Using AI-generated personas to pressure-test creative work, surveys, or interfaces before real humans see them. For ad agencies, synthetic testing is a faster, cheaper first pass of audience reaction. Not a replacement for real research, but a way to catch obvious problems before they reach a focus group.
Also known as synthetic research, synthetic user testing, AI-based testing
Synthetic testing uses AI-generated personas (detailed character profiles representing target audience segments) to simulate reactions to creative work, product concepts, ad copy, or research questions. Instead of recruiting twenty real people and waiting two weeks, an agency can spin up twenty AI personas representing different demographic, psychographic, and contextual profiles and get reactions in an hour.
The output is not real human feedback. It is a probabilistic best guess about how someone with that profile might respond, based on the model’s training data. Used well, it surfaces obvious creative problems and accelerates iteration. Used poorly, it gives a false sense of validation and replaces actual research that should have happened.
Real research is expensive and slow. Most creative decisions happen without it, which means most creative decisions are made on instinct and senior judgment. Synthetic testing changes the cost-benefit math of getting outside reaction. Three places this matters.
Internal pressure-testing before the client meeting. Generating ten personas and walking three creative directions past them surfaces flat reactions, confusing language, or misread tones before the client sees the deck. The team arrives at the meeting having already heard the obvious feedback.
Bias auditing. Personas representing audiences outside the agency’s lived experience catch assumptions that an all-internal review would miss. Done with care, this is a useful counterweight to monoculture in agency teams.
Real risk of hollow validation. AI personas reflect training-data stereotypes. They are not actual audience members. Hallucination applies to opinions too. Treating synthetic feedback as conclusive is a common mistake that produces work that “tested well” with simulated people and tanked with real ones. The discipline is using synthetic testing to ask better questions of real research, not to replace it.
A strategist drafts six personas representing different segments of the client’s target audience, each with detailed psychographics, recent context, and one specific frustration. She presents three campaign directions to each persona via an LLM and reads back the reactions. Two of the directions get consistent flat responses across personas; one sparks specific objections from a segment the team hadn’t considered. That third direction goes back for refinement before the work meets a real research panel.
Synthetic testing is a hypothesis machine, not a verdict machine. The team’s job is to know the difference.
The synthetic testing and feedback module of the workshop shows you how to pressure-test creative work using AI personas before it reaches the client, and where to stop trusting them.