GEO means generative engine optimization: getting cited by AI answer engines such as ChatGPT, Perplexity, Google AI Overviews, Gemini and Copilot. It has nothing to do with geography or maps. Answer engines do not rank pages; they retrieve passages from sources they can reach, parse and trust, and cite them. geo-score measures whether a site is ready for that.

Yes. It is MIT-licensed and has no paid edition. The score needs no key. Levels 2 and 3 call AI providers with your own keys: you pay the providers directly, and every run keeps a ledger of requests, tokens and cost.

Why did my score move a few points between two runs?

Each run samples up to 8 pages, and the sample can change from run to run. In a same-day retest of 317 sites, 99% stayed within ±5 (source: REPRODUCIBILITY.md ). To see whether the site changed, save a run with --json-out before.json and re-score the same pages with --baseline before.json .

Why does a check show a dash instead of a score?

It was not measured, so it left the denominator instead of scoring zero. That happens when an off-site lookup does not answer, a page is in a language the heuristics do not read, or the check needs off-site search or human judgement. Not measured is never zero.

Why is my score capped at 40?

One of the three gates scored zero: robots.txt disallows AI retrieval crawlers, most retrieval user-agents are refused, or the body text exists only after JavaScript runs. Until a crawler can get the content, nothing else on the page reaches an answer engine, so the normalised score cannot exceed 40. The report names the gate and its evidence first.

Why does the online score differ from my local run?

The online scorer reads up to 5 pages instead of 8, waits at most 10 seconds per request and 30 seconds per run, and fetches from a cloud data centre. Some sites answer cloud addresses differently, and a check the run had no time for is not measured. For a score you want to keep, run the CLI with its defaults; the result page gives you the exact command.

Does a higher score mean AI will cite my site more?

That has not been shown. The score measures whether engines can cite a site. In the preregistered test's first wave, 1,280 answers from one engine across 83 sites detected no association between score and citations (source: RESULTS-wave0.md ). Whether engines do cite you is measured separately, by levels 2 and 3, and never added to the score. geo-score never predicts traffic, rankings or revenue.

Does geo-score fix my site?

No. It measures and reports what the next tier of each check needs. Fix templates, content rewriting and per-engine tactics are left out of the repository on purpose, apart from two public-knowledge examples: a robots.txt that lets retrieval crawlers in and an Organization JSON-LD object. If you want the fixing done, Jianrun Tech's GEO service, CiteSpot , does that work.

What does geo-score send, and to whom?

Nothing to its maintainers: there is no telemetry. A level 1 run fetches the site you name (following its redirects), the URLs its markup and robots.txt point to, such as sameAs profiles, the logo and the sitemap, and Wikidata and Wikipedia search. Levels 2 and 3 call only the AI providers whose keys you set.

What does the online scorer keep?

It does not store the URL you submit or the result in any database, and its code writes no logs. The result for a URL is cached at the CDN for up to an hour so that repeated requests are not recomputed. Details are on the privacy page .

Which AI crawlers does geo-score check?

The 10 retrieval user-agents: OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Googlebot, Bingbot, Applebot and Amazonbot, each with the user-agent string its vendor documents. Training crawlers and opt-out tokens such as GPTBot, ClaudeBot, CCBot and Google-Extended are reported but never scored ( reference/ai-crawlers.md ).

How does a site get onto the public leaderboard?

Through the "Add a site" route in the contributing guide . A badge or a link back never moves a score and never decides whether or where a site appears on the leaderboard .

How do I challenge a weight in the rubric?

Open a rubric proposal with your evidence. Every weight already declares what it rests on in the evidence table , including where that support is thin.

geo-score FAQ: scores, gates, the online scorer and privacy

Short answers to the questions people ask most about geo-score, with links to the record behind each one. For a reading that looks wrong, the troubleshooting guide explains every error kind and every way a score can move.

What is GEO?: GEO means generative engine optimization: getting cited by AI answer engines such as ChatGPT, Perplexity, Google AI Overviews, Gemini and Copilot. It has nothing to do with geography or maps. Answer engines do not rank pages; they retrieve passages from sources they can reach, parse and trust, and cite them. geo-score measures whether a site is ready for that.
Is geo-score free?: Yes. It is MIT-licensed and has no paid edition. The score needs no key. Levels 2 and 3 call AI providers with your own keys: you pay the providers directly, and every run keeps a ledger of requests, tokens and cost.
Why did my score move a few points between two runs?: Each run samples up to 8 pages, and the sample can change from run to run. In a same-day retest of 317 sites, 99% stayed within ±5 (source: REPRODUCIBILITY.md). To see whether the site changed, save a run with --json-out before.json and re-score the same pages with --baseline before.json.
Why does a check show a dash instead of a score?: It was not measured, so it left the denominator instead of scoring zero. That happens when an off-site lookup does not answer, a page is in a language the heuristics do not read, or the check needs off-site search or human judgement. Not measured is never zero.
Why is my score capped at 40?: One of the three gates scored zero: robots.txt disallows AI retrieval crawlers, most retrieval user-agents are refused, or the body text exists only after JavaScript runs. Until a crawler can get the content, nothing else on the page reaches an answer engine, so the normalised score cannot exceed 40. The report names the gate and its evidence first.
Why does the online score differ from my local run?: The online scorer reads up to 5 pages instead of 8, waits at most 10 seconds per request and 30 seconds per run, and fetches from a cloud data centre. Some sites answer cloud addresses differently, and a check the run had no time for is not measured. For a score you want to keep, run the CLI with its defaults; the result page gives you the exact command.
Does a higher score mean AI will cite my site more?: That has not been shown. The score measures whether engines can cite a site. In the preregistered test's first wave, 1,280 answers from one engine across 83 sites detected no association between score and citations (source: RESULTS-wave0.md). Whether engines do cite you is measured separately, by levels 2 and 3, and never added to the score. geo-score never predicts traffic, rankings or revenue.
Does geo-score fix my site?: No. It measures and reports what the next tier of each check needs. Fix templates, content rewriting and per-engine tactics are left out of the repository on purpose, apart from two public-knowledge examples: a robots.txt that lets retrieval crawlers in and an Organization JSON-LD object. If you want the fixing done, Jianrun Tech's GEO service, CiteSpot, does that work.
What does geo-score send, and to whom?: Nothing to its maintainers: there is no telemetry. A level 1 run fetches the site you name (following its redirects), the URLs its markup and robots.txt point to, such as sameAs profiles, the logo and the sitemap, and Wikidata and Wikipedia search. Levels 2 and 3 call only the AI providers whose keys you set.
What does the online scorer keep?: It does not store the URL you submit or the result in any database, and its code writes no logs. The result for a URL is cached at the CDN for up to an hour so that repeated requests are not recomputed. Details are on the privacy page.
Which AI crawlers does geo-score check?: The 10 retrieval user-agents: OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Googlebot, Bingbot, Applebot and Amazonbot, each with the user-agent string its vendor documents. Training crawlers and opt-out tokens such as GPTBot, ClaudeBot, CCBot and Google-Extended are reported but never scored (reference/ai-crawlers.md).
How does a site get onto the public leaderboard?: Through the "Add a site" route in the contributing guide. A badge or a link back never moves a score and never decides whether or where a site appears on the leaderboard.
How do I challenge a weight in the rubric?: Open a rubric proposal with your evidence. Every weight already declares what it rests on in the evidence table, including where that support is thin.

Frequently asked questions