REVIEW METHOD
How we review and record evidence
Review method v1.0 · 7 September 2026. Working principles for assessing readiness separately from measured results.
Document updated: 2026.09.07
1. Define the scope
We review the accuracy of public web information, document structure and readiness for search access. We do not rate clinical quality or rank treatments. We distinguish official technical guidance from practical editorial and operational judgments.
2. Fix the comparison conditions
Before reviewing, record the criteria version, URL sample, language, access scope and review environment. Use the same URL sample and conditions for before-and-after comparisons, and identify any conditions that changed.
3. Attach evidence to each finding
Record the source URL, review date, necessary excerpts, observations and finding together. Exclude information identifying individuals or clients from public examples. Distinguish facts found in a source from the editor’s interpretation.
4. Use four finding categories
- Verified: evidence was examined and the criterion was met.
- Needs improvement: an error or omission was found; record where it needs correction.
- Needs verification: evidence was insufficient or access was restricted. Do not count this as a failure.
- Not applicable: record why the criterion does not apply.
5. Record relationships and re-reviews
Disclose relevant commissioning or material-supply relationships in the record. On re-review, retain the previous finding and reasons for any change. Record the evidence and unresolved issues when judgments differ. Mark a two-person cross-check complete only when it has actually taken place.
6. Measure search and AI results separately
When measuring AI answers, record the exact prompt, service, disclosed model name and settings, date and time, language, location and login conditions, repeat count, original response and citation URLs. Mark unavailable model information as undisclosed. Assess changes in search activity and enquiries alongside advertising and seasonality; a before-and-after difference alone does not establish that the site changes caused it.
Set the question list and counting method before measurement, separating questions that name an institution from those that do not. Report successful responses, errors or missing responses, and how many successful responses mention or cite the institution. Do not calculate rates from favourable responses alone.