AIO Copilot

AIO Score: What the Number Measures, and Why It Cannot Tell You Who Gets Cited

There is no standard AIO score. Ours is a 12-point checklist for one page, weighted by our own judgement, and no page score can say whether ChatGPT or Google's AI Overviews will cite you. Here is what each point measures, which ones Google cares about, and what to track instead.

Updated 3 November 202613 min read

What is an AIO score?

An AIO score is a readiness checklist for one web page: points awarded for things a parser can find in the HTML, such as structured data, a title of sensible length, a heading outline and links. No industry standard defines it. Our checker scores 12 page checks, while Semrush, Ahrefs and HubSpot attach similar-sounding scores to how often a brand turns up in AI answers. And no page score can predict whether ChatGPT, Perplexity or Google's AI Overviews will cite you, because that depends on whether the page ranks for the question, what else has been published on it, and how each assistant retrieves its sources.

Paste a page's source into our AIO readiness checker and it answers in well under a second with a letter grade, a number out of 100 and twelve findings. Every point comes from reading the text you pasted. Does a robots meta tag say noindex or nosnippet? Does the application/ld+json block parse, and what types does it declare? How many h2 tags are there, and how many of them are questions? The checker does not contact Google or any assistant, and it has no idea where the page ranks.

Google's page on AI features and your website supports that narrow reading. It says there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," beyond being indexed and eligible to show a snippet, and that there is "no special schema.org structured data that you need to add." A checklist can make a page easier to parse. The decision to use it happens where the checklist cannot see.

What follows covers what each of the twelve checks looks for, which of them matter for Google, and how to measure the outcome people hope the score predicts. One thing up front: the weights are ours, and Google publishes nothing like them. If the acronym itself is the puzzle, our explainer on AEO vs GEO vs SEO sorts out the overlapping names, and the guide to how Google's AI answers choose their sources covers the wider practice.

What do other tools mean by an AIO score or AI visibility score?

Search for the phrase and you find at least three different measurements wearing similar names. The large SEO suites score brand visibility in AI answers. Semrush's documentation defines its AI Visibility score as a benchmark from 0 to 100 showing "how often your brand appears in AI-generated answers compared to competitors." Ahrefs' Brand Radar counts mentions and citations across AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini and Copilot and turns them into an AI share of voice against competitors.

HubSpot's AEO Grader goes further from the page. It asks OpenAI's GPT-5.4 mini, Perplexity and Gemini about a brand and scores the answers out of 100, with up to 40 of those points for sentiment and 10 for share of voice. HubSpot describes it as a "free one-time diagnostic."

Those scores measure an outcome, by sampling what assistants say. A page checklist like ours measures inputs. The two can move in opposite directions in the same month without either being wrong, and a 70 on one tells you nothing about a 70 on another.

If you arrived from WordPress, there is a fourth meaning. The All in One SEO plugin, or AIOSEO, puts a TruSEO score out of 100 on every post, graded against a focus keyword and readability, and its documentation says following the recommendations does not guarantee rankings. It has nothing to do with AI answers.

What does the AIO readiness checker measure?

The checker on this site runs twelve checks on HTML you paste in. The points add up to 100. By category, eligibility is worth 14 points, structured data 16, meta tags 18, content structure 26, links 18 and image alt text 8. One rule overrides the sum: if the page carries a noindex or nosnippet directive, the score is capped at 20, because Google says such a page cannot be cited in its AI features. A score of 85 or more is an A, 70 a B, 50 a C and 30 a D; anything lower is an F. The table lists each check as the code scores it, as of November 2026.

