ai-messaging-reviewlisted

# AI Messaging Review ## The Law ``` AN AI CAPABILITY CLAIM WITHOUT EVIDENCE IS A LIABILITY, NOT A DIFFERENTIATOR. "10x productivity" without measurement is a promise users test on their first session and don't forgive when it fails. Every quantified claim sourced + every capability claim scoped + every "AI" label defined IS reviewed messaging. ``` ## When to Use Trigger before: - Publishing any marketing copy that makes claims about an AI product's capabilities - Launching any campaign that references AI performance, accuracy, or productivity impact - Publishing any press release, case study, or sales collateral about an AI product - Making any public quantified claim ("saves X hours", "Y% more accurate", "Z× faster") ## When NOT to Use - Internal technical documentation (accuracy matters but external credibility risk is lower) - Product UI copy describing what a feature does — tested separately with `copy-quality-gate` ## The Four Review Categories Every piece of AI marketing copy is reviewed against all four. One failure in any category is a fail. ### Category 1 — Quantified Claims Every number in AI marketing copy must have a source. **Required for any quantified claim:** - The study or measurement that produced the number - The population it was measured on (user segment, task type, timeframe) - The comparison baseline (X× faster than what? Y% more accurate than what?) - Whether it is reproducible by a new customer ``` ✗ "Save 10 hours per week" — no measur