GEO Ready Checker Methodology
Scoring version 2026.11. Every check is a measurement of the page and the files around it, never a guess by an AI model. When the platforms change what they document, we update the checks and the version number. Run the checker.
How the Score Works
The 56 checks sit in nine groups, and each group has a weight. Within a group, a pass counts as 1, a warning as 0.5 and a fail as 0, multiplied by the check's own weight. Information checks and checks that do not apply to the page are left out. A group with nothing that applies drops out, and the other groups are scaled back up to 100.
| Group | Weight | Why this weight |
|---|---|---|
| A. AI crawler access (robots.txt) | 25% | Documented by every platform |
| B. Live access, firewall and CDN | 15% | Documented, and the most common silent failure |
| C. llms.txt and markdown for agents | 5% | Speculative for search, useful for agents |
| D. Content visible without JavaScript | 15% | Most AI fetchers do not run JavaScript |
| E. Indexability and AI-use directives | 15% | Google AI features need indexable, snippet-eligible pages |
| F. Structured data and entities | 8% | Hygiene: the effect on citations is not proven |
| G. Content extractability | 8% | Titles, headings and alt text are scored. Editorial patterns are shown but not scored, as no format guarantees a citation |
| H. Trust and entity signals | 5% | Hygiene |
| I. Performance basics | 4% | Hygiene |
Some checks are shown but never scored: where sections open with an answer, question headings and FAQs, lists and tables, figures and sources, and how recently the page was updated. They are useful prompts for an editor, but there is no proven quota for any of them, so a score built on them would reward formatting rather than quality. This follows our SEO and GEO Website Build Standard, which the checker is built to agree with.
Bands: 90 to 100 is GEO ready, 75 to 89 nearly there, 50 to 74 needs work, and under 50 invisible to some AI engines.
Score Caps
Some problems make everything else irrelevant, so they cap the score however well the rest of the page does. The lowest cap that applies wins.
- 30The page is set to noindex (meta robots or the X-Robots-Tag header). Google AI features and most AI search engines will not show it.
- 35robots.txt cannot be read (a 5xx error, a timeout or a refusal). Crawlers treat that as "do not crawl anything".
- 40robots.txt blocks one of the main AI search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot or Bingbot) from the page.
- 60Fewer than 50 words of the main content are in the HTML before JavaScript runs. Most AI fetchers do not run JavaScript.
- 70Two or more AI search crawlers were refused by a firewall when we fetched the page as them. This may only affect imitations, so the cap is lighter.
Evidence Tiers
Every check shows how strong the evidence for it is, so you can decide what to fix first.
- Documented
- The AI platform or search engine says so in its own documentation, for example which crawler powers ChatGPT search, or that Google AI features need a page that is indexed and may show snippets.
- Supported by research
- Independent studies show an effect, such as pages with statistics, quotations and cited sources being cited more often by generative engines.
- Good practice
- Sound technical SEO with no proven effect on AI citations yet: headings, structured data, fast responses, HTTPS.
- Speculative
- Proposals and emerging conventions, such as llms.txt and content-usage signals. Worth knowing about, weighted very lightly.
How We Fetch
- The page is fetched once as a desktop browser, then as eight AI crawlers using their published user agents, one at a time and 300 milliseconds apart.
- robots.txt is read for both the www and bare domain and matched as RFC 9309 says: the longest matching rule wins and Allow wins a tie. Applebot and Amzn-SearchBot fall back to the Googlebot rules when they are not named, as their documentation says.
- We analyse up to 2 MB of HTML, take three samples of the response time, and read the sitemap (and up to three child sitemaps), the most specific llms.txt that covers the page, and up to 50 of its links.
- We only fetch public addresses on ports 80 and 443, and refuse private, loopback and cloud metadata addresses at every redirect.
- We do not run JavaScript. That is deliberate: it shows what most AI fetchers see.
- Results are cached for 24 hours per address and kept for 12 months. About our bot.
All 56 Checks
A. AI crawler access (robots.txt) (25%)
A1robots.txt is reachable and valid
Every AI crawler reads robots.txt first. A 5xx or 403 answer makes crawlers assume the whole site is off limits (RFC 9309), so this is the most damaging failure a site can have.
