Check your AI SEO Score

What an AI SEO score is

An AI SEO score is a readiness check for a single URL: it tells you whether the page is technically open to AI crawlers and whether its content is shaped in a way that retrieval systems can extract, summarize and attribute. It does not measure how often ChatGPT, Perplexity, Gemini or Claude actually mention your brand. Those are different questions, and it helps to keep them apart.

Think of the score as a list of preconditions. A page that blocks GPTBot, hides its main text behind client-side rendering, has no clear heading outline and gives no hint of who wrote it is unlikely to be used as a source, however good the underlying expertise is. A page that clears those hurdles is not guaranteed a citation, but it is no longer disqualifying itself.

Our checker scores five groups: AI crawler access, structure, content quality, entity and trust, and technical. Each group is built from individual checks you can verify yourself in the page source, which means every point in the score maps to something concrete you can change.

How AI engines find and use pages

AI assistants reach web content in two broad ways: through training data collected by crawlers in the past, and through live retrieval when a user asks a question. The second path matters most for visibility, because that is where a specific page gets fetched, quoted and linked.

Training crawlers

Crawlers such as GPTBot (OpenAI), ClaudeBot (Anthropic) and CCBot (Common Crawl) collect pages that may be used to train future models. Blocking them keeps your content out of future training sets. It does not, by itself, remove you from AI search answers, because the major providers run separate agents for search and for user-initiated fetches.

Search and retrieval agents

OAI-SearchBot indexes pages for ChatGPT search, PerplexityBot builds Perplexity's index, and Claude-SearchBot supports search in Claude. User-triggered agents like ChatGPT-User, Claude-User and Perplexity-User fetch a page when a person asks the assistant to look at it or when the assistant browses during an answer. Each of these tokens can be allowed or disallowed separately in robots.txt, which is why a single blanket rule is rarely the right choice.

Google is a special case

Gemini and Google's AI features draw on Google's existing index, which Googlebot builds. The Google-Extended token only controls whether content Googlebot has crawled may be used to train Gemini models and to ground answers in Gemini apps and Vertex AI. Disallowing Google-Extended does not affect Google Search crawling, indexing or ranking, and it does not keep a page out of AI Overviews, which are part of Search.

Traditional SEO vs AI visibility

Most of what makes a page rank also makes it usable by AI systems, so AI visibility is an extension of SEO rather than a replacement. The differences are in emphasis: retrieval systems reward pages that are easy to split into self-contained passages, and they care about whether a claim can be attributed to a clear source.

The practical conclusion: do not split your team or your content into an SEO track and an AI track. Fix the fundamentals once, then add the few AI-specific layers, mainly bot access rules and passage-level clarity.

Crawler access and robots.txt

Crawler access is the first group because it is binary: if a bot is disallowed, nothing else on the page matters to that bot. The most common problem we see is not a deliberate block but an inherited one, such as a security plugin, a CDN bot setting or a years-old Disallow: / under User-agent: * that nobody revisited.

Decide per agent. Many publishers allow search and user-triggered agents, which can send visitors and citations, while blocking training crawlers they have no agreement with. Others allow everything to maximize the chance of being used. Both are defensible; an accidental policy is not. Here is an example of an explicit, mixed policy:

# Allow AI search and user-triggered agents
User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: PerplexityBot
User-agent: Claude-SearchBot
User-agent: Claude-User
Allow: /

# Opt out of model training
User-agent: GPTBot
User-agent: ClaudeBot
User-agent: CCBot
User-agent: Google-Extended
Disallow: /

# Everyone else, including Googlebot and Bingbot
User-agent: *
Disallow: /wp-admin/
Allow: /wp-admin/admin-ajax.php

Sitemap: https://www.example.com/sitemap.xml

Two details trip people up. First, a crawler follows only the most specific group that names it, so a bot listed in its own group ignores the rules under User-agent: *. Second, robots.txt is a request, not an access control. Reputable operators honor it; to actually stop traffic you need server or firewall rules.

llms.txt

llms.txt is a proposed standard, published in 2024, for a Markdown file at your site root that summarizes the site and links to its most useful pages for language models. It is not an official web standard, and no major AI provider has committed to reading it the way search engines read robots.txt and sitemaps. It is cheap to add and harmless, so our score gives it modest credit, but do not expect it to change visibility on its own.

noindex and noai

A noindex meta robots tag or X-Robots-Tag header removes a page from search indexes, and search-based AI features inherit that exclusion. The noai and noimageai values are non-standard; some platforms respect them, many crawlers ignore them. The checker flags both so you know they are there.

To test every AI bot against both your homepage and a specific path, use the AI Crawl checker, which reports the rule that applies to each user agent and suggests a robots.txt block.