CheckPointsWhat earns the points
Indexing and snippet controls14No noindex, none, nosnippet or max-snippet:0 in a robots or googlebot meta tag. A positive max-snippet costs 4, data-nosnippet elements 2 and a missing canonical tag 2.
Structured data (JSON-LD)10JSON-LD that parses and declares at least one @type, including types inside an @graph. Invalid JSON earns 3, no @type 4, no JSON-LD 0. FAQPage is listed but adds nothing.
Article markup6Article, BlogPosting, NewsArticle or a similar type with an author and datePublished or dateModified. Without them, 3.
Title tag8A title of 15 to 60 characters. Longer earns 6, shorter 4, missing 0.
Meta description670 to 160 characters. Longer earns 5, shorter 4, missing 0.
Open Graph tags4og:title, og:description and og:image. One or two of the three earn 2.
Headings10An H1 and at least two H2s. One H2 earns 7, H2s without an H1 earn 5, an H1 alone 4. Several H1s are not penalised.
Content length101,000 words of visible text or more, with scripts and styles left out. 500 earns 8, 250 earns 5, 100 earns 2.
Question headings6Three or more H2 to H4 headings with a question mark. One or two earn 4; questions only in the text earn 2.
Internal links10Five or more unique links to the same site, relative or absolute. The site's domain is read from the canonical tag or og:url. Three or four earn 7, one or two 4.
External links8Three or more unique links to other domains, not counting social profiles. One or two earn 5.
Image alt text8No image without an alt attribute. An empty alt="" counts as correct, since that is how decorative images should be marked. Up to half missing earns 5, more earns 2; a page with no images gets full marks.

Each check is there because it gives a machine reading the page something it can use without guessing: permission to index and quote it, a declared page type, a title, an outline, a route to related pages and sources, a description of each image. That is a case about parsing and basic quality, and the point values are a judgement we made about relative importance. Google has never published weights for anything like this. Any formula for AI citation, ours included, is somebody's opinion expressed as arithmetic, and the next section sorts the checks by how much Google says about each.

Does a higher AIO score help you rank or get cited?

One check is a requirement Google documents. A few line up with things Google says it uses. The rest are housekeeping. Sorting the checks by what Google's own documentation says about each signal shows which is which.

Our checker spreads 100 points over 12 checks. Only the 14 for indexing and snippets match a requirement Google documents.
Indexing and snippets14
Structured data10
Headings10
Content length10
Internal links10
Title tag8
External links8
Image alt text8
Article markup6
Meta description6
Question headings6
Open Graph tags4
A requirement Google documentsOur own judgement
Points per check in the AIO readiness checker, out of 100, November 2026
Indexing and snippets14
Structured data10
Headings10
Content length10
Internal links10
Title tag8
External links8
Image alt text8
Article markup6
Meta description6
Question headings6
Open Graph tags4
Source: the scoring code of the AIO readiness checker, November 2026. The points are our own weighting; Google publishes no weights for anything like this. The requirement is from Google's page on AI features and your website: a page must be indexed and eligible to show a snippet.

The one check Google documents as a requirement

Google's AI features page says that "to be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet," and names the controls that take a page out: nosnippet, data-nosnippet, max-snippet and noindex, all described in its robots meta tag documentation. A noindex left in a template after a site migration is a common way for a whole section to vanish, and it is invisible on the page itself. That is why this check carries the most points and why failing it caps the score.

The checks that line up with how Google ranks

The title is the first source Google lists for the title links it shows in results. The same page says there is no limit on how long a title element can be, but that the displayed title is truncated "as needed, typically to fit the device width." So our 60 character line is about display, and a title cut off mid-promise can cost clicks.

Internal links matter for a stated reason: Google's link documentation says it uses links "as a signal when determining the relevancy of pages" and "to find new pages to crawl." What counts is which pages a guide links to and with what words, which a link count cannot judge.

The content matters most, and it is exactly what the length check cannot assess. Google's guidance on helpful content says its ranking systems are built to "prioritize helpful, reliable information that's created to benefit people," and its SEO starter guide is blunt: "The length of the content alone doesn't matter for ranking purposes." Read the ten points for length as a prompt to ask whether the page has room to answer the question, never as a target.

The checks that help indirectly

Meta descriptions sometimes become the snippet under your result, but Google's snippet documentation says "snippets are primarily created from the page content itself," and that "there's no limit on how long a meta description can be." The 70 to 160 character band in our checker is a display convention we chose.

Article markup, according to Google's Article documentation, "can help Google understand more about the web page," and it has no required properties. Alt text is used, Google's image guidelines say, "along with computer vision algorithms and the contents of the page" to understand an image, and it is what a screen reader reads aloud. The link documentation adds that "using external links can help establish trustworthiness," with citing sources as the example. Each makes a page easier to understand or to trust, which is the most Google claims for them.

