Persona Content Quality System

A brief template system that prevents AI from publishing unverified health and wellness claims. Every claim in a content brief requires an evidence tier and a source before it reaches the language model. If you cannot assign an evidence tier, the claim does not go in the brief.

The evidence tier system has six levels. T1 is strong RCT evidence or meta-analysis: state as fact. T2 is good replicated RCTs: 'Studies have found...' T3 is observational: 'Research suggests...' T4 is animal or in vitro studies: qualifier required. T5 is traditional use only: qualifier required. Claims without a tier are blocked at the brief stage.

Also includes a Claude Code post-creation verification prompt that checks each claim in the finished article against its brief-stage evidence tier and flags mismatches.

Validated against 20 live articles on a wellness site: caught a Van Dongen sleep deprivation misquote (stated 24-hour cognitive equivalence, correct is 48 hours), a CBT-I remission rate inflation (stated 75-80%, actual around 60-65%), and three supplement efficacy claims stated without required qualifiers.

What is included: the full brief template with tier and source fields, the six-tier classification guide with required wording per tier, the Claude Code post-creation verification prompt, and a list of the five most common LLM health content errors.

Best for anyone running AI-generated health, wellness, or nutrition content who needs to maintain factual accuracy at scale.