GEO · Measurement method
What is GEO readiness and how is it measured?
GEO readiness is a reproducible score of how easily a generative answer engine can find, parse and quote a page. Kestralia measures it with nine weighted checks over the HTML: entity markup, answer-first structure, question coverage, FAQ markup, authorship, citations, freshness, llms.txt and passage length. The score runs without any AI provider key, and it never predicts citations.
Last updated: 2026-11-07·By Kestralia editorial team
- Nine checks, weights summing to 100.
- Deterministic over public HTML: the same crawl always produces the same score.
- Grades: strong 75+, developing 45–74, weak below 45.
- No AI provider key required.
Definition
GEO readiness is the degree to which a page is ready to be extracted — not a forecast of ranking or citation. Answer engines do not treat a page as a link; they assemble a passage that has to make sense on its own, be verifiable, and be attributable. Readiness measures exactly those conditions: a resolvable entity, an answer near the top, question-shaped sections, a real author, primary sources, a visible date and short sentences.
The score is computed over the public HTML of crawled pages. Login, signup, password-reset, invite, verification and app screens are excluded so they do not dilute a marketing-content score.
Readiness is not visibility
These are two different measurements, and conflating them is where most "AI SEO" claims break. Readiness comes from page structure and anyone can reproduce it. Visibility comes from a real answer and depends on the provider, model, language, region and date.
| Question | GEO readiness | Answer visibility |
|---|---|---|
| What it answers | Can a machine extract and quote this page? | Did a specific engine mention or cite the brand in a specific answer? |
| Input | Public HTML of crawled pages | A query to an authorized API |
| Determinism | Same HTML, same score | Non-deterministic, personalized and dated |
| What it proves | Structural extractability | One observation, not a ranking |
The nine checks and their weights
The weights add up to 100. Each check resolves to one of three states: pass (full weight), warning (half weight) or fail (zero). The table below is what the scoring engine actually computes, not a marketing summary.
| Check | Weight | What it inspects, and the threshold |
|---|---|---|
| Entity markup identifies the brand | 16 | Homepage structured data contains Organization or LocalBusiness together with WebSite or WebPage. Name, URL and logo must resolve as one entity. Both groups = pass, only some types = warn, none = fail. |
| Pages open with a direct answer | 14 | Words in the first block after the H1, up to the next H2. 25–120 words = pass, zero = fail, anything else = warn. |
| Content covers real questions | 12 | Share of pages with at least two question-shaped headings (what, how, why, which, or a trailing question mark). Pass at 30% of pages. |
| FAQ or Q&A structure exists | 10 | At least one page exposes FAQPage or QAPage markup. |
| Authorship and evidence are visible | 12 | Pages with Person schema, an author field or a visible byline. Pass at 25% of pages. |
| Claims link to primary sources | 12 | Pages linking to primary sources: standards bodies, government, research or official documentation. Pass at 20% of pages. |
| Freshness is machine-readable | 10 | dateModified or datePublished in markup, or a visible review date. Pass at 30% of pages. |
| llms.txt guidance published | 6 | A /llms.txt file exists. This is an emerging convention: absence warns, it never fails. |
| Passages are self-contained | 8 | Share of pages averaging more than 34 words per sentence. Pass when at most 40% of pages exceed it. |
How the score is calculated
Score = earned weight ÷ 100 × 100, rounded to an integer. Three grades: 75 and above is strong, 45–74 is developing, below 45 is weak.
Every check is a deterministic rule over crawled HTML, so the same crawl always produces the same number. This score is separate from Search Health, which weighs technical findings by severity and prevalence. That method is documented in the scoring methodology.
How to audit GEO readiness yourself
You do not need a special tool for a real check. Run these eight steps on the ten most important pages of your site:
- Fetch the homepage HTML and read its structured data: Organization and WebSite with name, url, logo and sameAs.
- Read the first 100 words after the H1. They should answer the page's main question without needing earlier context.
- Count question-shaped headings on each page. Fewer than two is a missed opportunity.
- Add one genuinely useful FAQ — only where users actually ask those questions — and mark it up with FAQPage.
- Put a real author with credentials on substantive pages, not "admin".
- Link every statistic to its primary source and write the review date on the page.
