"People are getting lazy about friendships,” said a 13-year-old named Jordan in a study on Gen Alpha. “They'd rather have a fake, easy conversation with something that can't disagree with them than deal with a real person.”

The study, conducted by Acceleration Community of Companies (ACC) and USC Annenberg combined surveys, Think Tank discussions, and synthetic focus groups sourced by AI platform Definity. Jordan, credited as an AI child, was a part of the AI audience.

Marketers are embracing synthetic focus groups, AI models trained to respond to questions as specific demographics, as an extension and core component of their research.

“These models [like Jordan] were asked the same questions as live families, and their answers aligned more than 95% of the time,” said Michael Kittilson, who runs strategic growth and AI at ACC.

While marketers cite efficiency and creative flexibility as key advantages, they’re aware of the limitations of early technology powered by existing data.

Challenging research limits

Understanding their target audience has always been costly and time consuming for marketers, which is why identifying and segmenting audiences are the top AI use cases for both agencies and brands, according to a January IAB survey.

Researchers grappled with the ethical grey area that comes with researching children, and opted to use AI audiences to fill in those gaps. Throughout ACC's study, quotes from synthetic parents and children were labeled next to quotes from real people:

• "My 13-year-old asked for a cologne set from Dior. I'm still wrapping my head around that. Middle school basketball practice and luxury fragrance,” said Andre T, an AI parent.
• “My mom said she actually enjoys shopping for us now because we’ve introduced her to brands she’d wear herself, even though they’re technically for us,” said Avery, a real 11-year-old.

While traditional focus groups can be directionally helpful and color findings with case studies, they are not a main source of truth for marketers. When blending data sources, marketers in 2025 heavily rely on channels like search data (52.1%) and survey data (39.7%), over focus groups (17.4%), per a May SI Lab report.

“Synthetic focus groups don’t hand over truths,” said Monica Chun, president of ACC. “They generate scenarios, trajectories, and probabilities that need to be read with discipline. The role of the researcher is to treat them as directional, to test them against lived voices, and to draw out meaning without overstating certainty.”

Closing the cutting room floor

Marketers are using synthetic focus groups to speed up creative development. TV advertising agency Marketing Architects has launched a creative pre-testing tool called ScriptSooth, designed to evaluate campaign concepts before production.

The tool has given the agency more creative flexibility, said CMO Elena Jasper, noting that faster testing has allowed them to take “big swings” that once felt too risky or time-consuming.

“Because we’re saving so much time in that early creative development phase, we’re able to get brands on TV a lot faster,” said Jasper. “Our timelines have gone from months to weeks for full creative shoots.”

Instead of creating a video mockup, the team can plug a script into ScriptSooth, widening the pool of testable concepts.

“Before you would put together a concept, it would go through pre-testing, maybe it wouldn’t win, and that concept would just sort of sunset,” said Jasper. “Now you can workshop good ideas that just didn’t perform and get them to a place where they’re going to perform well.”

Embracing human chaos

AI can help marketers focus more on creativity by shortening timelines, but this can force them to miss unexpected findings and nuances from real conversations, said our analyst Paola Flores-Marquez.

“AI relies on humans being rational, but they’re just not,” she said, pointing to recent trends like Skibidi Toilet and Labubus. “You can kind of see a thread, but there is an element of human absurdity and nuance around cultural trends that gets lost.”

While synthetic focus groups have been more predictive of in-market results than online surveys for Jasper, building high-performing models requires extensive historical data and time. Even with advanced resources, relying too heavily on synthetic audiences can lead to stale findings.

“Human research supplies the grounding, the lived texture, and the unexpected spark that no model can manufacture,” said Chun. “It’s AI with humans, not AI versus humans.”