Your GEO Score, category by category

Citation Worthiness (34 of 100 points)

Citation Worthiness is the category that asks: does this page carry facts an answer engine can quote, and show where they came from?

Answer engines cite the page that supplies the number, the definition or the named source. v6.0 separates facts that are sourced (a link, a citation or an attribution next to the figure) from facts that are only stated, because a figure with a source can be checked and quoted with confidence.

• Sourced facts per 1,000 words, 8 pts: Specific facts (percentages, counts with units, money, measurements, dated events) with a link, a citation or an attribution phrase within 200 characters. Marketing claims such as 'up to', superlatives and adoption counts are not facts. 5 per 1,000 words earns 8; 3 earns 6; 2 earns 4; 1 earns 2.
• Fact density per 1,000 words, 4 pts: All specific facts, sourced or not. 12.5 or more earns 4; 8 earns 3; 5 earns 2; 2 earns 1. Copyright years, navigation and footer text and prices outside product pages are not counted.
• Topic fact coverage, 8 pts: The share of the facts that currently-cited pages for this page's topic agree on that this page also states. 40% earns 8; 30% earns 6; 20% earns 3; 10% earns 1. Not tested when no claim set exists, and then left out of the denominator.
• Attributed quotations, 5 pts: A blockquote, <q>, figure with a caption or a quoted span of ten words or more, with a citation, an attribution verb or a person or organisation name beside it. Three earn 5; two earn 3; one earns 2.
• External source citations, 5 pts: Unique outbound links in the main content placed next to a factual claim. 6 earns 5; 4 earns 4; 2 earns 2; 1 earns 1. Your own domain and social profiles do not count, and .gov, .edu and .org links are not weighted extra.
• Comparison table or grid, 2 pts: A real table with header cells and two or more data columns, or an aligned comparison grid of two or more items sharing three or more labelled attributes. Layout tables do not count.
• Quotable definitions, 2 pts: A <dfn>, a definition list, 'X is a ...' under a heading about X, or 'defined as', 'refers to' or 'means'. Three or more earn 2; one or two earn 1.

The second version states figures and links the source beside them, so each can be verified and quoted. The first makes claims that are excluded from the count.

• Replace adjectives and 'up to' claims with figures, and put the source in the same sentence.
• Link each statistic to its primary source.
• Quote named people or organisations, in their own words, with the attribution beside the quote.
• Add one real comparison table and a clear definition for each main term.

On paid plans, GeoTest writes this fix for your page, from your own content, with the points it recovers.

Answer Structure (22 of 100 points)

Answer Structure is the category that asks: can an engine lift a complete, self-contained answer from this page without reading all of it?

Retrieval works section by section. The score rewards sections that answer their own heading, passages that name their subject so they make sense when quoted alone, and Q&A that is visible on the page in whatever form your site uses, not only in one /faq path.

• Self-contained answer openings, 8 pts: The share of H2/H3 sections whose first sentence is at least 8 words, has no preamble ('In this section', 'Let's look at') and shares a term with its heading. 80% earns 8; 60% earns 6; 40% earns 4; 20% earns 2.
• Heading hierarchy, 4 pts: Exactly one H1 (+2), and enough H2 sections for the length of the page (one per 400 words, at least one) with no skipped levels (+2; +1 if there are some H2s but fewer).
• Q&A block on this page, 4 pts: FAQPage or QAPage schema backed by visible text, or three or more question-and-answer pairs from accordions, definition lists or headings with answers of 15 words or more. Outside a section or page labelled FAQ, questions or help, a heading counts only if it ends with a question mark. 3+ pairs earn 4; 1 to 2 earn 2.
• Question depth on this page, 2 pts: Unique question-and-answer pairs: 8 or more earns 2; 4 or more earns 1.
• FAQPage JSON-LD backed by visible text, 2 pts: Valid FAQPage markup where at least half the questions appear as visible text. Markup with no visible Q&A scores 0 and is flagged.
• Self-contained passages, 2 pts: Sections of 40 to 300 words that name their subject instead of opening with It, This or They. 70% earns 2; 40% earns 1.

