AI Visibility Audit: A Practical Checklist (with Real Data)

Traditional SEO audits check whether Google can crawl, index, and rank your pages. An AI visibility audit checks whether AI answer engines — ChatGPT, Perplexity, Claude, and Google AI Overviews — can find, understand, and cite your content when users ask relevant questions.

The two overlap, but they are not identical. A site can rank well in Google and score poorly on AI citation readiness. This checklist covers the five dimensions of AI visibility, what a passing state looks like, and how to fix common failures. At the end, you will know exactly where your site stands.

---

What is an AI visibility audit, and why does it matter?

An AI visibility audit evaluates how well your site is structured to be understood and cited by generative AI systems. These systems don't rank pages the way search engines do — they extract answers, attribute them to sources, and decide which sources are worth citing based on content structure, entity clarity, schema markup, and crawler access.

As of 2026, ChatGPT (via Bing and GPTBot), Perplexity (PerplexityBot), Claude (ClaudeBot), and Google AI Overviews collectively generate a material share of query responses that previously surfaced as traditional blue links. Getting cited in an AI answer is the new "ranking in position 1."

The DeepSEOAnalysis GEO score quantifies this across five dimensions. Across hundreds of audits in our dataset:

• The average AI visibility score is 64/100 — most sites have significant room to improve
• Over 70% of sites have no llms.txt file
• A surprisingly high share of robots.txt files block at least one major AI crawler
• Sites with FAQPage or HowTo schema score 15–20 points higher on AI citation readiness on average

This checklist follows the same five dimensions.

---

1. Are AI crawlers actually allowed to access your site?

The check: Are GPTBot (OpenAI/ChatGPT), ClaudeBot (Anthropic/Claude), and PerplexityBot explicitly allowed in your robots.txt?

Why it matters: An AI system cannot cite content it cannot crawl. Many sites added Disallow: / catch-alls during the AI training opt-out wave of 2023–2024 and forgot to re-allow inference-time crawlers. These bots crawl to answer user questions, not to train models — but a blanket disallow blocks both.

How to check: Fetch your /robots.txt and look for these user-agents. Use the free robots.txt tester to check how each bot is treated.

Passing state:

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

Failing state (common):

User-agent: *
Disallow: /

# No explicit Allow for AI crawlers above the catch-all

Fix: Add explicit Allow: / rules for each AI user-agent before any catch-all Disallow. Check the Crawl-delay as well — very low values can cause crawlers to give up.

---

2. Does your site have an llms.txt file?

The check: Does /llms.txt exist and return a 200 status?

Why it matters: llms.txt is a Markdown-style context file placed at your domain root. It gives AI crawlers a fast, citable summary of what your site is, who it serves, and which pages are authoritative. Without it, an AI system must infer your site's identity entirely from crawled content — a slower and less accurate process.

How to check: Fetch https://yourdomain.com/llms.txt. A 200 response with structured Markdown content passes; a 404 or empty file fails.

Passing state:

# Your Site Name

> One-sentence factual description of what the site does.

## Key pages

- [Homepage](URL): What it offers
- [Documentation](URL): What it covers
- [Blog](URL): What you publish

Fix: Use the free llms.txt generator to create a properly formatted file. Publish it at /llms.txt. See llms.txt: What It Is, Who Respects It for a complete walkthrough.

---

3. Does your site have FAQPage or HowTo schema on at least one page?

The check: Is there valid FAQPage or HowTo JSON-LD schema on at least one public page?

Why it matters: FAQPage and HowTo are the two schema types most directly used by AI systems to extract and cite structured question-answer pairs. When your FAQ is marked up with schema, an AI system doesn't have to infer that "What is X?" in a paragraph is a question-answer pair — the markup makes it explicit and machine-readable.

In our audit data, sites with FAQPage schema score significantly higher on AI citation readiness than sites without, even controlling for other factors.

How to check: View the page source and search for "@type": "FAQPage" or "@type": "HowTo". Or run a free audit and check the Structured Data section.

Example (FAQPage):

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "What does DeepSEOAnalysis check?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "DeepSEOAnalysis runs 100+ checks across technical SEO, Core Web Vitals, on-page SEO, structured data, internal linking, and AI visibility."
      }
    }
  ]
}

Fix: Add FAQPage schema to any page that already has a FAQ section — or add a FAQ section to your most important landing pages. Use the free schema generator to produce valid JSON-LD without hand-coding it.

