The End of Focus Groups? AI Replicates Consumer Responses With 85% Accuracy

Want to know how consumers really feel about a product? Soon, a company might not need to bribe them with pizza and gift cards — just their digital twins.

Stanford and Google DeepMind researchers have created artificial intelligence (AI) replicas of consumers from two-hour conversations, opening the door to endless virtual focus groups that never complain about parking or run overtime.

The breakthrough, detailed in a new paper, could change how companies understand and predict consumer behavior. Researchers found AI agents built from brief interviews matched their human counterparts’ responses to personality tests and social surveys with 85% accuracy, pointing to potential applications from focus groups to product testing — all without the need for real people in the room.

“By creating digital replicas of target customer personas, businesses can simulate how different personality types respond to various product features, designs or marketing messages,” Rob Lubeck, the chief revenue officer at RTS Labs, an enterprise AI consulting company, told PYMNTS. “This enables rapid testing of multiple variations simultaneously, identifying which resonate most effectively with different segments. Not only does this save time and resources compared to traditional methods, but it also provides deeper insights into customer motivations, preferences and pain points — all before engaging real customers.”

How They Did It

The study researchers recruited 1,000 participants varying in age, gender, race, region, education and political ideology, offering up to $100 compensation. Each subject completed a two-hour interview about topics ranging from childhood memories to immigration policy.

The team created AI replicas and tested accuracy by having participants complete personality tests, social surveys and logic games twice over two weeks. The AI agents then completed identical assessments. Results showed 85% similarity between human and AI responses, though the replicas performed less effectively on behavioral tests like the “dictator game,” which measures fairness values.

Lubeck said that support agents could practice handling various customer interactions using AI personality replicas. These digital customers would let agents safely experiment with different approaches — from defusing tense situations to suggesting new products — across diverse personality types and scenarios.

“This approach enables companies to anticipate and address customer needs with greater empathy and precision, ultimately improving satisfaction and loyalty,” he said. “Over time, AI replicas could also help refine automated support systems, ensuring that even chatbot interactions feel more human and tailored.”

Virtual Humans

AI personality simulations join a growing field of digital modeling. Digital twins, which are virtual models of physical systems or objects, are increasingly used in healthcare, manufacturing and construction industries. These simulations replicate real-world conditions, allowing for testing and analysis without physical prototypes. The growing interest stems from their potential to improve design, predict outcomes and streamline complex operational processes.

Companies across various industries increasingly utilize digital human simulations to enhance product testing and development. For instance, BMW’s Regensburg plant employs “3D human simulation” to digitally plan assembly processes, allowing for ergonomic analyses and operational efficiency assessments before physical implementation. Similarly, Dassault Systèmes introduced the Living Heart Project, which created a virtual twin of the human heart to simulate cardiac function.

Saul Marquez, founder and CEO of Outcomes Rocket, a digital healthcare marketing agency, told PYMNTS that the role of healthcare customer service is particularly compelling for AI personality replicas.

“One of the most common pain points I’ve seen come up over and over again for the healthcare executives I talk to on the Outcomes Rocket Podcast is the challenge of providing consistently great customer service without being compliant and without being able to answer complex medical questions,” he said. “Using AI-powered training simulations, teams can learn the intricacies of communication in the healthcare context, such as how to deal with sensitive patient data and how to explain complex medical devices to different stakeholder groups.”