The opening sentence of the second section names the product, answers the heading and carries the figures, so it still works when an engine quotes it alone.

• Rewrite the first sentence under each heading so it answers the heading and names its subject.
• Show questions and answers as visible text, and mirror them in FAQPage markup only if you do.
• Use one H1 and enough descriptive H2s for the length of the page.

On paid plans, GeoTest writes this fix for your page, from your own content, with the points it recovers.

Authority and Trust (17 of 100 points)

Authority and Trust is the category that asks: can an engine tell who is behind this page, how to reach them and who vouches for them?

These signals are now read from how the page is built, not from single words. A testimonial is recognised by its structure (a quote with a named person or company beside it), a byline by its position and markup, and contact details by working methods, not the word 'street'.

• Social proof, 5 pts: Evidence of six kinds: an attributed testimonial (a quote with a named person or company in the same block); a rating or review shown on the page or in Review or AggregateRating schema backed by a visible rating; a customer logo row of three or more named marks; a case study or result naming a customer; independent recognition (an award, accreditation, certification or a press row); or an adoption count of 100 or more. One kind earns 3; two or more earn 5. Legal pages never qualify.
• Author byline, 4 pts: A person author in JSON-LD, rel=author, a meta author tag or an author box, or 'written by' or 'reviewed by' with a name near the start or end of the main content. A name alone earns 2; with a role, credentials or an author page link, 4. Names in comments, related-post cards and sidebars do not count.
• Contact methods, 3 pts: On-page ways to reach you: a tel: link, a mailto: link, telephone, email, contact point or address in schema, or a working contact form. Two or more earn 3; one earns 2; a link to a contact page alone earns 1.
• Privacy and Terms links, 2 pts: Footer or navigation links to privacy and terms that resolve to a live page (HTML or PDF). One combined policy covering both earns 2; one earns 1.
• Contextual internal links, 3 pts: Links in the main content to indexable live pages with descriptive anchor text. On pages under 500 words one link earns 3; otherwise 3 or more earn 3 and one or two earn 1.

A quote with a named person beside it is recognised as an attributed testimonial without needing a keyword such as 'testimonial' or 'review'.

• Put a real name and a role beside every testimonial, in the same block as the quote.
• Show a visible rating if you publish Review or AggregateRating markup.
• Put a named author with a role on articles, near the top or the end.
• Give at least two working ways to reach you on every non-legal page.

On paid plans, GeoTest writes this fix for your page, from your own content, with the points it recovers.

Structured Data (12 of 100 points)

Structured Data is the category that asks: is the page's meaning declared in a form a machine can read, and does it match what the page says?

Schema states what a page is and who published it. The score checks that the type fits the page, that the required properties are present, and that names, ratings, prices and dates agree with the visible text. Every page type is held to the schema that fits it.

• Primary schema type matches the page, 4 pts: The JSON-LD type fits the page type: Organization or WebSite on the homepage, Article or BlogPosting on articles, Product, Service or SoftwareApplication on product pages, FAQPage on FAQ pages, LocalBusiness on location pages, and so on.
• JSON-LD parses, 1 pts: Every JSON-LD block is valid.
• Schema properties present and agree with the page, 1 pts: The required properties for each type are present, and names, ratings, prices and dates match the visible text.
• BreadcrumbList, 1 pts: BreadcrumbList on inner pages.
• Supporting schema types, 1 pts: Review, AggregateRating, Person, HowTo or ItemList where the page has that kind of content. Not applicable when it does not.
• Meta description + OG + Twitter card, 4 pts: A meta description of 80 to 200 characters, Open Graph title and description, and a Twitter card.

The second block carries the required name and URL. The first is valid JSON but has nothing in it a machine can use.

