Simulated audiences, market by market
Prefield generates responses with AI using audience profiles built from market data. The respondents are simulated. A study is an early research input for deciding what to investigate with real customers.
Each market has its own population inputs, interview language and limitations. The United States, the Philippines and Mexico are calibrated against each country’s own benchmark surveys. Saudi Arabia is available as an exploratory market: its population is constructed from official statistics, and it is not yet calibrated. Audience filters describe the requested panel; they do not establish that every subgroup has independently validated predictive accuracy.
How we build a market
Every market goes through the same six steps. The details differ by country because the data, language and benchmarks differ.
Start from official statistics
We build each market from its own country’s census, survey and population statistics. Each source’s terms of use are recorded before its data is used, and where a license requires it, source records stay on our servers.
Build the population from real records
Where a statistics office publishes person-level survey files, each profile starts from a real, de-identified record. The records are reweighted so the population matches current official totals, for example by age, sex and region. Characteristics the survey lacks, such as values, are matched from other surveys, and anything no source supports stays unstated. Each market has 100,000 profiles, and study panels are drawn from them at random.
Ask in the local language
Questions are written for each market in its own language and register, not translated from English: formal Filipino in the Philippines, Mexican Spanish in Mexico, English in the United States, and Arabic or English, chosen per study, in Saudi Arabia.
Interview each respondent individually
Every simulated respondent answers the same questionnaire from their own profile. Shares and intervals are calculated from those answers, and in hypothesis studies the support threshold is set before the panel responds.
Calibrate against a benchmark
Where a country publishes a recurring benchmark survey, the audience answers that survey’s own questions for the same population and period. We measure the gaps and correct them on those questions only; every other question is reported as the audience answered it.
Keep fit and forecast apart
Matching results that are already published is not a prediction. We separate fits to published results, tests on later results and forecasts sealed before the results exist, and describe each market by the strongest evidence it has. Mexico’s first sealed forecast, for October 2026, is registered on the track record.
The methodology paper documents the sources, checks and references behind each step, and where each market stands. Read the methodology paper
What calibration establishes
Historical benchmark comparisons measure behavior on specific reference questions and populations. Corrections apply only where a study meets the relevant market and question requirements. A fitted historical result is not a held-out test or a sealed prediction, and a benchmark result does not establish validity for every custom product question.
Read the evidence and limitations attached to your report. They identify the applicable sources, method and scope. Report source notices also describe required attribution and restrictions on underlying data.
What remains unproven
Do not interpret synthetic response shares as guaranteed demand, sales or conversion. Model knowledge, question wording, population coverage and compressed differences between groups can affect results. Compare concepts carefully and test consequential decisions with real customers and other independent evidence.
Reading a report responsibly
Keep the brief, audience, questionnaire, response basis and limitations together when sharing findings. Preserve source notices and label quotations as simulated. Public design examples are illustrative; they are not customer testimony or measured performance.
Privacy notice · Terms of service · Software and font notices