A better survey starts
with a
better draft.
Start questionnaire testing with simulated answers, then pretest with people. Keep the same question bank as you collect and compare their perspectives.
Write the question bank
Mix open questions, multiple choice, and scales. Add product context so the study is about your offering and the decision you face.
Describe hypothetical respondents
Use a label, age range, description, and context to explore different situations. Personas guide a model; they are not real participants or a representative sample.
Explore simulated answers
Read each response and compare the model configurations you selected. Use disagreements to decide which assumptions need closer attention. The local studio uses fixed demo responses.
Ask people you can reach
Open the survey, copy its link, or embed it in your product. Interviews use your authored question bank and can adjust the order within that bank.
Compare what you learned
Read simulated and human answers together. Choice and scale summaries use the same aggregation rules, with separate sample sizes. Open responses give you explanations to review.
Use the simulation to investigate.
Voxq has no published accuracy benchmark. A model’s response depends on its training, your framing, and the hypothetical person you describe. Agreement between models can reflect shared assumptions.
Your human sample matters
An embed reaches people who visit and choose to answer. Voxq does not recruit, apply quotas, or weight responses to a population. Customers can reveal problems in your product; they cannot automatically tell you what the whole market thinks.
Keep comparisons interpretable
Reports combine saved simulation runs and include saved answers from incomplete interviews. Each group has its own sample size. Adaptive interviews may alter order and framing within the authored bank. Revisit the underlying answers when you see a gap.
Know which mode you are using
Local demo simulations are deterministic fixtures and cost nothing. Configured providers produce model responses. Neither should be presented as evidence from people.
Read the results guide →