• Emit the schema type that matches each page type from your templates.
• Make sure every ratings, price and date value in schema is also visible on the page.
• Fill the meta description, Open Graph and Twitter card on every page.

On paid plans, GeoTest writes this fix for your page, from your own content, with the points it recovers.

Freshness (7 of 100 points)

Freshness is the category that asks: can an engine date this page, and does the date hold up?

Dates matter for articles, documentation and product pages with prices or specifications. Pages that are not expected to carry a date are not scored on it.

• Recency of last update, 5 pts: A visible date labelled updated, modified or reviewed, then JSON-LD dateModified, then the sitemap's lastmod. Within 3 months earns 5; 6 months 3; 12 months 1. A date known only from the sitemap earns at most 3. Documentation with a last-reviewed date or version marker within 12 months earns 5.
• Machine-readable date agreement, 2 pts: dateModified within 7 days of a visible date earns 2; a schema date with no visible date earns 1; a mismatch earns 0.

A visible date and a matching dateModified in the same week, so the date can be trusted.

• Show a 'Last updated' date on articles, docs and product pages, using the real edit date.
• Keep dateModified in schema within a week of the visible date.

On paid plans, GeoTest writes this fix for your page, from your own content, with the points it recovers.

Extractability (8 of 100 points)

Extractability is the category that asks: can a crawler reach the page, read its text in the first response and know which URL is the original?

If a page is noindexed no amount of content helps, and a noindex caps it at 30. v6.0 also compares the text in the initial HTML with the text after JavaScript runs: content that exists only after rendering counts as present but is flagged as an extraction risk here, never as missing content.

• Not noindex, 3 pts: No noindex in a meta robots tag or an X-Robots-Tag header.
• Rendered parity, 2 pts: The share of the page's rendered main-content words already present in the initial HTML. 90% earns 2; 60% earns 1.
• Canonical tag, 2 pts: Present, absolute, resolves to a live indexable page and matches the page or the sitemap URL. A canonical with a problem earns 1.
• Mobile viewport meta, 1 pts: A viewport meta tag.

The text is in the first response, so crawlers that do not run JavaScript still read it.

• Remove noindex from every page you want cited.
• Server-render or pre-render public pages.
• Point the canonical at the page's own live address.

On paid plans, GeoTest writes this fix for your page, from your own content, with the points it recovers.

What your total means

Your total falls in one of these bands, the one scale used everywhere in the product: the badge, the dashboard, the shared report and the PDF all read them. The GEO Score benchmark shows where public sites land on the same scale.

• 90–100, A: EXCELLENT
• 80–89, B: GOOD
• 65–79, C: FAIR
• 50–64, D: MODERATE
• 35–49, E: POOR
• 0–34, F: CRITICAL

The site layer (100 points)

The site layer adds 100 points for what only a crawl can prove, on top of the six page categories above. A site score is 70% the weighted average of its page scores (homepage counted twice, contact and location pages counted half, legal pages left out) and 30% this layer.

