geo-score checks whether ChatGPT can cite your site, in 20 seconds and with no install

The question of whether ChatGPT will quote your homepage is different from where you rank on Google, and until this week there was no free way to answer it. geo-score, an open-source tool published on GitHub by the Shenzhen consultancy Jianrun Tech, reads any homepage URL and prints a 0-to-100 readiness score against a published rubric, in about twenty seconds and without installing anything.

The tool is Python 3.8+, uses only the standard library, and runs from a single curl line. It carries an MIT licence and, as of 25 September 2026, 401 stars on GitHub. The rubric is versioned (v1.1), and every check reports the tier a page reached, the evidence, and the improvement the next tier asks for.

What it checks

21 checks add up to 100 points across five pillars, plus four bonus checks worth up to six points outside the denominator:

• Reachable (15). Does robots.txt allow the ten named AI crawlers, does the page respond to each one, and is the main content in the server-rendered HTML?
• Understandable (22). Is there Organization and WebSite JSON-LD, an llms.txt file, a fresh sitemap, breadcrumbs, and page-type schema?
• Content Citability (35). Do paragraphs stand on their own as answers, do headings match how a person asks the question, are statistics attributed, and is there a real named author?
• Brand Credibility (18). Is the company an entity in the Google Knowledge Graph, do sameAs links resolve, and is there video presence?
• Answer Fit (10). Is the content shaped so an engine can lift a passage into an answer?

Three of the checks are gates: score zero on crawler access, live reachability, or server-rendered content, and the total is capped at 40, because nothing else matters until a retrieval bot can read the page.

The 314-site benchmark

Jianrun Tech ran the tool against 314 named sites and published every result. The median score is 56 and the range runs from 12 to 98. Pulumi, MiniMaxi and Luma Labs top the table at 98, 95 and 93. Amazon lands at 12, because it disallows AI crawlers in robots.txt, which the report calls an editorial choice.

Eighty of the 314 sites failed a gate check. Nineteen block AI crawlers by name, and 45 serve a body that only exists after JavaScript runs, so a crawler gets an empty shell where a browser sees the page. Sites built for the Chinese market scored 19 points lower than the rest of the field, a gap that held across five separate samples. Running the whole benchmark twice, 96 percent of sites landed within five points of the first run and 45 percent scored identically, so Jianrun Tech says a single-site score should be read with a five-point margin.

What it does not do

The rubric measures readiness and stops there. The maintainer says on the project page that remediation templates, per-engine tactics and content rewriting all belong to a separate, non-open project. Brand credibility and the named-author check both need human judgement, and pages rendered mainly on the client will score low, because the pre-hydration HTML is close to what a crawler sees.

For a team publishing content they want an answer engine to quote, the score names the mechanical work that is missing, before spending on the ambiguous work that follows.

Source

• Jianrun Tech, geo-score on GitHub, rubric v1.1.
• Jianrun Tech, geo-score benchmark of 314 sites.
• Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024, cited by the rubric as one of the two findings that shaped the weights.