Structure and content signals

Retrieval systems break pages into chunks and pick the chunks that best match a question, so structure decides whether your best sentence survives extraction. A passage that starts with "As mentioned above" loses its meaning once it is lifted out of the page.

Write answer-first

Open each section with one or two sentences that directly answer the question in the heading, then add context, caveats and examples. This is the same inverted pyramid news writers use, and it serves skimming humans as well as it serves models.

Use a real heading outline

One H1 that names the topic, H2s for the main questions, H3s for sub-points. Phrase headings the way people ask: "How long does a lawn mower carburetor rebuild take?" is more matchable than "Timing". Do not skip levels for visual styling; use CSS for that.

Prefer lists and tables for structured facts

Specifications, comparisons, steps and prices belong in HTML lists and tables, not in images or long paragraphs. Tables in particular give a model clean rows and columns it can quote without guessing.

Content quality checks

The checker looks at word count, a readability score and lexical variety. None of these is a target in itself: a short page can be excellent when the question is simple, and repetitive wording often signals thin, templated text. It also checks for question-style headings, an FAQ block and visible published or updated dates, because assistants that answer time-sensitive questions favor content that shows when it was last reviewed. Only change the date when you actually update the page.

Entity and trust signals

Entity and trust signals tell a machine who is behind the page and whether the details are consistent. When an assistant decides which of several similar sources to cite, a page with a named author, a known publisher and matching structured data is the easier choice to justify.

• Schema markup: Organization on the site, Article or BlogPosting on editorial pages, Product or Service where relevant, and FAQPage when the page truly contains questions and answers. Keep it in JSON-LD and make sure it matches what is visible on the page.
• Authorship: a visible byline linked to an author page that explains the person's experience with the topic.
• About and contact links: plain evidence that a real organization stands behind the content.
• OpenGraph tags: a consistent title, description and image when the page is shared or previewed.
• Canonical URL and HTTPS: one preferred address for the content, served securely, so signals are not split across duplicates.

Note that Google limited FAQ rich results to a small set of authoritative government and health sites in 2023. FAQPage markup still describes your content accurately for any parser, but do not add it expecting the old expandable search snippet.

Technical signals

Technical checks cover what a crawler experiences when it requests the page. AI fetchers often work with tight timeouts, and a user-triggered agent answering a live question will not wait for a slow server.

• Response time: aim for a fast server response; slow time to first byte usually points to hosting, caching or database issues.
• Page size and compression: serve HTML with gzip or Brotli and avoid shipping megabytes of inline scripts before the content.
• Viewport and lang: a mobile viewport tag and a correct lang attribute help parsers classify the page.
• Image alt text: descriptive alt attributes make images understandable to systems that do not process pixels.

One issue worth checking manually: if your main content is injected by JavaScript after load, some AI crawlers will see an empty shell because not all of them render scripts. Server-side rendering or static generation removes that risk.

Steps to raise your AI SEO score

Work in the order below. It follows the dependency chain: access first, then parsing, then trust, then polish.

1. Confirm crawler access

   Read your live robots.txt, decide a policy for each AI agent, and remove unintended blocks, including CDN or firewall bot rules that return 403 errors to AI user agents.

2. Remove accidental exclusions

   Search templates for leftover noindex tags and check response headers for X-Robots-Tag values, especially on pages migrated from a staging site.

3. Rewrite the intro answer-first

   Make the first paragraph a complete, quotable answer to the page's main question in two to three sentences.

4. Fix the heading outline

   One H1, then question-style H2s and supporting H3s, with each section able to stand alone.

5. Add identity and schema

   Add an author byline, Organization and Article markup, and links to your about and contact pages.

6. Tighten the technical basics

   Enable compression, trim page weight, set the lang attribute and fill in missing alt text.

7. Recheck and share

   Run the checker again, confirm the fixes registered, and share the saved report with whoever owns the remaining items.

For site-wide issues such as security headers, sitemaps and broken links, follow up with the full SEO audit. To see which sources AI engines actually cite for your topic, use the AI Citation Analyzer.

What the score cannot tell you

The score measures one page against checks we can verify from its HTML, headers and robots.txt. It cannot see the private ranking logic of any AI provider, so it cannot promise a mention, and it does not measure how well known your brand is across the web, which strongly influences what models say.

It also does not judge whether your content is correct or original. A perfectly structured page that repeats what fifty other pages say gives an assistant no reason to cite it. Use the score to remove technical and structural obstacles, then spend the time you save on first-hand data, specific examples and expertise that other pages lack.

Finally, treat the number as a trend line. Recheck after theme updates, plugin changes, CDN configuration changes and robots.txt edits, since those are the moments when a page that scored 90 last month quietly drops to 40.