Synthetic personas vs customer research: when to use each
Synthetic personas and customer research answer different questions. Simulation helps teams generate and challenge hypotheses quickly; research with real people provides evidence about their language, experience, and behavior.
Choose methods by the claim you need to support. “This could confuse someone” and “this increases conversion” require different evidence.
- Prepare sharper interview questions
- Explore plausible reactions before recruiting
- Challenge internal consensus cheaply
- Identify variants worth testing
- Represent customers without validation
- Discover prevalence in a population
- Observe real product behavior
- Resolve high-stakes decisions alone
Worked example: choose evidence for a pricing-page decision
A team is deciding whether to lead with a monthly price or with the time a workflow may save.
Use simulation to list possible misunderstandings and objections under fixed audience assumptions. Then ask real customers how they compare the offer with their current workaround.
Simulation can suggest that the time-saving claim needs a baseline and that the monthly price needs scope. Customer conversations can reveal the language and constraints people actually use.
The two methods contribute different evidence. Neither the simulated objection nor one interview estimates conversion impact.
Use the findings to create a controlled page comparison and let observed behavior answer the performance question.
Compare the available plans
When you are ready to apply this research workflow, review the plan details and included capabilities.
Match the method to the evidence
Use synthetic personas when speed and breadth of hypothesis generation matter. Use interviews when you need customer language and causal explanation. Use surveys when prevalence matters and sampling is defensible. Use behavioral experiments when the claim concerns action.
- Synthetic personas: possible reactions and blind spots
- Interviews: lived context, language, and reasons
- Surveys: measured distribution of reported responses
- Experiments: causal impact on observed behavior
A stronger sequence than either/or
Begin with internal evidence and explicit assumptions. Use simulation to stress-test the message and improve the research instrument. Then speak with customers or observe behavior. Feed what you learn back into the audience definition and the next iteration.
This sequence makes simulation a preparation and synthesis layer, not a replacement for contact with the market.
Let decision risk set the burden of proof
A reversible headline draft can tolerate directional evidence. A positioning change, price decision, or major media commitment deserves stronger evidence. The higher the cost of being wrong, the closer the validation should be to real customers and real behavior.
Keep provenance visible
Document which findings came from simulation and which came from people. Do not merge them into an indistinguishable “customer insight” label. Clear provenance prevents confidence from outrunning the evidence.
Pre-launch checklist
- 01Name the claim behind the decision
- 02Choose the method that can actually observe that claim
- 03Use simulation findings as hypotheses
- 04Keep synthetic and human evidence labeled
- 05Increase validation strength with decision risk
Sources and further reading
- Hu & Collier (2024), arXiv
Quantifies how persona variables affect LLM simulation performance across subjective NLP datasets.
- Li et al. (2025), arXiv
Analyzes when synthetic-user fidelity is and is not sufficient.
- Market Research Society, Delphi report
Offers industry guidance for responsible synthetic-data practice.
Bring one message. Leave with clearer next moves.
Explore a reviewed example and see how DoesItClick turns plausible reactions into a focused revision list.