AI can explain evidence. It cannot invent it.

Permitted uses

AI may help review wording, translate with human checking, code open responses, detect anomalies, summarize approved aggregate findings, retrieve methodology records, and evaluate forecast performance.

Prohibited uses

AI-generated people cannot be counted as survey respondents. Demosight does not generate synthetic answers and present them as measured opinion, infer sensitive traits for ad targeting, or allow a model to publish a political claim without human review.

Predictive marketing boundary

Predictive products use approved aggregate survey indices and separately governed aggregate client outcomes. Panel identities, contact details, and individual responses do not become campaign audiences, lead lists, customer scores, or individual ad-targeting features.

Accountability

Material models are versioned, documented, backtested, monitored for drift, and compared with actual outcomes. Public forecasts state their data cutoff, horizon, method, uncertainty, and evaluation history. Each material AI use records its task, data class, human owner, validation evidence, failure modes, and disclosure treatment in line with AAPOR’s 2026 responsible-AI direction.