What Is Your Site’s AI Visibility Score?

For decades, SEO meant building a page a person could read and a search engine could rank. Answer engines changed the job. ChatGPT, Perplexity, and Gemini do not only hand someone a URL. They read a page, decide what it is about, and answer in their own words. GEO is the name many teams use for that shift: being cited, not only ranked.

What the AI Visibility Score Measures

The AI Visibility Score is a 0 to 100 reading of one public page. It asks whether a crawler is allowed in, whether the HTML says what the page is, and whether an AI agent can discover tools without guessing at buttons. The score is built from the HTML that was served, from robots.txt, from a request with Accept: text/markdown, and, when the page itself does not publish tools, from the site’s /.well-known/mcp files. It does not run the page in a browser, and it does not execute the page’s scripts.

What You Can Do with the Report

After the score finishes, two actions sit under the results.

• Download PDF Report: A branded report with a dark header, the page title and domain, the time it ran, section scores, and a resources panel under each section. The file name uses letters, numbers, and hyphens only, so dknewmedia.com becomes dknewmedia-com.
• Copy Markdown to Clipboard: Copies the same results as Markdown. Nothing is downloaded. The copy opens with an explanation of what the report is, how it was built, and what PASSED, IMPROVE, FAILED, and BONUS mean, so you can paste it into your own AI assistant with the context already in place.

If a direct fetch is blocked, the checker sends that URL through Bright Data and scores the HTML that comes back. When that HTML names a canonical address on the same host, the report treats that address as the end page.

How to Read a Result

Each check carries a status pill on the left so you can scan the page.

• PASSED: The check is in place.
• IMPROVE: The check is partial, or it is worth fixing.
• FAILED: The check is missing, or it is blocking discovery.
• BONUS: The item is not required. It can still help visibility when it is present. Blockquotes, author signals, <time>, <address>, <aside>, an Apple touch icon, a web app manifest, hreflang, and resource hints are bonus checks.

A single landmark such as <main>, <header>, <footer>, or <nav> says Found. Repeating items are named. Headings, sections, quotes, and the first item of each bulleted list (ul) or numbered list (ol) appear after Found: as a comma-separated list.

What Each Section Covers

The report groups checks so you can see which layer is helping you and which layer is in the way. Each section in the PDF opens with a short paragraph on why that layer matters for search and for AI visibility. On the page, platform sources sit as footnotes. In the PDF they sit in a gray panel with rounded corners.

Crawlability

A crawler that is blocked never reads the page. This section checks robots.txt, whether an XML sitemap is declared there, and whether the site root is open to major search crawlers and major AI crawlers. Search crawlers are listed first. AI crawlers follow. A crawler with its own User-agent group is judged by that group. Everyone else inherits User-agent: *.

Core Meta

The title and meta description are the first text a search result or an AI citation can show. A clear title tells Google and Bing what the page is about. A description gives them a sentence they can quote.

Social Graph

Open Graph and X card tags set the title, description, and image used when a URL is shared. Chat apps such as Slack and Discord read those tags too. Missing tags force a guess, and the preview is often empty.

HTML5 Structure

An LLM does not see a layout. It reads the document. Landmarks such as <header>, <main>, <nav>, and <footer> show where the real content starts. Sections, figures, and captions give the rest of the page a shape.

Content Structure

One H1, then H2 and H3 headings, give an answer engine clean chunks to lift. Bulleted lists (ul) and numbered lists (ol) are easier to cite than the same items buried in a paragraph. Author signals, including sameAs profile links, support experience and trust. Those author checks are bonus items: useful, not required for the rest of the score.

Structured Data

Schema.org markup, usually JSON-LD, hands a crawler facts it does not have to infer from prose. Google and Bing use markup that matches the visible page. Markup for something the page does not show is ignored.

Technical Signals

A canonical link tells a crawler which URL is the real page when several addresses serve the same content. This section also looks at the document itself: a valid page, a favicon, and optional signals such as hreflang and resource hints.

Links

Crawlers discover the rest of a site by following links whose addresses are in the HTML. A link that exists only inside a script click handler is much harder to follow.

Agent Interaction

WebMCP is how a page tells an AI agent which actions it can take, instead of making the agent guess. That check is 25 points. The checker looks for a discoverable signal: a webmcp-manifest or mcp-server-card link, a service link that points at a WebMCP document, a WebMCP script, document.modelContext or registerTool() in an inline script, or toolname and tooldescription on a form. If the HTML is silent, it also requests /.well-known/mcp/server-card.json, /.well-known/mcp.json, and /.well-known/mcp-server.

When none of those are present, the report says WebMCP was not detected. It does not claim the site has no tools. A tool that appears only after an ordinary external script runs is not discoverable, and the best practice is to put the signal in the HTML or in a well-known file. Read the longer explanation in WebMCP: Give AI Agents a Real Door Into Your Site, Store, or Platform.

Cloudflare Markdown for Agents is the other 25-point check. The report requests the page with Accept: text/markdown. It passes when the response comes back as Content-Type: text/markdown and the body is Markdown. A site can turn this on at the Cloudflare edge, or any server can answer that header the same way.

llms.txt is a bonus and does not change the score. You may see checklists that recommend llms.txt, a proposed Markdown file meant to summarize a site for language models. As of 2026, OpenAI, Google, and Anthropic have not adopted it as a requirement. A file does not hurt, but schema, HTML structure, and an explicit robots.txt policy do more for citations today.

Platform Sources, Not Third-Party Recaps

Each section ends with resources from the companies that publish the rules: Google, Bing, OpenAI, Anthropic, Cloudflare, and, for social cards, Meta and X. Each bullet is the source name as a link, then the description on the same line. These are the platform’s own pages for site owners, not a recap from another blog.

What This Score Does Not Measure

The score also stays on the page you submitted. It does not audit directories, knowledge graphs, or news mentions. Those off-site signals still matter for whether an answer engine trusts a claim. This report only tells you whether the page itself can be fetched and read.