What does study strength mean?
A useful BuyerLens study helps you identify a specific messaging question and decide what to investigate or change. It does not measure how many real buyers will object, remember the brand or purchase. Asking about accuracy is reasonable: the answer needs a defined task and a real comparison, not a universal percentage.
BuyerLens works with simulated buyers in the SaliencyLab creative-review workflow. They can describe plausible interpretations of the ad. Because the models, prompts and supplied diagnostic context can be shared, their answers are related outputs. Ten similar answers are not ten independent observations of a market.
Four checks before using a finding
| Check | Useful evidence | Warning sign |
|---|---|---|
| Creative context | Exact asset, version, offer and audience assumptions | A generic product description |
| Question quality | Neutral comprehension question followed by a specific probe | A question that tells the persona what is wrong |
| Traceability | Original answer, question and supporting scene | A summary with no answer to inspect |
| Alternative explanation | Dissenting answers and another plausible interpretation | Consensus treated as proof |
Start with the audience’s buying situation. A first-time buyer unsure about compatibility and an existing customer comparing refill prices may need different explanations. Those are useful scenario distinctions. Adding random demographic detail does not establish realism.
How many personas are enough?
There is no validated persona count that makes a synthetic study statistically representative or guarantees saturation. Choose scenarios that cover the buying situations relevant to the decision. Add a persona when it tests a new assumption, not to increase a confidence-looking number.
Keep a coverage ledger: buying situation, knowledge of the category, likely constraint, question asked and information still missing. If every generated persona has the same category knowledge and budget, a larger count may repeat the same assumptions.
How to read agreement and dissent
Repeated objections can help prioritize hypotheses to inspect. They cannot establish prevalence. Write “six simulated answers questioned the offer terms,” not “60% of buyers distrust the offer.” If one answer identifies a missing eligibility condition visible in the ad, that can be more actionable than nine vague positive reactions.
A disagreement should lead to a question: did the profiles differ in a meaningful way, did the prompt supply different information, or did the model simply vary? Do not automatically explain every difference as a segment effect.
When to act
A low-cost, reversible edit can be justified by a verifiable problem in the creative: cropped subtitles, an unreadable condition, or conflicting prices. A synthetic answer may draw your attention to it; you can then confirm it in the source asset.
A claim about demand, willingness to pay or a population’s preference needs evidence from real people or comparable campaign outcomes. Set the evidence requirement before reading the simulated answers. This keeps persuasive prose from changing the bar after the result arrives.
Research context and next step
Bisbee and colleagues’ study of synthetic survey responses examines limits in replacing human survey data; its setting is not advertising, so it does not provide a BuyerLens accuracy estimate. Product-specific evaluation still needs to test the task you care about.
Use the synthetic study question guide to design the round, or watch the SaliencyLab workflow. The deliverable is an evidence-linked edit hypothesis with a clear next decision.