A simulated customer can produce an answer without anyone recruiting an interviewee. The difficult part is knowing which answers deserve to influence a product, price or advertising decision.
This is already a question for India’s research industry. The Market Research Society of India included synthetic respondents in its 2026 seminar themes. NoahAIC markets simulated consumer populations for India and the United States. Those are signs of an emerging research offer, not proof that its outputs predict purchases.
The Breakdown
Synthetic research can draw on general model training, customer-segment profiles or an individual’s previous answers. These approaches are not interchangeable.
A September working paper by Oded Netzer and Rajan Sambandam tested GPT-5.4-mini against withheld responses from a 2025 US survey of 3,063 people. Adding relevant attitudes improved individual predictions, yet demographics alone produced smaller errors in survey averages. Matching toplines did not mean identifying differences between respondents.
The test concerned survey answers, not purchases or Indian consumers. It does not establish that simulated respondents can replace human research.
Why This Matters
Packaging and advertising introduce another challenge: recognising a visual feature is not the same as responding to it as a person would.
A separate September preprint compared GPT-4o-mini and GPT-5.4-mini, using text and structured prompts, with six published visual-marketing experiments. Without supplied benchmark evidence, no configuration reproduced more than two of the six reported human effects, despite passing checks on recognising the visual differences.
Supplying the original studies’ statistics improved alignment: the newer model reproduced effects in six of seven outcome tests. But that demonstrates responsiveness to supplied evidence, not independent prediction. The benchmarks involved US adult and undergraduate samples, not Indian shoppers. Published findings are comparison points, not infallible ground truth; the results concern these models and tests, not every synthetic-research system.
For Indian brands, neither paper removes the need to test a method on their own customers.
The Big Picture
Kellogg consumer anthropologist Gina Fong suggests using synthetic personas to prepare for conversations with real customers, rather than treating the preparation as a replacement for those conversations. Observation can reveal how people actually use a product and what they struggle to explain.
A research brief should identify generated responses and specify validation. Matching averages cannot validate customer differences; agreement with stated preferences does not establish sales predictions.
The Crunch
Synthetic customers can help a brand ask better questions. Whether they can answer those questions reliably remains a separate claim to prove.