How WebAnalyzer tests and scores a page
WebAnalyzer inspects a public page and reports evidence across eight areas. A score summarizes the checks available in that run. It does not measure search position, AI citations, conversion rate or complete accessibility compliance.
Scoring version: 2026-10-06.2. Published October 6, 2026. Maintained by Eduardo Gil Calvo.
What the report actually measures
- Observed: a fact found in the fetched HTML, response headers, robots rules or an explicitly recorded timing.
- Estimate: an inference from available markup, such as an apparent call to action or opening content structure. This is not a usability study or a measured AI ranking factor.
- Not tested: the required evidence was unavailable. Static HTML cannot establish rendered layout, real-user Core Web Vitals, Google indexing or the site's Search Console AI inclusion setting.
- Not applicable: an optional feature or a check that does not fit the inferred page purpose. These checks do not lower the score.
Static HTML is the default mode. Browser recovery is labeled separately when automated access requires it. Optional browser measurements are synthetic lab observations from one run, not real-user field data. Lab and field measurements answer different questions.
How the score is calculated
Each applicable check has a weight and a value from 0 to 1. A dimension uses round(100 × (weighted earned / applicable weight)^1.7). Informational checks are excluded. A dimension without applicable measurements displays N/A and is excluded from the overall weighted average.
| Area | Weight |
|---|---|
| SEO | 1.40 |
| GEO / AI access and content signals | 1.15 |
| Performance | 1.30 |
| Accessibility | 1.20 |
| Code quality | 0.90 |
| UI / UX | 1.05 |
| Data and tracking | 0.80 |
| Security | 1.05 |
Selected critical failures cap a dimension at 55 for one failure, 40 for two and 25 for three or more. Examples include a noindex directive in SEO or missing HTTPS in security. Optional AI score adjustments cannot bypass these caps. Confirmed blockers appear before inferred improvements; a requested checker focus affects recommendation order rather than the scoring formula.
A reproducible scoring example
If applicable checks earn 80% of their total weight, the curved score is round(100 × 0.8^1.7) = 68. If a critical SEO check fails while that ratio remains 0.8, the SEO score is capped at 55. Two critical failures cap it at 40. Adding an informational check changes none of these numbers.
Crawler permissions are evaluated for the submitted URL
The parser follows the matching group and most specific path rule, supports Allow, Disallow, wildcard paths and terminal anchors, and prefers Allow on an equal match. A specific agent group overrides the wildcard group. An unreachable robots file is reported as untested. These rules are based on RFC 9309.
- Googlebot access concerns crawling. The report cannot confirm indexing or Google AI Search eligibility from public markup alone.
- OAI-SearchBot concerns ChatGPT Search. ChatGPT-User represents user-triggered requests and does not determine Search inclusion. See OpenAI's crawler documentation.
- Google-Extended controls Gemini-related uses and has no effect on Google Search inclusion. See Google's crawler documentation.
- Training and user-triggered permissions are displayed separately from search crawler restrictions. A site's policy choice is not a general quality failure.
Content suggestions and page purpose
Short server-delivered pages do not fail merely because they contain fewer than 50 words. Calls to action are recognized in the site's ten supported languages. Article and general page checks do not require a sales funnel; absence of client-side analytics does not establish missing measurement.
Answer formatting, paragraph length, chunks and statistics are optional editorial observations with no score penalty. We do not treat a 200-word chunk limit, a statistics quota or an llms.txt file as a requirement for AI citations. Google describes normal Search eligibility and helpful content as the foundation for its AI features.
Comparisons, retries and public benchmarks
A comparison uses one successful analysis per URL. If one side fails, its completed partner is retained for retry. Database reservations enforce the shared allowance across the website, REST API and MCP; completed requests can be replayed with the same idempotency key without another analysis. Failed scans release their reservation.
The public benchmark uses complete static-mode reports from this scoring version only. Older scores remain available in their original reports and are not mixed into a new scoring cohort. The submitted sample is self-selected and does not represent the whole web.
The score is available without signup. An email unlocks the detailed report; actionable fixes remain in the administrator view. Submitter IPs and emails are excluded from public results. See our privacy policy.