Synthetic Audiences in Market Research: Hype, Reality, and Outlook for 2026
What Are Synthetic Audiences?
Synthetic audiences are AI-generated consumer panels, essentially virtual personas, that mimic the behaviour and preferences of real people for research purposes . Instead of recruiting a focus group or survey panel, a marketer can feed existing audience data (demographics, purchase history, survey responses, etc.) into an AI system which then generates stand-in respondents who behave statistically like the real audience . These synthetic personas can be queried as if they were real consumers, providing instant answers to questions about product concepts, messaging or other “what if” scenarios. Importantly, a synthetic audience represents segments or archetypes (e.g. “urban Gen Z tech enthusiasts”) rather than specific identifiable individuals, so it leverages patterns in the data without exposing any one person’s information. Synthetic research uses AI to generate new data that mirrors real customer insights, offering a virtual test audience on-demand . This concept has rapidly emerged in 2024–2025 as generative AI technology improved, with major brands and research firms beginning to experiment with it in their innovation and insights processes.
How Altair have been using Synthetic Audiences
We have been building and experimenting with synthetic audiences for a little while now.
Each is built on real life data, to ensure we are not relying on LLM behaviour. In fact, when our AI teams are building new synthetic audiences a key step in the process is to put strict guardrails in to stop the LLM using its own knowledge rather then the vast amount of qualitative data we are feeding it.
We have had great success and a few nice surprises along the way when using our audiences to:
- Critique our media strategy
- Understand how they consume different media channels
- What they feel about our creative
- What motivates them, including why and when
When using synthetic audiences it's important, as with anything AI, that it is built with human insight and the last mile is always with the human.
We use a range of synthetic audiences as a focus group to ensure rounded insight too, which is extremely useful.
Each synthetic audience is built bespoke. We have client specific synthetic audiences as well as broader synthetic Audiences we use to give wider insights.
Interestingly they have proved effective, whether for our B2B clients or sustainability clients or our arts and culture clients.
As they all follow the rules of:
- Quality data in
- Strong guardrails to remove LLM learning swamping the structured data we have inputted
- Strong refinements before release
- The human does the last mile
Here is an example from our culture, heritage and tourism synthetic audiences:
Meet Elizabeth - Elizabeth lives in Harrogate and is a member of multiple cultural and arts institutions and is fuelled by 5 years worth of bespoke research and qualitative data.
We asked Elizabeth:
What makes you choose a live cultural experience over staying at home? Is it discovery, connection, or the sense of supporting something meaningful?
Key Benefits: Speed, Scale and Savings
The appeal of synthetic audiences largely comes down to speed, scalability, and cost-efficiency. Traditional focus groups or surveys can take weeks to recruit, field, and analyse. In contrast, querying a well-built synthetic audience yields insights almost instantly – responses in seconds or minutes rather than days. This acceleration means marketers can test ideas on Monday and have feedback by Tuesday, which is invaluable in fast-moving sectors like entertainment or retail where waiting weeks might mean missing a trend.
Cost is another major factor. Running large focus groups or panels is expensive, often involving incentives, facility rentals, and researcher hours. Synthetic research, once the system is set up, only incurs the software cost and can be used repeatedly. The speed and cost advantages are especially important for non-profit or niche fields (e.g. heritage organisations with limited research budgets), where these efficiencies are a compelling advantage.
Beyond raw speed and cost, scalability and flexibility set synthetic audiences apart. There’s no risk of respondent fatigue, you can ask a virtual audience 100 questions, or re-run the same question in 10 different ways, without participants dropping out or getting tired.
Need to test five different poster designs for an upcoming theatre festival?
A synthetic panel can score or react to all five, anytime, without scheduling conflicts.
Planning to test multiple value-prop statements for your enterprise SaaS solution?
A synthetic panel can review and rank each option on demand, without waiting for hard-to-book stakeholder calls. This “always-on” availability also means research can happen outside of normal hours, useful for global marketing teams working across time zones.
Synthetic audiences can also represent hard-to-reach or specialised segments much more easily. For instance, if you require feedback from hard-to-reach roles, for example CISOs in fintech or IT directors evaluating cloud migration, synthetic audiences can model their viewpoints instantly, avoiding the logistical challenge of securing time with these decision-makers. This democratises insights that previously only large companies with big research budgets could obtain.
Encouragingly, recent studies suggest that synthetic respondents often produce insights comparable to traditional research. In one 2025 experiment, AI-generated “digital twins” of readers provided answers that matched a real survey’s results with 94% accuracy in one case. Another research team found that after training on a person’s data, a digital twin’s survey responses replicated the human’s own answers nearly as closely as the person’s answers matched themselves in a test-retest scenario. While not perfect, these outcomes indicate that, when built on robust data, synthetic audiences can mirror real consumer opinions to a high degree. In short, they’re proving to be a useful early indicator: a way to weed out weaker ideas and highlight promising ones quickly, which can then be confirmed with real people.
Limitations and the Need for Human Insight
Despite their promise, synthetic audiences come with important caveats. First, they are only as good as the data and models behind them. “Garbage in, garbage out” applies: if the underlying data is biased, unrepresentative or outdated, the AI’s responses will be too . For instance, researchers have shown that some generative models today skew toward the perspectives of younger, more educated and liberal demographics, likely a reflection of their internet training data, and underrepresent views of older or more conservative groups . If a synthetic audience isn’t carefully trained on diverse, relevant consumer data, it might give a false or skewed view of your market. An AI might miss nuances important to minority or local communities if those voices weren’t in the training mix .
Another limitation is plausibility vs. reality. AI responses can sound very confident and detailed, which is impressive, but sometimes they may be convincingly wrong. NielsenIQ cautions that many rushed synthetic feedback tools generate outputs that “pass a gut check” but aren’t backed by real evidence . Basically, an AI panelist might give you a neatly phrased explanation for why customers would love your new eco-friendly product, but that doesn’t mean real customers actually would. This risk of hallucination or simply incorrect inference means marketers must apply a healthy dose of skepticism. No one should be replacing their market research department with ChatGPT, precisely because current AI can confidently make up information . Human oversight is needed to discern insight from illusion. Always run the last mile yourself.
Synthetic audiences also struggle with predicting new behaviour or “unknown unknowns.” They excel at interpolating within known patterns (e.g. projecting likely opinions based on past data), but if you’re introducing something truly unprecedented, a groundbreaking theatre experience or a new piece of SaaS technology, AI might not anticipate reactions well since it has no precedent to learn from. Human creativity and intuition are still essential to gauge those leaps.
Finally, ethical and creative judgment calls require humans. An AI might tell you a highly provocative ad will get attention, but a decent marketer will consider brand reputation, context and ethical implications that an algorithm might not grasp. Synthetic audiences can’t (yet) fully comprehend cultural nuance or moral values beyond what’s in their data. Hence, the “last mile” of insight, interpretation, strategy, and implementation.