AI Visibility Score: Methodology and 130+ Check Items

The AI Visibility Score is a 0-to-100 score, created by Madbotz, that measures how visible your site is to generative AI. This post opens up the methodology behind the number.

We introduced it when we presented the Searchability framework and the AI Visibility Score as the thermometer of your presence with the models. Here we get into the detail: how it is calculated, what the 8 categories measure, and how to read your score to act.

If you already ran your site through Visibility and saw a number, this is the guide to understanding what it means and what to do with it — without it being a black box.

Why a Score — and Why a Transparent One

Whoever owns AI at a company cannot report "we improved AI visibility" without a number. A 0-to-100 score gives you a single auditable figure you can take to the board and track over time.

The transparency is a deliberate decision. Visibility to AI matters because more and more people are replacing traditional search with answers from models: according to Gartner, traditional search volume will drop 25% by 2026 as users shift to chatbots and AI agents. If your customers ask a model instead of a search engine, you need to know whether the model includes you — and a diagnosis that makes it measurable.

A score is not a medal. It is a prioritizable diagnosis — it tells you where the gap is, not how good you are.

How the AI Visibility Score Is Calculated

The mechanics are simple on purpose. Every check is worth the same: one point, with no severity weighting. That decision keeps the score auditable — nobody has to trust a secret weighting: a check is either met or it is not.

Your score is the proportion of applicable checks that your page passes. The key word is "applicable": many checks apply conditionally depending on the page type.

A single-language site is not penalized for lacking hreflang. A landing page without an article is not penalized for lacking Article schema. The engine does not punish you for what does not apply to your page.

Each check rests on a public, verifiable standard, not an opinion. The robots.txt check rests on RFC 9309, the IETF (Internet Engineering Task Force) standard for the Robots Exclusion Protocol; the structured data check, on the schema.org vocabulary. That is why the score is reproducible: anyone can review the same standard.

And it is per page, not per site. That is why you should analyze your key pages — not just the home.

The 8 Categories of the AI Visibility Score

The score groups the 130+ check items into eight categories. Each one maps to a signal that models use to discover, understand, and trust your content.

Table 1 — the 8 categories of the AI Visibility Score. The last column, "Where to go deeper", links to the spine post that opens each category.

Indexability, Context, and Content — Discover and Understand

Indexability measures whether AI crawlers can get in: per-bot robots.txt, noindex, canonical, sitemap, and the llms.txt check. It is the foundation, because if bots cannot get in there is nothing to cite. It rests on RFC 9309 and on each crawler's official docs, like OpenAI's GPTBot and Anthropic's ClaudeBot; go deeper in crawlability for AI bots and in the llms.txt cargo cult.

Context measures the structured data that tells the model what each thing is: Organization, Article, FAQPage, BreadcrumbList, Product. According to Google, structured data helps engines understand the content of the page; go deeper in schema.org JSON-LD for AI.

Content measures the readable-quality signals: single H1, heading hierarchy, adequate length, alt text, and declared language. A clean structure lets the model extract passages with confidence, and according to Google, content should be written for people first; go deeper in citation-worthy content for generative AI.

Trust, Performance, and Accessibility — The Technical Base

Trust measures the site's basic trust signals: HTTPS, valid SSL, no mixed content, privacy and terms pages, visible contact. Google has treated HTTPS as a ranking signal since 2014, and its quality guidelines describe trust as the most important pillar of E-E-A-T (experience, expertise, authoritativeness, and trust).

Performance measures the mobile loading experience with Core Web Vitals: LCP (Largest Contentful Paint), which measures when the largest element appears, and CLS (Cumulative Layout Shift), which measures visual stability. According to web.dev, a good experience requires an LCP of 2.5 seconds or less and a CLS of 0.1 or less. A slow page is crawled worse and frustrates the person.

Accessibility measures whether a machine can read the structure: alt text, form labels, ARIA roles, landmarks, and skip links. The accessibility tree that WCAG 2.2 (Web Content Accessibility Guidelines), from the W3C, defines is, in practice, the same tree an AI walks to understand your page.