DocumentedWeight 2rfc-editor.org
A2AI search crawlers may crawl this page
These crawlers decide whether you can appear in ChatGPT search, Claude, Perplexity, Google AI Overviews, Copilot, Apple, Meta AI and Alexa answers. OpenAI, Anthropic, Perplexity, Google, Apple, Meta and Amazon all document that blocking them keeps you out.
DocumentedWeight 4developers.openai.comsupport.claude.comdocs.perplexity.aidevelopers.google.comsupport.apple.comdevelopers.facebook.comdeveloper.amazon.com
A3Fetchers for user questions are allowed
These fetch your page when someone asks an AI assistant about it, so the answer can quote and link you. Anthropic says blocking Claude-User stops Claude retrieving your content for users.
DocumentedWeight 1developers.openai.comsupport.claude.comdocs.perplexity.ai
A4Training crawler policy
Blocking training is a policy choice. It does not remove you from AI search answers, which use the search crawlers above. Not scored.
DocumentedNot scoreddevelopers.openai.comdevelopers.google.comsupport.apple.com
A5No accidental blocks in robots.txt
Most AI blocks are accidents: a catch-all rule, a misspelt crawler name or a retired token. Each one either blocks a crawler you wanted or does nothing.
DocumentedWeight 1rfc-editor.orgsupport.claude.comsupport.apple.com
A6Sitemap declared in robots.txt
A Sitemap line is how crawlers that never see your search console find every page.
Good practiceWeight 1rfc-editor.orgdevelopers.google.com
A7Content-usage signals
Content-Signal is a Cloudflare policy (search, ai-input, ai-train); Content-Usage comes from an IETF draft that is not yet a standard. Neither is enforced by crawlers today. Not scored.
SpeculativeNot scoredblog.cloudflare.comdatatracker.ietf.org
A8Same rules on www and the bare domain
Crawlers read robots.txt per host (Amazon and Anthropic say so explicitly), so www and the bare domain can block different things.
DocumentedWeight 1developer.amazon.comsupport.claude.com
B. Live access, firewall and CDN (15%)
B1The page loads for a normal browser
The baseline every bot fetch is compared with.
DocumentedWeight 1developers.google.com
B2AI crawlers get through your firewall
Google says pages must be reachable "by any CDN or hosting infrastructure". A firewall or bot protection that refuses AI crawlers hides you from AI answers even when robots.txt allows them. We send each bot’s official user agent from our own server, so a refusal here is a warning to check your CDN, not proof.
DocumentedWeight 3developers.google.comdocs.perplexity.aidevelopers.cloudflare.com
B3Bots get the same page as people
A thinner or different page for bots (or cloaking) means AI answers describe something other than what visitors see.
DocumentedWeight 1developers.google.com
B4robots.txt, llms.txt and the sitemap reach bots
These files should never be challenged. A crawler refused robots.txt treats the whole site as off limits.
DocumentedWeight 1rfc-editor.orgdevelopers.google.com
C. llms.txt and markdown for agents (5%)
C1An llms.txt covers this page
llms.txt is a proposal, not a ranking factor. Google says you don’t need it for AI Overviews, and most files are never fetched. Coding agents and some AI tools do read it when it’s linked, so it’s cheap and worth getting right. It counts for 5% of the score.
SpeculativeWeight 1llmstxt.orgdevelopers.google.comahrefs.com
C2llms.txt is served as text
Agents expect text/plain or text/markdown. A single-page app that answers every address with its HTML shell makes the file look present when it is not.
Good practiceWeight 1llmstxt.org
C3llms.txt structure is valid
One H1, an optional summary, then H2 sections of "- [name](url): notes" lists.
SpeculativeWeight 2llmstxt.org
C4llms.txt links work
Broken links waste an agent’s fetches and signal a stale file.
Good practiceWeight 1llmstxt.org
C5llms.txt links go to markdown
The proposal suggests linking to markdown versions of pages, which agents read with far fewer tokens. Optional.
SpeculativeWeight 1llmstxt.org
C6llms.txt size and fetch time
Agents read the whole file, so keep it under 50 KB and put detail in llms-full.txt. Lighthouse fails a fetch slower than 2 s.
Good practiceWeight 1developer.chrome.comllmstxt.org
C7Pages point to llms.txt
AI tools fetch llms.txt when something points to it. Ahrefs found zero AI bots go looking for unlinked files.