- Publish dateModified in Article schema and visibly on the page.
- Keep sentences short, one idea per paragraph, and write descriptive H2s so each section makes sense when lifted out.
How visibility in AI answers is measured
There is no queryable "memory of AI". You choose a fixed prompt set, run it against an authorized API, and store the provenance of every observation. Kestralia records, per row: the prompt, language, provider, model, date, whether the brand appeared in the answer, whether the site was cited, which competitors appeared and which domains were cited.
One run is a sample. A sample is not a rank, and a missing mention is not proof of invisibility. When no authorized API is available, the record stays "not measured" — an estimate is never substituted for an observation.
What this score does not tell you
- It does not predict or guarantee a citation or mention by ChatGPT, Gemini, Perplexity, Copilot or Google AI Overviews.
- It does not measure rankings, traffic or revenue.
- AI answers are non-deterministic and personalized; one observation is not a ranking.
- llms.txt is an emerging convention, not a control over model behaviour.
- A high score will not rescue a page that has nothing original to say.
How to evaluate any GEO vendor's claim
- Does it name the model and the date for every observation? Without both, the number cannot be reproduced.
- Does it separate deterministic readiness from a sampled observation?
- When it has no access, does it say "not measured" — or present an estimate as fact?
- Does it publish the prompt set and its language? One English prompt tells you nothing about Persian visibility.
- Does it guarantee citations? No honest tool can.
- Can you export the evidence — URL, HTML snippet and timestamp?
Tooling landscape: which question each category answers
This is not a ranking and inclusion is not an endorsement; features change. Each description is taken from the vendor's own public page on 2026-11-07.
| Category | Examples, described by the vendor | Question it answers |
|---|---|---|
| Desktop crawler | Screaming Frog SEO Spider: a site crawler for Windows, macOS and Linux; the free version crawls up to 500 URLs and a licence is £199 per year. Sitebulb: a website crawler for SEO audits, on desktop and in the cloud. | What is technically broken on the site? |
| Search data platform | Ahrefs describes itself as an AI marketing platform covering AI search, SEO, content and social. | How are search demand and visibility trending? |
| AI search tracker | Peec AI describes itself as an AI search platform for marketing teams. | Where does the brand appear in AI answers over time? |
| Combined audit platform | Kestralia: technical crawl, nine-check GEO readiness and direct observation with provenance on every record. | What should we fix first, and what actually changed? |
Sources
- Google Search Central documentation — official Google documentation for crawling, indexing and structured data.
- Schema.org Organization — the official definition of the Organization entity and its name, url, logo and sameAs properties.
- llmstxt.org — the proposal to standardise an /llms.txt file; explicitly a proposal, not a finished standard.
- Screaming Frog SEO Spider — product page: site crawler, free up to 500 URLs, £199 per year licence (2026-11-07).
- Sitebulb — product page: website crawler for SEO audits, desktop and cloud (2026-11-07).
- Ahrefs — homepage: describes itself as an AI marketing platform (2026-11-07).
- Peec AI — homepage: describes itself as an AI search platform for marketing teams (2026-11-07).
Frequently asked questions
What is GEO readiness?
A reproducible score of how well answer engines can find, extract and quote your pages. It is computed from public HTML and does not need an AI provider key.
How is GEO readiness measured?
With nine weighted checks that add up to 100: entity markup, answer-first structure, question coverage, FAQ markup, authorship, citations, freshness, llms.txt and passage length. Each check is a pass (full weight), a warning (half weight) or a fail (zero).
Is GEO readiness the same as AI answer visibility?
No. Readiness comes from page structure and is deterministic. Visibility comes from a real answer, is non-deterministic, and depends on the provider, model, language and date.
Does a high GEO readiness score guarantee AI citations?
No. No score can guarantee model behaviour. The score says the content is extractable; it does not say it will be quoted.
Does measuring GEO readiness require an API key?
No. Readiness is deterministic and works without a key. A key is only needed for direct observation of answers; without it the record stays "not measured".
What is a good GEO readiness score?
75 and above is treated as strong, 45–74 as developing and below 45 as weak. Those thresholds are relative to the crawled sample and the version of the ruleset.