A synthetic consumer panel is a small set of AI-generated personas built to represent a target audience, so you can ask them questions before real customers exist or before you have time to recruit real ones. Each persona has a name, a background, values, pain points, and a distinct voice, and, if it's built well, at least one persona who actively pushes back on whatever you're proposing, not just people who'd say yes to anything.
The mechanics are simpler than the term suggests: a brief goes in ("solo accountants at small Indian firms who still track invoices in Excel"), a language model generates a handful of personas against explicit rules meant to fight its own default bias toward young, urban, agreeable characters, and then each persona answers your questions independently, in character. What comes back is a set of individual responses plus a synthesis of where they agree and where they don't.
Where this is genuinely useful: catching the obvious version of a problem before it costs you real time. Would this pricing line read as reasonable or as a scam? Does this onboarding paragraph explain itself, or does it assume context a stranger wouldn't have? Is there an objection so obvious that five different generated people all raise it independently? A synthetic panel answers these in minutes instead of the days or weeks a real recruiting-and-interviewing cycle takes.
Where it stops being useful, or starts being actively misleading, is as a stand-in for knowing what your actual customers think. A panel's answers come from a language model's training data and the constraints you gave it, not from your market. It cannot tell you whether real adoption will happen, what your actual churn will look like, or whether a specific real customer will pay a specific real price.
“Anyone selling a synthetic panel as a replacement for talking to real people is selling you something the underlying technology can't deliver.
The honest way to think about the boundary:
“A synthetic panel is a filter, not an oracle.
It's good at surfacing which three questions are worth taking to real customers first, and bad at telling you the answer to those questions with any real confidence. Use it to sharpen what you ask real people, not to skip asking them.
If you're deciding whether to bother: the lower the cost of being wrong and the earlier the stage, the more sense a synthetic panel makes. A five-minute check before you write a real survey is close to free. The higher the stakes and the more a real decision hinges on the answer, the more that five-minute check needs to be followed by the slower, real version, not replaced by it.