The checks that are housekeeping

FAQ schema used to carry ten points on this checker, for a feature Google no longer shows. In August 2023 Google said FAQ rich results "will only be shown for well-known, authoritative government and health websites." Its documentation updates log then recorded the end: "This feature will no longer appear in Google Search starting May 7, 2026." Existing markup can stay, since the 2023 post notes that "structured data that's not being used does not cause problems for Search," but adding it now buys nothing visible in Google, and since 22 October it buys nothing on our checker either.

Open Graph tags exist for social sharing. The protocol's own site says it "enables any web page to become a rich object in a social graph," which in practice means the card shown when someone pastes your link into a feed or a chat. They are worth having for that alone, which is why they carry only 4 points.

Heading order belongs in the same bin. The starter guide says headings in semantic order are "fantastic for screen readers" but that for Google Search the order does not matter, and that "there's also no magical, ideal amount of headings a given page should have." The checker no longer asks for an H3 or marks down a page with several H1s. Question-style headings have no Google documentation behind them either, though they are a sound writing habit: a heading phrased the way people ask tells the reader what the next paragraph answers. The same goes for how structured data is packaged. Google's structured data introduction says it "uses structured data that it finds on the web to understand the content of the page," and whether that arrives in one script tag or three changes nothing, so the checker now reads the types inside the JSON-LD rather than counting the tags.

What the checker cannot see

The larger blind spot is everything outside the pasted HTML. The checker does not know whether robots.txt blocks the page, whether an X-Robots-Tag header says noindex, whether Google has indexed it, or whether text injected by JavaScript ever reaches a crawler (paste the raw source and that text is missing from the analysis too, so paste the rendered HTML from the browser's inspector instead). It cannot see rankings or competing pages, or how an assistant chose its sources this morning.

Inside the HTML, it still reads literally. Five limits are worth knowing before you act on a finding:

  • It checks that JSON-LD parses and names a type, not that it carries the properties Google requires for a rich result. Google's Rich Results Test does that.
  • The word count and the internal link count include the menu and the footer.
  • It cannot tell whether a link's anchor text describes its target, or whether a meta description says anything a rival's could not.
  • Absolute links count as internal only when the page declares its own address in a canonical tag or og:url.
  • It sees where the canonical tag points but not the address of the page you pasted, so it cannot tell whether a canonical points somewhere else by mistake.

Earlier versions of this article listed six worse habits, and on 22 October 2026 we fixed the checker rather than keep explaining them. It used to count JSON-LD script blocks instead of reading them, treat absolute links to your own domain as external, count text inside scripts as words, mark down pages with more than one H1, give FAQPage markup ten points, and count decorative images marked alt="" as missing alt text, although the W3C's accessibility tutorial says decorative images should carry exactly that empty value so screen readers skip them. It now does the opposite of each, and checks for the indexing directives it used to ignore.

That leaves it useful as a lint check, good at catching a template that shipped with a noindex, a page with no title or JSON-LD that does not parse. It is no judge of whether the page deserves to be read.

How do you measure AI visibility instead?

The outcome people hope an AIO score predicts can be observed directly, at least in part, and two of the best sources cost nothing.

Search Console's generative AI report

Google counts traffic from AI features inside the ordinary Performance report, under the Web search type, according to its AI features page. On 3 June 2026 it added a dedicated view: the Search Generative AI performance reports, one for Search and one for Discover. A note on that announcement says that as of 31 August 2026 they have rolled out to all websites worldwide.

The Search report covers AI Overviews and AI Mode. Its one metric is impressions, which the Search Console help page defines as "how many times links to your site were shown to a user in a generative AI feature on Google Search," broken down by page, country, device and date. Google describes impressions only, with no click figure and no query dimension, and the report may not appear at all for a site with too few AI feature impressions. Even so, it answers the first real question: which pages show up in Google's AI answers and which never do. Compare the two lists before you change a single tag, and when you do change something, measure the edited pages against pages nobody touched, as our piece on proving an SEO change worked explains.

