We do not infer what AI sees from theme source or screenshots. We collect from a real browser: rendered JSON-LD, FAQ anchors, raw-md and visible text. The server re-validates the payload; client-provided verdicts are never trusted.
2. Four dimensions
Readability: robots access, status codes, raw-md availability.
Consistency: the same SKU has the same price, weight and return window across pages, Admin and raw-md.
Freshness: every finding carries a timestamp and collection method.
3. Measurement
For AI visibility we use a fixed prompt set, versioned by hash. Each run records engine, model version, answer hash, mentions, cited URLs and sentiment. We report rates with sample size and time window; single answers are anecdotes.
4. Reproducibility
fixed URL lists and template versions;
raw evidence retained for at least 90 days;
benchmark protocols published before results;
scripts and schemas documented so a third party can replicate.
5. Limitations
AI answers vary; correlation is not causation;
first-party evidence reflects collection time, not history;
no ranking or traffic guarantees;
platform rules and model behaviour change.
6. Corrections
Errors are corrected in place with a changelog entry. Substantive changes update dateModified. We publish corrections rather than silently editing.