• Topical cluster depth, 12 pts: Groups of three or more substantive pages that share two or more terms from their titles, H1s or URL slugs, with two or more contextual links between them and no pair mostly duplicated. Three clusters earn 12; two earn 8; one earns 4.
• About or Team page, 10 pts: A page found by URL, title or AboutPage schema with 100 or more words about the organization. 10, or 5 if thin.
• Entity completeness, 10 pts: Organization JSON-LD with name and URL (2), a logo (2), two or more profile links that resolve and match the brand (2), a contact point or address (2), and a founding date or founder (2). Capped at 6, with a flag, if the schema name, the visible brand, the hostname and the profiles disagree.
• robots.txt allows search-indexing crawlers, 10 pts: 2 points for each of the five search-indexing crawlers (Googlebot, Bingbot, OAI-SearchBot, PerplexityBot, Claude-SearchBot) that is allowed, 1 if allowed at the root but blocked on substantive paths, 0 if blocked. A robots.txt that cannot be read is unknown, not blocked.
• Page breadth, 8 pts: Substantive pages (indexable, canonical, 150 or more words, not tag, utility or legal pages): 10 or more earns 8, 5 earns 5, 3 earns 3.
• Q&A surface, 8 pts: Any help, FAQ or knowledge-base page, any page with five or more question-and-answer pairs, or FAQPage markup with five or more visible questions. No /faq path is required.
• sitemap.xml present, healthy and complete, 8 pts: Exists and parses (3), 90% or more of its URLs return 200 and are indexable and canonical (3), and it covers 80% or more of the crawled substantive pages (2).
• Site-wide fact density, 7 pts: The median fact density of editorial pages: 12.5 per 1,000 words earns 7, 8 earns 5, 5 earns 3, 2 earns 1.
• Cross-page schema consistency, 7 pts: Organization on the homepage (3), valid JSON-LD on half or more of inner pages (2), and no conflicting entity names or URLs across pages (2).
• Internal link health, 6 pts: Orphaned substantive pages at 10% or fewer (3), and no broken internal links (3).
• Site Q&A depth, 5 pts: Unique question-and-answer pairs across the site: 15 or more earns 5, 8 earns 3, 3 earns 1.
• Contact surface, 5 pts: A contact page that returns 200, or two or more contact methods across the site.
• Privacy and Terms pages, 4 pts: Privacy and terms both crawled and returning 200 after redirects (HTML or PDF). One combined policy counts as both; one document earns 2.
• llms.txt (reported, not scored), 0 pts: Reported as a recommendation. It carries no points until citation data shows it matters.

Six things override a total

Each cap below overrides a total: while its trigger is present, the page or site cannot score above the cap, whatever the categories add up to.

• This page is set to noindex caps the page score at 30. A noindex tag in the page or in an X-Robots-Tag header caps the page at 30 until it is removed.
• Under 150 words of main content caps the page score at 40. A page with fewer than 150 words of main content has little to quote. This applies to articles, documentation, product, case study and about pages, not to short pages such as contact and legal.
• 150 to 299 words of main content caps the page score at 70. A short page can still be dense and well structured, but it cannot score above 70. The same page types apply.
• Placeholder or Lorem copy detected caps the page score at 50. Lorem ipsum, or three or more unfinished-site phrases, mean the page is unfinished. It is capped at 50 until the copy is real.
• robots.txt disallows all crawlers, or blocks Googlebot or Bingbot site-wide caps the site score at 30. If the main search crawlers cannot read the site, the site layer and the site score cannot rise above 30.
• Two or more AI search-indexing crawlers blocked caps the site score at 70. Blocking two or more of OAI-SearchBot, PerplexityBot and Claude-SearchBot, site-wide or on substantive paths, caps the site layer and site score at 70. User-triggered fetchers and training crawlers are reported but never penalized.

When a cap applies, your report also shows the score you had earned before it, so you can see exactly what clearing the cap is worth.

Quick Scan vs Site Analysis

A Quick Scan scores one page by the same rules a Site Analysis applies to every page in its crawl. The free GEO Score checker runs a Quick Scan in about 15 seconds. A single-page scan reports that page score, runs the robots.txt, sitemap and llms.txt probes as uncounted context, and names the site-layer checks it could not run: Topical cluster depth, About or Team page, Entity completeness, Page breadth, Q&A surface, Site-wide fact density, Cross-page schema consistency, Internal link health, Site Q&A depth, Contact surface, Privacy and Terms pages.

Questions about the score

Each answer below about the score is taken from the published v6.0 rubric; the GEO glossary defines the terms it uses.

What does the GEO Score measure?

The GEO Score measures how reliably a site exposes useful, attributable, structured and technically accessible information to search engines and AI retrieval systems. It does not measure business quality, rankings, traffic, or guaranteed AI citation. Crawler access raises eligibility; it does not guarantee inclusion in any AI answer.

