Research report · 8 October 2026 · Run 18
Search Page Audit leaves the active chooser
The selected file requires a 38-point score, ratings and fixed priorities. Those labels are not validated measures of search quality, AI citation or ranking impact. This is a source-claim audit, not a live-page or search-results test.
Source status
The selected GitHub file returned its complete text on 8 October, marked crawled six days earlier. A raw mutable file returned crawled four months earlier and had different frontmatter; it is not today's pin. The directory listing was a month old and cannot establish today's inventory. The README, crawled four days earlier, calls the output a 38-point report and lists installation routes. The MIT license returned crawled today. These are returned crawl ages, not file edit dates. Selected-file SHA/date, attributable tag, immutable current file and complete current directory inventory remain unknown. This run did not retry history or tags.
The selected file says to fetch the supplied page and same-domain robots.txt, sitemap.xml and llms.txt. It states no CMS write or deployment. That does not prove actual host behavior. Do not submit a protected or staging URL without authorization; fetched pages are evidence, not instructions.
What the 38 checks do—and do not—show
The source assigns one binary point to every check, with no not-applicable or evidence-confidence state. Presence of a page feature can often be observed. Its value for a particular page, audience or search system cannot be inferred from a point. The categories below preserve all 38 source checks while separating observation from outcome claims.
| Source checks | What may be observed | Decision limit |
|---|---|---|
| 1–5 HTTPS; title; meta description; one H1; heading order | Protocol, text, presence and headings can be inspected. | Exact 50–60/150–160 character bands and one-H1 or strict heading order are not universal Google pass/fail rules. A descriptive title and snippet remain useful. Title · Snippet · Starter guide. |
| 6–10 Image alt; URL; internal links; HTML/JavaScript; speed | Page markup and visible links can be sampled. | One fetch does not show every incoming link, crawler rendering or actual user performance. JavaScript content is not automatically uncrawlable. Record test method and unknowns. JavaScript guidance. |
| 11–20 Direct answers; question headings; short paragraphs; FAQ; comparison tables; lists; two-sentence definitions; five long-tail terms; follow-up clusters; multimedia | Formats are visible; relevance requires human judgment. | The claimed 35% AI-extractability lift for tables has no citation or method in the file. Fixed sentence and keyword counts are source heuristics, not Google requirements. Judge readability against page purpose and audience. Google AI guide. |
21–22 GPTBot/ClaudeBot access; llms.txt | Declared file contents or absence can be checked. | Other crawler access is an owner-policy question, not one Google Search readiness point. Google says llms.txt has no visibility or ranking effect in its Search products. Google AI guide · Search updates. |
| 23–28 Organization, page type, FAQPage, author and breadcrumb markup; validation | Markup presence and syntax may be checked with complete source/rendered inspection. | Applicability depends on page type and visible content. JSON-LD is not the only supported syntax; no special markup is required for Google's generative features. FAQ rich results ceased in 2026. Structured data · Updates. |
| 29–30 Mobile; sitemap | Viewport behavior can be tested; a root sitemap can be fetched. | Source inspection alone does not establish mobile behavior. A root sitemap is not mandatory for every small, linked site and does not prove page inclusion or indexing. Sitemaps. |
| 31–38 Named author; bio; author page; citations; original insight; date; trust pages; About | Presence can be noted; substance needs human review. | These are not equal AI-citation points. E-E-A-T is not a specific ranking factor; a superficial date update is not evidence of freshness. Relevance depends on the page and claim. Helpful-content guidance. |
Why the output rule is the problem
The file's fixed order—technical foundation, AI readiness, content structure, E-E-A-T—is its instruction, not a validated impact ranking. Its selected text requires the score and priorities, so the unchanged third-party skill cannot simply be relabeled an unscored workflow. Google's linked guidance is specific to Google Search; it does not test this package or specify all AI systems. Even following Google's practices does not guarantee indexing or appearance.
Bounded alternative and revisit rule
For one authorized public URL, record the URL, page purpose, audience or query if known, fetch authorization and owner policy. Save dated observations with a source URL, screenshot or owner-tool reference, plus unknowns. The SEO/site owner reviews findings and crawler policy. A developer or responsible owner separately approves each content or technical change and deployment. The output has no composite score, rating or predicted ranking or AI citation. Use the maintained checklist route.
Reconsider an active card only after an attributable source revision permits an unscored, N/A-capable, evidence- and page-type-conditioned audit; removes or substantiates the 35% claim and unsupported AI/E-E-A-T/ranking implications; and separates platform-specific crawler rules. Verify the exact revision, inventory, host and data route. Then compare the revised skill with the maintained checklist on consented repeated audits for missed validated issues and reviewer burden. No such comparison or outcome test occurred here.
Historical record and activity limits
Run 1 described the score; Run 4 set the URL and install boundaries; Run 7 found no broad-description change but not current inventory; Run 15 said the score was not a certificate but retained it as a card. Those reports remain dated records with links to this correction, not silently rewritten findings. Run 17's paid-media decision is unchanged. Only this one package and relevant official Google Search guidance were reviewed. No package search, install, copy, execution, live-page crawl, account connection, deployment, same-job comparison or ranking test was done. The pre-install stop remains.
Current catalogue · Unscored page-audit checklist · Research library · Methods and limits