HOW WE MEASURE

Every conclusion needs the right evidence.

Our method separates what a website exposes, what search engines report and what customers actually do. A technical readiness check is a useful starting point; it is not a ranking forecast or a measured AI citation rate.

1. Establish the website baseline

Record the reviewed URLs, collection date, response codes, metadata, content, internal links and crawl directives. Important issues are checked against the actual response. A missing sitemap that returns a homepage with HTTP 200 needs a routing fix, not an assumption that the sitemap is valid.

2. Review first-party search evidence

With permission, use Search Console to review indexing and search performance. Keep a stored indexing result separate from a live URL test: the first reports Google’s recorded state; the second tests whether a page is accessible now. An eligible live test does not establish that a page is indexed.

Record date range, search type, country and device filters when comparing impressions, clicks, click-through rate and average position. These are aggregated search measurements, not a complete record of every query or customer journey.

3. Check performance without mixing lab and field data

Lighthouse is a controlled lab test. Chrome UX Report field data describes eligible real-user samples. A fast lab result is helpful, but it cannot substitute for missing field data. We report the device profile, test date and available metrics; we do not infer real-user responsiveness from a fetch duration.

4. Connect visits to meaningful actions

GA4 can report visits and successful enquiry events where the visitor permits analytics and collection works. The website records a generate_lead event only after the enquiry API confirms a saved request. Failed requests and simple button clicks are not leads.

Consent choices, blockers and missing referrers limit the observed sample. The operational enquiry database remains the record of submitted requests. Search Console and GA4 also use different definitions and can use different reporting timezones.

5. Observe actual AI answers

Agree a fixed question panel and record exact prompts, platform, date, language, market context, search mode, answers and citations. Preserve evidence and compare like-for-like runs. Report mentions, linked citations and referral visits separately.

Our own initial prompt panel is a research plan, not completed evidence. We will not present its unrun questions as a visibility score. Read the AI-search service scope for details.

6. Review changes and uncertainty

Use the release date and backlog to track what changed. Review unexpected outcomes and changes in demand, competitors or search systems before claiming causation. Automated checks support this process; important business claims and publishing decisions need review.

Primary references

Method reviewed 14 November 2026. See the EntAI pilot baseline for a first-party example.