What Are Synthetic Audiences in Marketing?

What Are Synthetic Audiences?

Imagine having a focus group available at 2 a.m. No scheduling. No recruitment budget. No waiting weeks for a debrief.

That's the basic idea behind synthetic audiences.

A synthetic audience is a virtual panel created by training AI models on real human behavior data. The information comes from sources like CRM systems, surveys, customer interactions, and publicly available datasets. The result is what researchers call a "digital twin" of an audience segment: a model that mimics how real people think, respond, and behave.

The concept emerged as a distinct practice in 2024 and 2025, as generative AI improved enough to make the outputs actually useful. Major brands and research firms began experimenting with it in their innovation and insights processes. The terminology is still settling. Synthetic data, synthetic research, and virtual panels are all used interchangeably, but they generally refer to the same thing: using AI to generate data that mirrors the real audience it's modeled after.

Here's the thing. This isn't some fringe experiment. The AI-based research services market is projected to grow from USD 7.97 billion in 2025 to USD 35.4 billion by 2035, at a compound annual growth rate of 16.1%, according to Future Market Insights. The acceleration is already underway.

How Synthetic Audiences Work: Data Sources and AI Models

The engine behind synthetic audiences is a blend of diverse data and advanced language models. Understanding where the data comes from is important, because the output is only as reliable as what goes in.

The Three Data Layers

Advanced integration techniques, including Retrieval-Augmented Generation (RAG), allow brands to upload their own proprietary information directly into the model. This is what makes the resulting personas feel specific rather than generic.

The AI powering these platforms includes foundation models like GPT variants and refined generative architectures. Electric Twin, for example, combines real-world survey data with large language models, social science research from the London School of Economics, and behavioral machine learning. The platform has run over 40,000 evaluations across 155 countries.

Cohort Boosting: Scaling Small Samples

One technique worth knowing is "cohort boosting," a term used by Cynthia Vega, Global Analytics and AI Director at Kantar.

Here's how it works. A company has 50 real survey participants. That's not enough to test across different market segments. Cohort boosting uses AI to expand that dataset to 500 synthetic participants, extrapolating behavioral patterns while keeping the statistical integrity of the original group.

The resulting personas don't rely on real people's personal information. They're built on probabilistic data, which means they address privacy concerns while still producing actionable insights.

GWI, the consumer intelligence platform, takes this at scale. They currently give brands conversational access to around 7 million digital twins, all validated against real consumers as a ground truth.

Real-World Use Cases and Examples

Synthetic audiences aren't theoretical anymore. Publishers, banks, and global agencies are using them right now for decisions that actually matter.

The Times: Editorial Planning and Podcast Naming

News UK, publisher of The Times and The Sunday Times, used Electric Twin to guide editorial product decisions. The team tested names for a new business podcast and asked what content would drive deeper engagement with different reader segments. They got fast, directional feedback before committing to any launch.

Electric Twin embedded with The Times for months, analyzing their audience datasets deeply enough to ensure the synthetic panel genuinely represented their readership.

US Bank: National Campaign Strategy with Supernatural AI

US Bank worked with Supernatural AI to build six synthetic audience profiles: young affluent, midlife affluent, high net worth, small business owners, corporate, and commercial customers. The profiles were built using multiple third-party data sources and trained across different AI models.

The team used these to develop the baseline strategy for their national "The Power of US" campaign, testing messaging across all six segments before production began.

"We did some tests with humans and the themes came back with a 90% to 95% overlap with the synthetic audiences."

Michael Lacorazza, CMO, US Bank (AdExchanger, 2025)

The speed-to-market for the project was under four months. A comparable process using traditional research would have taken six months or more.

Dentsu: Media Planning for Hard-to-Reach Audiences

Agencies like Dentsu use synthetic panels for media planning, particularly when targeting audiences that are expensive or difficult to recruit through traditional research. This allows creative and media teams to get directional feedback on campaigns before committing to production budgets.

Telecom Providers: Market Entry Without Brand Risk

A major telecom provider used synthetic audiences to test features and pricing for a new market segment. The goal: understand how to enter a new category without cannibalizing or damaging their premium brand. Running these tests on real customers first would have been too expensive and too risky.

Key Benefits: Speed, Cost, and Scale

Here's the truth about traditional market research. Surveys take weeks. Focus groups cost tens of thousands. You get one shot at the questions. And by the time the debrief lands, the market has already moved.

Synthetic audiences cut through most of that friction.

Speed

Alex Cooper, CEO and co-founder of Electric Twin, described the time advantage as "tens of thousands of times faster" with "almost instantaneous response" compared to traditional surveys (Digiday, 2025). The platform's LSE-validated research puts that at 10,000 times faster with 95% accuracy.

A human survey could take weeks from design to debrief. A synthetic panel can return findings in minutes.

Cost

Cooper also described synthetic research as "one-thousand times cheaper" than doing the same work with human participants. The agency OLIVER, which uses synthetic personas for client research, puts the cost at roughly 20% of conventional focus groups while delivering accuracy in the 70% to 85% range, per Futureweek (2025).

For initial idea screening, that math is compelling. You don't need 100% accuracy at the screening stage. You need to eliminate the worst five ideas and identify the best one. Synthetic audiences do that efficiently.

Scale and Unlimited Questions

Real focus groups suffer from survey fatigue. Ask too many questions and participants disengage. Synthetic panels don't.

Brands can ask unlimited questions to entire segments or to specific personas within those segments. They can test what happens if they change pricing, reframe messaging, or enter a new geography, all without recruiting a new group each time.