---

4. Are at least 20% of your H2/H3 headings phrased as questions?

The check: On each page with ≥3 headings, are at least 20% of H2/H3 headings phrased as questions (ending in "?" or starting with "How", "What", "Why", "Which", "When", "Can", "Does", "Is")?

Why it matters: AI answer engines extract answers to user questions. A heading phrased as a question — "How does the audit work?" — directly signals to AI systems that the following content answers that question. Declarative headings ("How it works") are less explicit. AI systems can still extract content from declarative headings, but question-form headings make the mapping clearer and more citable.

How to check: Scan your most important pages. Count H2/H3 headings. Count how many start with a question word or end in "?". Divide.

Example conversion:

• "Our approach to performance" → "How does the performance audit work?"
• "Supported platforms" → "Which platforms does DeepSEOAnalysis support?"
• "Pricing overview" → "What does each plan include?"

Target: ≥20% of H2/H3s per page. For a page with 10 headings, at least 2 should be questions. For a page with 5 headings, 1 question heading is the minimum.

Fix: Rewrite declarative headings to question form where natural. Do not force-convert every heading — pages where 20–40% of headings are questions read naturally; pages where 80% are questions feel like a FAQ, which is only appropriate in an actual FAQ section.

---

5. Is your content structured in short, citable chunks?

The check: Is the average number of words between headings under 400?

Why it matters: AI systems extract answers at the section level. When a section runs 600+ words, the system must make judgment calls about which paragraph answers which question. When sections are short (150–350 words) and clearly labeled, extraction is more reliable and attribution is more accurate.

How to check: Count the words between consecutive H2/H3 headings on your key pages. Divide total content words by the number of headings to get an average.

Benchmark: In our audit data, pages that fail the chunkability check average ~620 words between headings. Passing pages average ~240 words between headings.

Fix:

• Add more subheadings to break up long sections. A 600-word section can usually become two 250-word sections with clearer topic focus.
• Prioritize pages that rank for informational queries — these are the pages AI systems are most likely to pull from.
• Use subheadings that label what each section answers, not just what it covers.

---

How to run a complete AI visibility audit

Manual checking covers the basics. For a complete automated AI visibility audit — with scores, specific failed checks, and copy-paste fixes — run a free DeepSEOAnalysis audit. The report includes:

• A 0–100 GEO score (separate from the main SEO score)
• Per-check pass/fail results for all five dimensions above
• Prioritized fix list sorted by impact
• Your robots.txt evaluation for specific AI crawlers
• llms.txt detection and structure check
• Schema coverage across all crawled pages

The AI visibility methodology page documents the exact pass rules for each check.

---

FAQ

How is the AI visibility score different from an SEO score?

The SEO score measures traditional search ranking signals: crawlability, on-page optimization, Core Web Vitals, structured data, and internal linking. The AI visibility (GEO) score measures citation readiness for AI answer engines. A site can score 95 on SEO and 60 on GEO — they optimize for different systems with different extraction patterns.

Which AI systems does this audit cover?

The audit checks specifically for GPTBot (ChatGPT/OpenAI), ClaudeBot (Anthropic/Claude), and PerplexityBot. Google AI Overviews uses Googlebot, which is already covered by the Technical SEO pillar in the main audit.

How quickly can I improve my AI visibility score?

Adding llms.txt and fixing a robots.txt bot block are same-day fixes. Adding FAQPage schema to existing content takes a few hours. Question-heading rewrites across key pages can be done in a day. In our experience, sites that address the three main checks (crawler access + llms.txt + FAQPage schema) typically see their GEO score jump 15–25 points.

Does a higher AI visibility score mean more AI citations?

Not directly — it means your site is better structured to be cited. Whether a specific AI system cites you also depends on the quality and uniqueness of your content, the competitiveness of the query, and how recently the AI's knowledge was updated. But fixing structural barriers is the prerequisite: you cannot be cited if you are blocked or structurally illegible to the AI.

Is AI visibility important for every site?

It matters most for sites that compete for informational queries ("how to...", "what is...", "best X for Y"). Product landing pages and conversion-focused pages benefit less — users asking AI systems about those queries often want recommendations, not the vendor's own page. Audit content-heavy pages first.