Bing Webmaster Tools AI Performance

Microsoft launched AI Performance in Bing Webmaster Tools as a public preview on 10 February 2026. Covering Microsoft Copilot, AI summaries in Bing and select partner integrations, it reports total citations, the average number of your pages cited each day, and grounding queries, the phrases the AI used to retrieve your content. Microsoft notes that the grounding query data "represents a sample of overall citation activity," and that page-level counts reflect "how often pages are cited, not page importance, ranking, or placement." Clicks are not among the metrics it lists.

Sampling the assistants that report nothing

For ChatGPT, Perplexity, Claude and the Gemini app, neither report covers you. Referral clicks show up in your analytics, but a mention without a link leaves no trace there. So you ask: write 20 to 50 questions your buyers actually put to assistants, run each several times per assistant, since answers vary from run to run, and count how often your brand and each rival is named.

Then put a margin of error on the share. Say 200 answers name you 60 times, a 30% share. At 95% confidence the margin is 1.96 times the square root of (0.3 times 0.7, divided by 200), about 6.4 points, so the true share sits somewhere between roughly 24% and 36%. If next month's sample says 34%, nothing has been shown to change. Our explainer on measuring share of model walks through the method, and the comparison of AEO tools checks how ten trackers sample answers and whether they show the uncertainty. AIO Copilot runs this sampling for a client against its rivals and prints the margin of error next to every share; the AI visibility report shows how.

Example: scoring a sample Harbour & Finch page

Harbour & Finch LLP is a Toronto employment law firm that exists only as our sample client; the page and every figure below are invented to show the method. Its guide to severance pay in Ontario runs about 1,200 words, sits on a template whose menu and footer use absolute addresses, and carries one JSON-LD block holding Organization, WebPage, Article and BreadcrumbList types. We built the page and ran it through the checker's scoring code twice. The version retired on 22 October scored it 64, a C. The current one scores it 88, an A.

CheckScoreWhat the checker saw
Indexing and snippet controls14 / 14No noindex or nosnippet; canonical tag present
Structured data (JSON-LD)10 / 10One valid block declaring four types
Article markup6 / 6Article with an author and datePublished
Title tag6 / 891 characters, long enough to be cut off in results
Meta description6 / 690 characters, generic
Open Graph tags4 / 4og:title, og:description and og:image set
Headings10 / 10One H1, seven H2s, no H3
Content length10 / 101,237 words of visible text, menu and footer included
Question headings4 / 6Two question headings
Internal links10 / 10Fifteen, most of them the menu and footer, now counted as internal
External links0 / 8None
Image alt text8 / 8Three described, two decorative dividers marked empty

The 24 extra points came from fixing the checker, not the page. The menu's absolute links now count as internal, the single block of JSON-LD gets full marks for the four types inside it, the empty alt text on the dividers counts as correct, and the old ten points for FAQ markup no longer exist to be missed.

What the checker flags now is what an editor would flag. The guide paraphrases Ontario's Employment Standards Act throughout and never links to the government's own guide to it, so a reader checking a severance figure has nowhere to go. The old checker gave this page six of six for external links, because it mistook the firm's own menu for outside sources; the new one gives it zero. The title, at 91 characters, loses its end in results. Link the government guide and rewrite the title so the part that matters survives, for instance "Severance Pay in Ontario: How Much You Are Owed" at 47 characters, and the page scores 95.

The sample page gained 24 points when we fixed the checker, and 7 more from two edits to the page.
Retired checker64
Current checker88
After two page edits95
After fixing the pageSame page, before the edits
AIO readiness score of a sample Harbour and Finch page, out of 100: retired checker, current checker, and current checker after two edits (worked example)
Retired checker64
Current checker88
After two page edits95
Worked example on sample data: a severance guide we built for Harbour & Finch LLP, a sample Toronto law firm, run through the checker's scoring code. The retired version was replaced on 22 October 2026. The two edits are one link to the Ontario government's guide (external links, 0 to 5 points) and a 47-character title (title tag, 6 to 8 points).