Authority and Freshness — Authority and Currency

Authority measures the external authority signals: author bio, Person schema, social profiles, citations, and publisher data. Google added Experience to E-E-A-T in 2022, and external mentions are what correlate most with presence in AI; go deeper in authority for answer engines and in brand mention engineering.

Freshness measures how current your content looks: sitemap lastmod, dateModified, current copyright year, and RSS. The sitemaps.org protocol defines lastmod to signal the last significant modification, and clarifies that updating only the copyright year does not count as a relevant change.

How to Read Your Score and Act

Start with the weakest category. That is where the highest return on investment is — the average hides the bottleneck, and AI fails in your worst category, not your average.

Read the range by color. A green / yellow / red traffic light tells you the severity at a glance, the same pattern Google uses for Core Web Vitals: according to web.dev, each metric is classified as "good", "needs improvement", and "poor".

The goal is not 100%. Some checks do not apply to your page, so the realistic ceiling is closing the gaps that do apply. And since the score is per page, review the ones that matter most to the business (blog posts, product pages, news) — not just the home.

Internal Score vs External Monitoring

The AI Visibility Score measures your site's readiness — the internal part, the part you control. Multi-LLM monitoring measures your real presence in the models' answers — the external part, the part you do not fully control.

A high score is a necessary condition, not a sufficient one. According to Profound, which analyzed 680 million citations, the sources each model cites diverge so much that "a one-size-fits-all approach to AI visibility cannot succeed." Your site can be flawless and still not be the source the model picks.

They are two layers, not one. This post covers the first; monitoring covers the second.

Anti-Patterns When Optimizing the Score

Six ways to misread the score that we see operating on real sites.

• Chasing 100% — some checks do not apply to your page type; closing the applicable gaps is the goal, not the round number.
• Optimizing the number without the underlying intent — slipping in a fake FAQPage to tick the check does not make you citable. According to Google, content should be for people, not to game a system.
• Treating one page's score as the site's — it is per page; your home can score 90 and your product pages 40.
• Ignoring the weakest category — the average hides the bottleneck; AI fails where you are worst, not at your average.
• Confusing score with presence — the score is internal readiness; real citation is external. You need both layers.
• Analyzing only once — content and standards change; the score is a trend signal, not a one-time certificate.

How Madbotz Applies This to Its Own Blog

Honesty before hype. Madbotz runs its own blog against the AI Visibility Score — the same engine that evaluates the 130+ check items across the 8 categories you just read.

We do not say it to show off a number. The Searchability framework we document is only worth it if we apply it to ourselves first. Each spine post — from crawlability to mentions — maps to a category of the score.

Our MadbotzVisibilityBot walks your page the way an AI crawler would, and evaluates it against those eight categories. Run your site, start with your key pages, and see your own score in under 60 seconds.

Frequently Asked Questions

What is a good AI Visibility Score?

It depends on your page, but as a quick reference: the green range means you pass most of the applicable checks, yellow means you have clear gaps, and red means foundations are missing. 100% is not the goal, because some checks do not apply to your page type. Aim to close the gaps that do apply.

Why is the score per page and not per site?

Because each page serves a different purpose and is evaluated against the checks that apply to it. Your home, an article, and a landing page have different profiles. That is why you should analyze the pages that matter most to the business, instead of assuming the home score represents the whole site.

Why do some checks not count toward my score?

Many checks are conditional. A single-language site does not need hreflang; a landing page without an article does not need Article schema. The engine only counts the checks applicable to that page, so it does not penalize you for what does not fit your page type.

Does the score guarantee that AI will cite me?

No. The score measures your site's internal readiness: a necessary condition, not a sufficient one. Whether you get cited also depends on your real external presence, which you measure with multi-LLM monitoring. They are two complementary layers, and this score covers only the first.

Closing

Three takeaways:

• The AI Visibility Score is a prioritizable diagnosis, not a medal — start with your weakest category.
• It is per page and per applicable checks — 100% is not the goal; closing the gaps that do apply is.
• The score measures internal readiness; real citation is measured externally. You need both layers.