AI Citation Score: What It Measures and How It Is Worked Out

The AI Citation Score is a 0 to 100 rating of how likely an AI engine is to quote a page, built from five areas: can AI reach it (25 points), is there an answer to lift (30), evidence (20), entity and trust (15) and says something specific (10). It is a prediction, not a count of citations.

• Google says nosnippet keeps a page from being a direct input for AI Overviews and AI Mode, and max-snippet:0 is equivalent, so both cost reach points (Google's robots meta tag documentation).
• OpenAI lists 4 crawlers with separate jobs, and only OAI-SearchBot decides whether pages appear in ChatGPT search answers (OpenAI's documentation on its crawlers).
• Google needs 0 special schema.org markup or AI text files for its AI features, which is why structured data carries few of the 100 points (Google Search Central on optimizing for generative AI search).
• 3 caps hold a page down whatever its other strengths: unreachable pages score at most 10, noindex pages at most 15 and pages under 80 words at most 30.

What does the AI Citation Score measure?

The AI Citation Score measures how easy a page is for ChatGPT, Perplexity, Claude, Google AI and Copilot to reach, understand and quote. It is a prediction built from the page itself, not a record of what the engines did.

The AI Search Visibility Checker from AI Web Maker reads each page as it arrives, without running JavaScript, because that is how most AI bots read it. It then makes 25 checks across the five areas, and every point a page loses is shown with the fix that would win it back. Most checks are counted from the page and the website, while two are judged by a language model: which question the page is best placed to answer and how directly it answers it, and how specific its wording is.

None of the engines publishes how it picks the pages it cites, and Google states that no third-party tool has access to its internal ranking or AI systems. The score therefore rests on what the engines do document, such as which bots they send and which robots rules they obey, and on the qualities their guidance keeps naming: a direct answer, evidence and a clear business behind the page. The weights are a design choice by AI Web Maker, explained in full on the method page, and they will be checked against real citations over time.

How are the 25 points for reach worked out?

The reach area checks whether each engine's search bot is let in and gets the page, whether the page may be indexed and quoted, and whether its words need JavaScript.

Check	Points	How it is counted
AI search bots let in	12	Each of the five engines counts equally: allowed and not refused (full), not confirmed (60%), quoting limited by nosnippet (50%), shut out (0)
Indexable and quotable	5	No noindex; nosnippet or max-snippet:0 keeps 2 of the 5
Words without JavaScript	4	150 words or more in the page as it arrives (4), 60 to 149 (2), a page drawn by scripts (0)
Canonical	2	Empty, or pointing to the page itself
In the sitemap	2	Listed in the XML sitemap; the homepage always counts

The bot check reads robots.txt for each engine's search bot, not its training bot: OAI-SearchBot for ChatGPT, PerplexityBot for Perplexity, Claude-SearchBot for Claude, Googlebot for Google AI and Bingbot for Copilot. OpenAI's documentation explains why the difference matters, since blocking GPTBot only signals that content should not be used for training, while blocking OAI-SearchBot keeps pages out of ChatGPT search answers. Training bots such as GPTBot, ClaudeBot and Google-Extended are listed separately in the report and never cost points.

The checker also asks for the homepage as each bot, which catches a firewall that turns a bot away even when robots.txt welcomes it. A firewall may refuse a request made as Googlebot or Bingbot from the checker's servers while still letting the real crawlers in, so those two are marked "could not confirm" rather than counted as shut out. User fetchers such as ChatGPT-User and Perplexity-User do not decide the result either, because their makers say robots.txt may not govern them.

How does the score decide whether there is an answer to lift?

The answer area, worth 30 points, checks whether the page answers a clear question near the top and whether its sections are built so a passage can be lifted whole.

