Establish the technical baseline
Record crawler directives, status codes, canonicals, hreflang, sitemap coverage, structured data and visible source pages before testing model responses.
Design the prompt sample
Group prompts by brand, category, problem and comparison intent. Fix the language, geography and evaluation date so later runs can be compared honestly.
Publish the evidence trail
Store prompt, response, citations, screenshots or exports, and the audit version. Separate confirmed facts from interpretation and list any inaccessible systems or unknowns.
Practical checklist
- Version the audit scope
- Preserve technical and response evidence
- Disclose sampling limits
- Compare like-for-like reruns