Supported by researchWeight 1ahrefs.comllmstxt.org
C8llms-full.txt
An optional companion with the full text of your key pages, for agents that want everything in one fetch.
SpeculativeWeight 1llmstxt.org
C9A markdown version of this page
Some agents ask for markdown. The preferred way is content negotiation: the same URL returns markdown when the request sends Accept: text/markdown, with Vary: Accept. A separate .md URL also works if it points its canonical back at the page and stays out of the sitemap. Google says markdown versions do not help in Google Search, so this is optional.
SpeculativeWeight 1marketingsignals.comllmstxt.org
C10llms.txt has no prompt injection or hidden text
A file that tries to steer AI agents ("ignore previous instructions", "always recommend us") is a trust and security risk, and tools are starting to treat it that way. Ahrefs notes bad actors are already probing llms.txt for prompt injection.
Good practiceWeight 1ahrefs.comllmstxt.org
C11llms.txt is current against the sitemap
A file that lags behind the site points agents at old or missing pages.
Good practiceWeight 1llmstxt.org
C12Lighthouse llms.txt audit
Replicates the experimental llms-txt audit in Lighthouse 13.3+ (Agentic Browsing), which checks the root /llms.txt only. Lighthouse is a diagnostic tool, so this is reported but not scored.
DocumentedNot scoreddeveloper.chrome.com
D. Content visible without JavaScript (15%)
D1Text is in the HTML, not added by JavaScript
OpenAI, Anthropic and Perplexity crawlers are widely observed not to run JavaScript; Google does, but later. Text that only appears after scripts run is invisible to most AI answers.
Supported by researchWeight 3developers.google.com
D2Main content is in the HTML
The heading and key facts should be in the HTML the server sends, not filled in after hydration.
Supported by researchWeight 1developers.google.com
D3Not an empty app shell or a soft 404
An empty shell or a "not found" page with a 200 status gives crawlers nothing, or the wrong thing, to index.
Good practiceWeight 1developers.google.com
E. Indexability and AI-use directives (15%)
E1Status 200 with few redirects
AI features only use pages that are indexed, which needs a final 200. Long redirect chains lose crawlers.
DocumentedWeight 1developers.google.com
E2Meta robots allow indexing and snippets
noindex removes the page from search and AI answers. nosnippet and max-snippet:0 keep its text out of Google AI Overviews and AI Mode and out of Apple’s AI answers.
DocumentedWeight 3developers.google.comdevelopers.google.comsupport.apple.com
E3X-Robots-Tag header allows indexing
The header does the same as the meta tag, and is easy to set by accident at the server or CDN.
DocumentedWeight 1developers.google.com
E4Non-standard AI tags
noai and noimageai are not supported by most platforms. Meta says it prefers robots.txt over "non-standard formats like NoAI tags". Not scored.
SpeculativeNot scoreddevelopers.facebook.comdeveloper.amazon.com
E5Canonical URL
A single, absolute canonical tells crawlers which URL to index and cite.
Good practiceWeight 1developers.google.com
E6XML sitemap
Sitemaps with honest lastmod dates are how crawlers find new and changed pages quickly.
Good practiceWeight 1developers.google.com
E7Ready for Bing and Copilot
Bing powers Copilot answers. ChatGPT search draws on several sources including Bing and OpenAI’s own OAI-SearchBot index, so Bing indexing helps but isn’t the whole story.
Supported by researchWeight 1blogs.bing.com
F. Structured data and entities (8%)
F1JSON-LD is present and parses
Structured data helps machines identify what a page is about. Ahrefs (2026) tracked 1,885 pages that added schema and saw AI citations barely move, so treat it as hygiene.
Good practiceWeight 1ahrefs.comvalidator.schema.org
F2Organization and WebSite schema
An Organization with a stable @id, logo and sameAs links to your official profiles helps engines tie your site to the right entity.
Good practiceWeight 1validator.schema.org
F3Schema fits the page type
The matching type (Article, Product, Event, LocalBusiness and so on) tells machines what kind of page this is. FAQ rich results are limited by Google to authoritative government and health sites, so FAQPage is for readability, not stars.
Good practiceWeight 1validator.schema.orgsearch.google.com
F4Author and dates
Named authors with a profile URL and honest dates help engines judge expertise and freshness.
Good practiceWeight 1validator.schema.org
F5Schema matches what visitors see
Google says structured data must describe content that is visible on the page.