Check	Points	How it is counted
Answers its question	10	Rated by a language model: answered in the opening lines (10), clearly but lower (8), vaguely (4), not at all (0)
Opening answer	6	First paragraph of 40 to 60 words (6), 30 to 80 (4), 20 to 120 (2)
Question headings	5	Share of section headings written as questions: 60% or more (5), 30% (3), any (1)
Sections open with an answer	4	Share of sections whose first sentence is 30 words or fewer
Questions and answers	3	4 or more (3), 2 or 3 (2)
Lists and tables	2	One point each

The largest single check is the first one, because a page that never answers its question gives an engine nothing to quote. The checker names the question it thinks the page can win and picks the passage an AI engine would quote, and that passage is kept only if those exact words appear on the page. If the named question is not the one you meant the page to answer, that alone tells you the page is unclear.

The length bands reflect a judgement about what lifts cleanly, not an engine rule. Google says there is no ideal page length and no need to break content into tiny pieces, and the score agrees: it does not reward short pages, only a self-contained answer at the top and sections that open with their point. A long, detailed page with a 50 word answer first scores as well here as a short one.

What counts as evidence in the score?

The evidence area, worth 20 points, rewards figures, primary sources, a visible date, a named author and something the business found or measured itself. Articles and business pages are weighted differently.

Check	Article points	Business page points	How it is counted
Facts with numbers	5	7	Figures per 300 words: 3 or more (full), 1.5 (60%), 0.5 (25%); years alone do not count
Primary sources	6	2	Links to official guidance, standards or research: 2 or more (full), 1 (half)
Updated date	4	3	A visible date and a date in the page data (full), one of them (75%)
Author	3	3	A named author; on a business page, the business data counts
Own data	2	5	Something the business found or measured itself

The split exists because nobody expects a bakery's homepage to cite research. A guide or blog post is judged mainly on its sources and dates, while a home, service, product or about page is judged on its own facts: prices, hours, sizes, counts and results. A hypothetical bakery in Leeds that lists its opening times, its loaf sizes and how many wedding cakes it made last year scores well on evidence without linking anywhere.

What do entity and trust and specific wording check?

Entity and trust, worth 15 points, checks that a clear business stands behind the page. Specific wording, worth 10, checks that the page says something only this business could say.

The entity checks are business data naming the business on the homepage (4 points), sameAs links to its profiles elsewhere (2 for two or more, 1 for one), an About page (2), a Contact page (2), an address or phone in a form machines read (2), structured data describing the page itself (2) and a secure https address (1). Structured data carries only a few points on purpose. Google says its AI features need no special schema.org markup, and adding markup alone has not been shown to raise how often a page is quoted, so the score treats it as a way to state facts clearly rather than a lever.

The specific area gives 8 points from a language model's rating of the wording: facts only this business could give (8), concrete details (6), a few details (3) or generic wording (0). The last 2 points go to pages that do not repeat another page the checker read almost word for word. Generic copy such as "quality service at competitive prices" can be found on thousands of websites, so an engine has no reason to quote yours for it.

What caps a page's AI Citation Score?

Three problems cap a page's score whatever its other strengths: no engine can reach it, it is marked noindex, or it has fewer than 80 words. The caps are 10, 15 and 30 points.

The caps exist because the other checks mean little when one of these is true. A page every AI search bot is shut out of cannot be quoted however good its answer is, and Google's robots meta tag documentation says noindex keeps a page out of search results, which also keeps it away from the AI features built on them. A page under 80 words rarely holds an answer and the evidence behind it, so a high score there would mislead.

When a cap applies, the report says so and the fix comes first. Removing a stray noindex or letting a blocked bot in can lift a page by dozens of points at once, which is why reach problems sit at the top of any list of fixes. OpenAI says a robots.txt change takes about 24 hours to reach its search systems, so access fixes are also the quickest to take effect.

How is the website's AI Visibility Score worked out?

The website's AI Visibility Score is 55% its pages, 20% its access for AI search bots and 25% how often other websites name it.

The pages part is the average of the pages read, with the homepage counted twice because engines and customers both start there. The access part gives 85 points for the five engines, counted the same way as on each page, 10 for a sitemap and 5 for a robots.txt file. The off-site part comes from one web search for the website's address and its name: a website naming the address counts fully, the name alone counts half and only when it is two words or more, and a mention on a community site, a reference site or a review site adds 10 points for each kind.

