SEO·AEO·GEO Reports: How to Read a Finding and Turn It Into a Fix

An SEO·AEO·GEO report scores one page out of 100 across four categories (Retrieval, Extractability, Authority and Hygiene) and lists every check behind the number. Each finding says what was checked, what was found, what to change and how strong the evidence is that it matters. Start with anything marked as capping the score, then work down the biggest wins. When you want help applying the fixes, export the JSON or copy the report as Markdown and hand it to an LLM or a developer.

Anatomy of a finding

This is a real finding shape, as it appears in the JSON export for an article with no named author:

{
  "category": "authority",
  "check": "Author attribution",
  "status": "fail",
  "message": "No author is named.",
  "recommendation": "Add a visible byline, and an `author` in your Article schema.",
  "value": false
}

Around the findings, the report records the URL it landed on, the scoring model (modelVersion, currently 4.1.0), the detected page type, the score and grade, each category’s points and the cap they were scored against, and the checks that were skipped. In the report page, each finding that needs work also has a “why this is checked” note with its evidence grade and sources, and a “what we found” panel with the measured value.

How the score is built

• Points by category: Retrieval 67, Extractability 28, Authority 26 on articles and 18 on other pages, Hygiene up to 15. Retrieval carries the most weight because nothing else matters if an engine can’t fetch and read the page.
• Blocking checks: failing Page fetch, Indexability, robots.txt AI rules, Raw HTML visibility, Snippet controls, HTTPS or Soft 404 caps the whole score at 39, however well everything else does.
• Page type: the engine classifies the page as a homepage, article, product page or general page. Bylines and article dates are scored only on articles; other pages are judged on freshness signals instead.
• Skipped checks: anything that doesn’t apply, such as alt text on a page with no images, is removed from both the points earned and the points available, so it can’t pull the score down.
• Grades: A from 90, B from 80, C from 70, D from 60, F below that.

Some checks are reported and never scored: llms.txt, schema.org markup, FAQ schema, security headers and others. They sit at zero either because the evidence says they don’t affect AI visibility (97% of llms.txt files got no requests at all in Ahrefs’ server logs) or because no test has shown how much they matter (adding JSON-LD showed no citation lift). The engine never marks them warn or fail, because that would ask you to fix something that can’t move your grade.

Exports, and who each one is for

1. The report page: a shareable link with the score, any blockers, the three biggest wins by points, and every finding with its evidence note.
2. Copy for LLM: the report as Markdown on your clipboard, with blockers first, category scores, key signals, and every fail and warn with its fix and, where there is one, the measured value. No email needed.
3. Email me the full report: HTML, Markdown or JSON, sent once you confirm a six-digit code. The JSON is the complete report object described above.
4. Save as PDF: your browser’s print dialog, with a print layout that lists the scored findings in severity order.

Agencies can put their own name on the HTML and Markdown exports; see the white-label audit for agencies.

From research lens to report line

The seven research lenses on this site map onto the report like this. Each research post explains the evidence behind its checks.

What a one-page audit can’t tell you

The best-evidenced drivers of AI citations sit off your site: how often your brand is mentioned elsewhere, review sites, forums and “best of” lists, and where you already rank. The report measures the part you control directly (whether engines can reach, read and attribute the page) and checks for a Wikidata entity as the one off-site signal it can verify. It doesn’t tell you whether you are cited today. For Google, the guide points to Search Console’s Generative AI performance report for that.

If you have a report, export the JSON and work through the fails with the matching research post. If you don’t, run a free audit; it finishes in under half a minute.