Field notes on synthetic audience research.
Every study ships with the methodology. This is where the methodology gets explained.
A plain definition of synthetic audiences: AI-generated populations that behave like real ones, how they are built and calibrated, what they are good for, and what separates a defensible one from a generic LLM wrapper.
What a cohort is across four research traditions, what AAPOR disclosure elements require, and why cohort precision determines synthetic audience fidelity.
The individual record in a synthetic study: how it differs from a persona, cohort, or audience; grounded vs prompted approaches; where it works and where it fails.
The person-level building block of a synthetic study: how personas differ from respondents and audiences, why grounding in real data matters, and what separates a defendable persona from a stereotyped one.
AI focus groups come in two distinct formats: AI moderators running parallel interviews with real humans, and fully simulated discussions among LLM-generated personas. This article disambiguates the two, covers what each can and cannot do, and explains how both sit relative to quantitative synthetic audiences.
A plain definition of likely voters: how pollsters construct them, why the screen is contested, and how a likely-voter cohort maps onto synthetic audience research.
A crosstab (cross-tabulation) is a table that shows how survey answers break down across respondent segments. The most common artifact in market research, and the output format Replism delivers.
The empirical ceiling on survey reliability, derived from test-retest research and Park et al. (2024). A reference for evaluating synthetic audience accuracy claims.
Put the platform in front of a real decision.
Bring a decision your team is working on. A research engineer will draft the cohort, the sample, and the study with you, in one working session. The methodology comes out with the result.