To show the arithmetic with made-up figures, a website whose pages average 60, whose access part is 85 and whose off-site part is 40 would score 0.55 × 60 + 0.20 × 85 + 0.25 × 40, which is 33 + 17 + 10, or 60. When the web search cannot run, the score is the pages and access parts alone, weighted again to fill 100. The 25% for mentions is there because AI engines answer "which business should I use" questions largely from what other websites say, and no amount of work on your own pages replaces that.

What can the AI Citation Score not tell you?

The AI Citation Score cannot tell you whether any engine quotes your page today; it only predicts how quotable the page is.

AI engines weigh signals no single website shows, such as how a page compares with every other candidate for the same question, and they change how they choose sources. That is why the score belongs beside the measures that see real outcomes: Search Console's Generative AI performance report, Bing Webmaster Tools' AI Performance report, AI visits in Google Analytics 4 and a monthly panel of customer questions. To measure AI search visibility properly, use the score to decide what to fix and those reports to see whether the fixes worked.

Two things are reported without being scored. An llms.txt file is noted but earns nothing, because no AI search engine has said it uses one, and Google says such files are not needed to appear in its Search. Training bots are listed so you can see your choices, but blocking or allowing them changes nothing in the score, since they do not decide citations.

How do you raise a low AI Citation Score?

Raise a low score by fixing the gaps that cost the most points first, starting with reach, then the opening answer, then evidence.

You can scan your website with the AI Search Visibility Checker to get each page's score, its gaps sorted by points lost and the fix for each. Five pages are read with no account, and up to 50 with an account for 1 credit, about $0.65, and the checker changes nothing on the website. Members can have fixes written for their own pages in AI Website SEO Tools and apply them themselves.

Work through a page in the order the score rewards. Open it with a 40 to 60 word answer that names the subject, turn the section headings into the questions customers ask, add your own figures and a date, and replace any sentence a competitor could also have written. Then scan the page again, and watch the engines' reports over the following months to see whether citations follow.

Questions business owners ask about the AI Citation Score

Is a score of 100 needed to be cited?

No. The score predicts how quotable a page is, and engines cite pages with all kinds of scores when they answer a question well. Aim to remove the caps and the largest gaps on your most important pages rather than chasing every point.

Why did my page lose points for Google when it is indexed?

The most common reason is a nosnippet rule or max-snippet:0, which Google says keeps a page from being a direct input for AI Overviews and AI Mode. The page can rank and still lose those reach points. Remove the rule unless you mean to stop Google quoting the page.

Does blocking GPTBot lower my AI Citation Score?

No. GPTBot, ClaudeBot and Google-Extended are training bots, listed apart and never scored. Only the search bots, such as OAI-SearchBot, Claude-SearchBot and PerplexityBot, affect the reach points.

Why do Google AI and Copilot show "could not confirm"?

A firewall may refuse a request made as Googlebot or Bingbot from the checker's servers while letting the real crawlers in. Rather than count that against you, the checker marks those two engines as not confirmed, worth 60% of their reach points.

Does adding schema raise the score a lot?

No. Business data on the homepage is worth 4 points, sameAs links 2 and page data 2, out of 100. Add it to state your facts clearly, but expect larger gains from a clear opening answer and specific evidence.

Score your five most important pages first

Start with the homepage and the four pages that answer the questions customers ask before they buy, because they carry the most weight in the website score.

Scan them, fix any cap first, then rewrite the opening of the lowest scoring page so it answers its question in 40 to 60 words with a figure only you could give. Add a visible updated date and the business data the entity checks look for, and scan again to confirm the gaps have closed. Keep a note of each page's score beside its citations in Search Console, Bing Webmaster Tools and your question panel.

Repeat the scan every month, and move on to the next five pages once the first five score well. Over a quarter you will see which kinds of fixes are followed by citations on your own website, which is worth more than any general rule. The score points to the work; the engines' own reports show whether it paid off.