- 0 of 9GEO audit repositories publish a test of their score against AI answersGitHub, searched 10 Nov 2026
- 27,863stars on the largest of the nineGitHub API, 10 Nov 2026
- 94 / 100what one access audit gives example.com, a placeholder pageOur pipeline test, 10 Nov 2026
- 20brands in the test we registeredSV-001, registered 10 Nov 2026
Short answerNone of the nine tools we checked has shown that its score does. On 10 November 2026 we searched nine open-source GEO audit repositories and found no published test of an audit score against what AI engines answer, so we registered one before any data exists.
Nine repositories, no test of the score
FindingNone of the nine open-source GEO audit repositories we searched on 10 November 2026 publishes a test of its score against how often AI engines name or cite the audited site.
| Repository | GitHub stars | Files mentioning a score “/100” | Test of the score against AI answers |
|---|---|---|---|
| AgriciDaniel/claude-seo | 27,863 | 35 | None found |
| zubair-trabzada/geo-seo-claude | 11,606 | 23 | None found |
| Ryze-AI-Adgent/open-seo-mcp-skills | 9,085 | 0 | None found |
| onvoyage-ai/gtm-engineer-skills | 1,359 | 5 | None found |
| Fangx-AI/geo-doctor | 606 | 4 | None found |
| seranking/seo-skills | 203 | 14 | None found |
| yaojingang/GEOHub | 169 | 3 | None found |
| TheSmokeDev/geo-skills | 32 | 23 | None found |
| sunshine-lang/geo-ai-skillkit | 1 | 0 | None found |
One figure, no source
FindingThe client report template in zubair-trabzada/geo-seo-claude says “A 10-point GEO score improvement typically correlates with a 15-25% increase in AI citation frequency” and gives no source for that figure (checked 10 November 2026).
- Where it is. The line is in
skills/geo-report/SKILL.md, in the instructions for estimating a dollar value for a client. The same line appears in TheSmokeDev/geo-skills. - One tool says the opposite about itself. Fangx-AI/geo-doctor’s own output says “Technical audit only, not an AI ranking”.
- An access score is not a visibility score. In our pipeline test, that audit gave example.com, a placeholder page, 94 out of 100. The audit checks whether crawlers can read a page, and they can read that one, so the score is right by its own design.
- Three repositories repeat one passage-length figure. “134-167 words” as the best length for a cited passage appears in three of the nine. We have not checked its source.
The test we registered
Our readingA score its makers have not tested is a checklist with a number on it. The checklist may be useful. The number is untested.
SV-001 compares two fixed scoring scripts with how often ChatGPT, Gemini and Google AI Mode name 20 project management brands in the collection of 13 and 17 November 2026 (registered as the Census100™, now part of CensusRank™). The scripts, the pages, the test and the rule for reading it were fixed on 10 November 2026, before any score or answer existed. With 20 brands the test can find a strong relationship and will usually miss a moderate one, and the entry says so. Our prediction, on the record: no detectable relationship.
What to do
Been handed a GEO score? Ask what it was tested against. Then check the things that block a page outright, which no score should average away:
- Is there a
noindextag or header on the page? - Does robots.txt block OAI-SearchBot, Googlebot or Bingbot?
- With JavaScript off, how much of the text is left?
Our free GEO check does these in one command and gives findings, not a score.
Selling audits? Report the findings and what was fixed. If you quote a score, say plainly that it has not been tested against AI answers.
Next: the SV-001 result, published with the CensusRank™ project management list whichever way it comes out. Get it by email.
None of nine open-source GEO audit repositories searched on 10 November 2026 publishes a test of its score against AI answers.
Source: AI Visibility Census, “Do GEO audit scores predict AI visibility?”, https://aivisibilitycensus.com/do-geo-audit-scores-predict-ai-visibility/ (CC BY 4.0)How we did it, and limits
On 10 November 2026 we cloned the nine repositories in the table at their latest commit and read their star counts from the GitHub API. We searched every Markdown file for a score or rubric described as validated, correlated, predictive or tested against an outcome, and read each match. Matches about AI answers cited outside studies of single signals, or asserted a link without evidence. None reported a test of the repository’s own score. Two make an unsourced link between their score and AI citations: the line quoted above, and seranking/seo-skills, which says of its CITE score that “the correlation is strong”. “Files mentioning a score” counts Markdown and Python files containing “/100”. The example.com figure comes from running Fangx-AI/geo-doctor’s audit at commit 3c15897 in a GitHub-hosted job; the raw output is in our public repository. AI was used to run the searches and draft this page. On 11 November 2026 a separate AI check, run without sight of this draft, rechecked every figure, quote and link against the repositories themselves. One figure changed: claude-seo had 31 files containing “/100” earlier on 10 November and 35 at its last commit that day (734998e), so the table says 35. Star counts were confirmed through the GitHub API on 11 November, a little higher than the day before. Everything else matched.
Limits: nine repositories chosen from GitHub searches for GEO skills, not every audit tool. A text search can miss a test described in other words, or one published outside the repository, such as on a blog. Paid tools were not looked at. Star counts change daily. “None found” means we did not find one, and the authors may know of evidence we missed: if so, email us and we will add it.
Our interest: we publish our own free GEO skills, including a page check that gives no score. A result against scores would suit that design, which is why SV-001 was fixed before any data existed.
The table is free to reuse under CC BY 4.0; please credit the AI Visibility Census with a link to this page. Spotted an error? Email hello@aivisibilitycensus.com and we will correct it.
Published by the AI Visibility Census.
Newsletter
Get the next Census by email
Each new edition, list and experiment result, sent when it publishes. About one email a week. One click to leave.