responses in minutes, it allows strategists to work faster, with greater insight and confidence.
Does this sound too good to be true? I’ve been testing various systems over the last five months and, honestly, I’ve been blown away at times. But how does synthetic research actually work? What are its limitations? And how might its adoption change the strategy process?
Prompts for deeper thinking
Synthetic research uses generative AI to simulate qualitative interviews with lifelike user personas. These are not real people, but data-informed constructs — in true sci-fi fashion, they’re referred to as “synths” or “generative agents” — trained on enormous collections of behavioural and attitudinal information, from forums and review sites to social sentiment and ethnographic studies.
Users input a research goal and an audience description. The system then returns a set of full qualitative interviews, each with a distinct persona, a set of inferred insights, and, if needed, a structured strategic summary. You can start with a research goal and let the AI create a set of interview questions or you can hardwire in your own custom questions. Or you can even test concepts.
In one example my team tested, we wanted to understand new and emerging trends with those interested in luxury and fashion. It highlighted things like the vibe shift from quiet luxury to boom boom that trend forecaster Scott Monahan predicted earlier this year. “Emma Thompson” (a synthetic university student aged 19 from Manchester) told us, “Boom boom gives me similar energy to that mob wife trend — just unapologetically extra, but with even more fun or chaos.”
Equally, when we wanted to understand the notoriously difficult-to-find high-net-worth audience and their motivations, it created synthetic personas like “James Rothchild-Smith” (a 52-year-old hedge fund manager who lives in Mayfair) who gave us the great quote: “A car isn’t just transportation — it’s practically a business card on wheels.”
These are very small examples. But the output wasn’t just plausible. It was fast, nuanced and full of prompts for genuinely deeper, strategic thinking. It moved us quickly to where we needed to be, and in surprising but entirely workable ways. “Unapologetically extra” and “a business card on wheels” are great creative stimuli.
This is particularly useful because it’s becoming less common to plan in quarterly cycles. Messaging evolves across the year, briefs arrive with less notice, and often demand more precision. Synthetic research makes it possible to explore smaller, more tactical questions and with hard-to-talk-to audiences. Planners can test creative mechanics with highly specific audience types. They can explore regional variations or price sensitivities without commissioning a full study. It opens up the possibility of running insight loops throughout the planning or campaign cycle, rather than once at the start.
There are practical benefits too. Cost is a major factor. One synthetic interview might cost a tenth of a traditional one. That changes the economics completely, especially for mid-sized or fast-turnaround briefs.
And there is speed: what once took three weeks can now be done in under an hour. That means we’re finally loosening the traditional tension between quality, speed and cost — the familiar triangle that usually forces a trade-off.
What makes this more than just a party trick or digital mirage is the science behind it. The most sophisticated synthetic research systems I’ve tested don’t rely on a single AI model. They pull from multiple language models, layer in psychographic and attitudinal datasets, and can also, if necessary, use a technique called retrieval-augmented generation (RAG) to ground responses in relevant real-world sources. They are also often benchmarked against traditional panel data to ensure accuracy.
In some cases, the parity between “synthetic” and real-human output (which in another sci-fi nod is usually referred to as “organic”) is now measured at 90 percent.
Not a substitute to traditional research, but additive
Despite that, it doesn’t make synthetic research a substitute for human insight. It can’t replicate the true nuance of any lived experience. It doesn’t catch the self-awareness in a research session, or the revealing aside that changes a creative brief; those eureka moments when you uncover something new about human behaviour.
It perhaps also lessens the chance of that magical serendipity that has defined some groundbreaking campaigns throughout the history of advertising. And biases, long-established as a risk for all AI systems, should also give us some pause before diving in headfirst. We do not want to risk amplifying skewed assumptions or structural imbalances baked into the datasets that synthetic research draws upon.
However, what it can do is prompt better questions. Over the years, strategists have sometimes become distanced from the art of proper research, yet synthetic research is not only a great way to re-engage teams with the discipline, but it also enables them to fine-tune how they think about interviews and eliciting sparkling insight. It can act as a flexible way of adding to overall processes; whether as the start, mid or end point.
In that sense, I believe something like the 80/20 rule should apply. Synthetic research delivers rapid inputs across the process, while traditional methods should be reserved for cultural depth, emotional resonance and giving a face and a voice to the work.
Looking ahead, I expect to see proprietary synthetic panels emerge, shaped by academic stringency and calibrated with live-market data. It’s certainly easy to imagine strategy teams embedding these tools within their planning routines, combining them — in a distinctly hybrid model — with real-human inputs to create faster, more adaptive workflows.
Traditional research providers may view this with concern or perhaps scepticism. But I don’t see synthetic research as something set to undercut them. Rather, I see it as supporting the reality of how we now work; it allows insight to be applied more widely, more often, and more imaginatively. It gives planners tools to test, refine and learn without waiting for a specific budget or a sign-off. There’s a reason Hearts & Science has an ampersand in its name; the best stuff happens when we combine the two.
The best thing we can do now is test the systems and look for faults and limitations, and learn exactly how to best use these tools. That will, with absolute necessity, require real human data — from real-life, not online — to effectively train the models and ensure we approach this future with a foundation based on reality.
But in the meantime, I do have some questions I must pose to my new synth friends…
Simon Carr is chief strategy officer at Omnicom Media Group agency Hearts & Science.