"72% of teams using synthetic responses report that their organizations depend on their research significantly more than a year ago."

Qualtrics 2026 Market Research Trends Report (surveying 3,000+ researchers across 14 countries)

The same report found that research teams not using purpose-built AI are four times more likely to lose organizational influence than those embracing it.

That gap is widening. Fast.

Synthetic Audiences, Privacy, and the End of Third-Party Signals

There's a structural reason synthetic audiences are gaining traction beyond just speed and cost. Traditional data-driven research is getting harder.

According to eMarketer, 67% of US adults have already turned off cookies or website tracking. Safari and Firefox block third-party cookies by default. Privacy regulations like GDPR and CCPA have made it increasingly difficult and risky to use real individual-level customer data for granular analysis. The Conversion Alchemy State of Synthetic Research 2025 report describes privacy as a "key driver of adoption" for this exact reason.

Synthetic audiences sidestep the problem entirely.

Because they're built from aggregated, probabilistic data rather than real personal information, synthetic profiles contain no personally identifiable information (PII). That puts them outside the scope of most GDPR requirements. They are, by design, fully compliant with GDPR, CCPA, and other major privacy frameworks.

Brands can explore new segments, test sensitive topics (like financial behavior or health decisions), and reach audiences that would be hard to recruit through traditional channels, all without privacy risk or compliance overhead.

Limitations and Risks: What Synthetic Audiences Get Wrong

This technology is a powerful assistant. It's not a final judge.

Here are the real limitations that responsible marketers need to understand before building a strategy around synthetic research.

The Sycophancy Problem

This is the biggest practical risk and the least talked about.

AI models are often optimized to be helpful and agreeable. In a research context, that tendency makes synthetic respondents unrealistically positive. They tend to like your new product. They tend to approve of your messaging. They rarely express strong negative opinions or flag genuine product flaws.

The Nielsen Norman Group documented this directly: AI users are "much more positive than real humans." A synthetic panelist asked whether they completed an online course said yes, always, without hesitation. Real humans rarely did.

NielsenIQ warns that many synthetic feedback tools generate outputs that "pass a gut check" but aren't grounded in real evidence. Confidently wrong is still wrong.

Hallucinations

Generative models can fabricate information that sounds statistically plausible but is factually incorrect. When Kantar's Cynthia Vega asked a digital twin whether it drinks alcohol in the morning, the model said yes and provided a sophisticated explanation involving work events and weddings. Plausible on the surface. Not what a fast human response would have said.

A synthetic panelist also skipped a question because it was mimicking survey fatigue behavior. That kind of guardrail failure produces unreliable data without obvious warning signs.

Bias Amplification

If the training data skews toward Western consumer behavior, the synthetic audience will reflect that bias. Models are only as good as what they're trained on. A dataset with gaps produces a panel with blind spots. Garbage in, garbage out still applies.

Unknown Unknowns

Synthetic audiences are good at interpolating within known patterns. They're weak at predicting genuinely new behavior or emergent trends that don't exist in the training data. They can't tell you about something they've never seen.

Best Practices: Getting the Hybrid Approach Right

The teams getting the most value from synthetic audiences aren't replacing traditional research. They're using synthetic panels to make traditional research faster and cheaper.

That's the hybrid approach. Use synthetic audiences first. Validate with real people second.

Here's how to apply it in practice.

Start with low-stakes use cases. Test a product name or a headline variation. Build familiarity with how the models respond and where they behave unexpectedly before using them for anything that affects a major budget decision.

Ask specific questions. Vague prompts produce vague answers. The more precise the research question, the more useful the synthetic output. Define your success metric before you start.

Watch for sycophancy actively. Ask adversarial questions. Ask the synthetic panel why they wouldn't buy, what they dislike, or what would make them switch to a competitor. If every answer is positive, your guardrails need adjustment.

Refresh your data regularly. Consumer perceptions shift. A synthetic audience trained on 18-month-old data reflects 18-month-old behavior. The model needs current inputs to stay accurate.

Validate before any major launch. US Bank ran real audience tests alongside their synthetic panels specifically to build confidence before committing to full campaign production. That validation step is not optional.

Evaluate your vendor carefully. Look for providers with transparent data sourcing, documented bias detection, and a clear validation methodology. Electric Twin publishes its LSE collaboration and accuracy benchmarks publicly. That level of transparency is what responsible adoption looks like.

The Qualtrics 2026 Market Research Trends report put it plainly: researchers who have integrated synthetic data are 11% more likely to engage in early-stage innovation and 7% more likely to conduct go-to-market research compared to those still using only traditional methods. The competitive gap is structural, not temporary.

The Bottom Line on Synthetic Audiences

Getting instant feedback on your next big idea no longer requires waiting for a focus group to be scheduled, recruited, and debriefed.

Synthetic audiences give you a fast, privacy-safe, cost-effective way to test products, campaigns, and messaging before they hit the real world. The accuracy benchmarks from Electric Twin, Kantar, US Bank, and GWI all point in the same direction: when built on robust data, these virtual panels mirror real consumer behavior closely enough to be genuinely useful.

They're not a replacement for talking to real people. They're a first filter. A way to narrow five ideas to one before you spend the budget validating it.

Teams that figure out the hybrid approach now will have a structural research advantage that compounds over time. Those still scheduling focus groups for every preliminary question will always be slower.

The same logic applies to content strategy. Understanding what your audience responds to before you publish, rather than guessing and hoping, is what separates consistent thought leaders from people who post sporadically and wonder why nothing lands.