Fetch metrics
Probe each page with the 10 canonical user agents and record:
- HTTP status per user agent
- Bytes and visible text received
- Main text ratio and first content offset
- Sentence survival against the rendered page
Answer metrics
Run the same buyer-intent prompts across ChatGPT, Claude, Gemini and Perplexity on a fixed schedule and record:
- Mention: the answer names your brand
- Citation: the answer links your domain
- Owned answer: a cited answer, not a mention
- Competitors named in the same answer
Label what you have not measured
A metric you did not collect should read NOT YET MEASURED, never a pass. That keeps every number defensible.