What is a good GEO Score?

On the GEO Score, 90 and above is the top band (EXCELLENT). 80 to 89 is GOOD, 65 to 79 FAIR, 50 to 64 MODERATE, 35 to 49 POOR, and 0 to 34 CRITICAL. The bands are the same ones used everywhere in the product, including the badge and the PDF report.

What are the four result states?

Every check ends in one of four states. Present: verified evidence was found. Absent: the page was examined and nothing qualified; it scores 0 and stays in the total. Not applicable: the check does not fit that page type, so it leaves the total. Unknown: the page could not be read reliably (a failed render, a bot wall, a file that did not answer), so it leaves the total and is never shown as a missing element.

What is coverage?

Coverage is the share of the expected evidence that was actually evaluated. Every report shows it beside the score. A single page under 75% coverage shows 'Insufficient coverage' instead of a grade, and a site score under 75% is labelled provisional.

How does GEO v6.0 decide something is present?

To decide something is present, GEO v6.0 looks at schema first, then page structure, then narrow language, and finally an AI second opinion on anything still absent. The AI can only raise an absent result, only with a quote that GeoTest verifies is on the page, and its input and result are cached so the same page scores the same. Counts, densities and technical checks use their own fixed rules.

What if my testimonials load with JavaScript?

Testimonials that load with JavaScript still count as present. If they are visible only after JavaScript runs and are absent from the initial HTML, the report says so and the cost falls under Rendered parity in Extractability, never under social proof, because some crawlers and AI systems may extract that content inconsistently.

Which crawlers affect the robots.txt score?

Only search-indexing crawlers: Googlebot, Bingbot, OAI-SearchBot, PerplexityBot and Claude-SearchBot. User-triggered fetchers (ChatGPT-User, Perplexity-User, Claude-User) and training crawlers (GPTBot, ClaudeBot, Google-Extended and others) are reported but never penalized.

Why did my score change from v5.0?

Your score can differ from v5.0 because v6.0 is a new scoring engine. v5.0 scores stay archived under their own version and are not compared with v6.0 scores, so a step in your history marks a change of instrument, not a change in your site.

Which scoring version is this?

The current scoring version is GEO Score v6.0: a page layer out of 100 and a site layer out of 100, combined 70/30 for a site score. Weights, caps, bands, rounding and the crawler registry version are pinned per version, and every scan stores the detector, content extractor, registry, model and prompt versions it was produced under.

AI search in figures

These AI search figures show how fast answers are replacing clicks, and each one is dated and sourced. Generative Engine Optimization (GEO) is a method for structuring a website so AI answer engines can read, trust and cite it. A citation is defined as a link to your page inside an AI answer, and the GEO Score refers to a 0 to 100 rating of how ready a page is to earn one. These published figures show why it matters; the reasoning behind every weight is in the methodology.

How much do AI summaries reduce clicks?

AI summaries cut clicks by about half: Pew Research Center found searchers clicked a result 8% of the time when an AI summary appeared, versus 15% without one, a 47% drop (July 2025).

How common are zero-click searches in 2025?

Zero-click searches rose from 56% to 69% between May 2024 and May 2025, and 83% of searches end without a click when AI Overviews show, according to Similarweb (2025).

What share of websites are ready for AI search?

Very few websites are ready: only 2% of the 537 sites in the GeoTest benchmark scored 80 or more in September 2026, and the median score was 47. Those scores came from the rubrics in use before GEO v6.0.

How much search traffic are publishers losing?

Publishers are losing about a third of their Google referrals: across 2,500+ publisher websites, Google search referral traffic was down 33% globally and 38% in the US year over year as of early 2026 (Chartbeat).

How many buyers research with AI chatbots in 2026?

71% of B2B software buyers rely on AI chatbots for software research (G2, 2026), which is why being cited matters as much as ranking. See what GEO is.

Find GeoTest on