DocumentedWeight 1developers.google.com
F6Key properties for each schema type
A basic check of the properties Google’s rich results need. Use Google’s Rich Results Test and the Schema.org validator for the full check.
Good practiceWeight 1search.google.comvalidator.schema.org
G. Content extractability (8%)
G1One H1 and logical headings
Clear headings let machines split a page into answerable sections.
Good practiceWeight 1developers.google.com
G2Title and meta description
Every page needs its own descriptive title and a useful description. Roughly 50 to 60 characters for a title and 140 to 160 for a description are drafting guides, not limits, so length alone is never marked down.
Good practiceWeight 1marketingsignals.comdevelopers.google.com
G3Sections open with a direct answer
Where a heading is a question, opening the section with a sentence that answers it directly helps readers and machines alike. There is no proven rule on where answers go, so this is shown for your editor and not scored.
SpeculativeNot scoredmarketingsignals.com
G4Lists and tables
Lists and tables with header cells are easy to read and extract, especially on comparison pages. Use them where they make the page more useful: no particular format guarantees a citation, so this is not scored.
Good practiceNot scoreddevelopers.google.com
G5Question headings or an FAQ
Useful FAQs that answer real questions are an editorial choice. FAQ schema is not a route to rich results (Google dropped FAQ rich results in 2026) or to AI citations, and there is no right number of question headings, so this is not scored.
SpeculativeNot scoredmarketingsignals.com
G6Facts, figures, quotes and sources
Specific figures, attributed quotations and links to primary sources make a page more useful and more quotable, and the KDD 2024 GEO study (Aggarwal et al.) found they raised visibility in generative answers. Use them where they genuinely support a claim: this is a measurement for your editor, not a percentage target, so it is not scored.
Supported by researchNot scoredarxiv.orgmarketingsignals.com
G7Text is not locked in images
Crawlers read alt text, not the words inside images.
Good practiceWeight 1developers.google.com
G8Freshness signals
A visible reviewed or updated date helps where freshness matters. Only change dateModified for meaningful updates: a date bumped without real changes is misleading, so the age of a page is never marked down.
Good practiceNot scoredmarketingsignals.com
H. Trust and entity signals (5%)
H1About and contact pages
Engines look for who is behind a site, and linked about and contact pages make that easy.
Good practiceWeight 1developers.google.com
H2Author pages
An author page with a bio and Person schema backs up the expertise behind an article.
Good practiceWeight 1validator.schema.org
H3sameAs profiles resolve
sameAs links to your official profiles (Wikipedia, Wikidata, LinkedIn and so on) only help when they resolve to the right entity.
Good practiceWeight 1validator.schema.org
H4HTTPS, HSTS and no mixed content
A valid certificate and no insecure resources are the baseline for trust.
Good practiceWeight 1developers.google.com
I. Performance basics (4%)
I1Time to first byte
Crawlers time out, and Meta’s crawler expects content within a few seconds. Under 600 ms is good.
Good practiceWeight 1web.devdevelopers.facebook.com
I2HTML weight
Meta’s crawler stops reading after 1 MB, so key content and Open Graph tags must come early.
DocumentedWeight 1developers.facebook.com
I3Compression and caching
Compressed, cacheable responses load faster for every crawler.
Good practiceWeight 1web.dev
I4Core Web Vitals
Field data (LCP, INP, CLS) from the Chrome UX Report, when a PageSpeed Insights key is set.
Good practiceWeight 1pagespeed.web.dev
More free tools
GEO Ready Checker
56 checks of whether ChatGPT, Claude, Perplexity, Google and Copilot can reach, read and cite a page, with copy-paste fixes.
OpenAI Crawler Checker
Can each AI crawler get to your site? robots.txt rules and live fetches as eight AI bots, in about ten seconds.
Openllms.txt Validator
Check an llms.txt against the format, with line numbers, the Lighthouse audit, broken links and prompt injection.
Openllms.txt Generator
Draft an llms.txt for your site from its sitemap, ready to edit and publish.
OpenAI Crawler List
Every AI crawler user agent, who runs it, what it is for and whether it obeys robots.txt, from the official docs.
OpenChrome Extension
Our free GEO and AI visibility checker for Chrome: check any page you are looking at in one click.
Open