Two jobs are still beyond it. Read the two links in the body: they point to the firm's wrongful dismissal and severance review pages with anchors that say so, which is right, but the checker would score them the same if they said "click here". And the description scores full marks at 90 characters while saying nothing a rival firm's could not. The housekeeping that used to earn 20 points earns nothing now. We added FAQPage markup, split the JSON-LD into four blocks, wrote alt text for the dividers and added an H3 for its own sake, and the page still scored 88.

Priority still has to come from elsewhere. If Search Console shows the guide earning impressions for severance queries at position 9 with a weak click-through rate, the title is the first job. If the generative AI report shows the guide never appears in AI features, and asking the question yourself turns up a rival firm's guide, the useful question is what that guide answers that this one does not. No tag fixes that.

What is a good AIO score?

On our checker, a well-built page lands in the 80s, as the Harbour & Finch guide does, and the last points come from sourcing its facts and tightening its title. Look at the eligibility line first: a noindex or nosnippet left in a template caps the score at 20 and keeps every page on that template out of AI Overviews. A score below 50 usually means something structural is missing, such as no title, no headings or a near-empty page (often the raw source of a page that renders with JavaScript), and those are worth fixing for their own sake. If you want the markup side done properly rather than to a score, our schema markup guide covers the types that still produce rich results.

Past that, stop working on the number. A readiness score is most useful the first time you run it on a page, when it catches what a template got wrong. After that, the figures worth watching are the ones Google and Bing report about your pages and the ones a sample of real answers gives you, margin of error attached, and none of them goes up because someone added a script tag.

Questions

What is an AIO score?

An AIO score is a checklist grade for a single web page. There is no industry standard: the AIO readiness checker on aiocopilot.com runs 12 checks on pasted HTML worth 100 points in total, while tools such as Semrush and HubSpot use similar-sounding scores for how often AI answers mention a brand. A page score says whether a page is well formed for machines to parse. It cannot say whether an AI assistant will cite the page.

How is the AIO readiness checker score calculated?

The AIO readiness checker awards points for 12 checks on the HTML you paste: indexing and snippet controls (14), valid JSON-LD (10), Article markup with an author and date (6), title tag (8), meta description (6), Open Graph tags (4), headings (10), content length (10), question headings (6), internal links (10), external links (8) and image alt text (8). The points add up to 100. A noindex or nosnippet directive caps the score at 20, because Google says such a page cannot be cited in AI Overviews. The weights are our own judgement; Google publishes nothing similar.

What is a good AIO score?

On our checker, 85 and above is an A, 70 to 84 a B, 50 to 69 a C, 30 to 49 a D and below 30 an F. A well-built page usually lands in the 80s, and the points above that come from sourcing facts and tightening the title. A page blocked by noindex or nosnippet is capped at 20 whatever else it gets right. A score below 50 usually means something structural is missing, like a title, headings or most of the text, which often happens when a page renders with JavaScript and the raw source is pasted.

Does FAQ schema still help in 2026?

Not for rich results. Google limited FAQ rich results to well-known government and health sites in August 2023, and its documentation says the feature stopped appearing in Google Search for all sites from 7 May 2026. Google also says structured data that is not being used does not cause problems, so existing FAQPage markup can stay. Google's guidance on AI Overviews and AI Mode says no special schema.org markup is needed to appear in them.

Can an AIO score predict whether ChatGPT or AI Overviews will cite my page?

No. Whether an assistant cites a page depends on whether the page ranks or is retrieved for the question, what else has been published on the same question, and how that assistant selects sources. None of that is visible in a page's HTML. To measure citations, use Search Console's generative AI report for Google, Bing Webmaster Tools' AI Performance report for Copilot and Bing, and a repeated sample of questions for other assistants.

How can I see whether my pages appear in Google AI Overviews?

Search Console has a Search generative AI performance report, introduced on 3 June 2026 and rolled out to all websites worldwide as of 31 August 2026, according to Google. It shows how often links to your site were shown in AI Overviews and AI Mode, by page, country, device and date. It reports impressions only. It may not appear for sites with very few AI feature impressions.

New studies, by email

Our own studies, and what each new release changes for SEO, AEO and Google Ads.

Or get your price for the app