An AI SEO audit step by step: the five layers you work through
An AI SEO audit checks how ready your site is to be cited and mentioned in AI answers. It works through five layers in a logical order: whether AI search tools mention you at all, whether they can get their crawlers to your site, whether machines understand your content, whether your copy is written so it can be cited, and how to measure all of it over time. This guide is a DIY checklist — a basic pass on a small or mid-sized site usually takes a single afternoon, mostly with free tools.
Why handle this separately from classic SEO? Because Google rankings and citations in AI answers don’t necessarily correlate. According to 2025 analyses, a significant share of the pages frequently cited in ChatGPT have weak to nonexistent organic visibility in Google. A solid classic SEO audit is the foundation, but it won’t cover the AI layer.
Why AI visibility matters right now
The numbers in this article come from public analyses of AI search published in 2025–2026, and they vary widely by market, language, and industry — treat them as an order-of-magnitude trend, not an exact value for your vertical. The trend itself is consistent: a share of Google searches now return an AI Overview, and traffic from AI search is growing fast according to the available analyses. Even so, a large share of companies still aren’t visible in AI answers — AI systems often don’t cite them because their sites lack content that is clear, structured, and verifiable enough.
The way people search is changing too: instead of ten blue links, they get one composed answer with a handful of cited sources. If your brand isn’t among those sources, it’s worth working out which sources AI systems prefer and why. What each discipline actually covers is broken down in the SEO vs GEO vs AEO guide; this audit is the practical check on whether your site measures up.
Step 1: Map whether AI mentions you at all
Start with where you actually stand. Write down 15–20 queries the way your customers phrase them (not the way a marketer would) and run them on ChatGPT, Perplexity, and Google. On Google, also watch whether an AI Overview appears for the query and which sources it cites. For each query, note three things: whether you’re mentioned, how you’re described, and who shows up instead of you.
It helps to know where each platform draws from. According to public 2025–2026 analyses, sources differ significantly between platforms (the specific shares shift by market, language, and query type):
| Platform | Sources it draws on most, per the analyses |
|---|---|
| ChatGPT | authoritative and encyclopedic sources (Wikipedia and the like) |
| Perplexity | a broad spread of sites including community content; usually many citations per answer |
| Google AI Overviews | classic organic results supplemented with multimedia |
A few prompts and a spreadsheet are enough for manual mapping. For ongoing measurement (more queries, competitive tracking) there are paid tools like Otterly, Profound, and AirOps — but that’s a step for later, not a precondition for your first audit.
Step 2: Are you letting AI crawlers in? (robots.txt and llms.txt)
If you want to be citable, AI systems have to be able to reach your content in the first place. In your robots.txt file, confirm you aren’t accidentally blocking the relevant AI crawlers and user agents:
GPTBot(OpenAI / ChatGPT)ClaudeBot(Anthropic)PerplexityBot(Perplexity)Google-Extended(controls how your content is used in some of Google’s AI features)
Draw a distinction between crawlers for classic indexing, for model training, and for live content retrieval — Google-Extended, for instance, isn’t an indexing crawler like Googlebot but a control token for content use in some of Google’s AI products. Allowing access doesn’t guarantee a citation on its own; it only removes the technical obstacle. Plenty of sites block these crawlers without knowing it — typically through a rule inherited from a template. If you’d rather restrict access, that’s a legitimate choice; how to do it and what it costs you is covered in the guide to limiting AI Overviews.
The llms.txt file comes up a lot too. Take it with a grain of salt: it describes what’s on your site, but according to the available 2025 analyses, widespread use of it by the major AI platforms hasn’t been demonstrated. In 2026, treat it as a nice-to-have — structured data and content are the priority.
Step 3: Can machines understand you? (structured data)
Structured data is a machine-readable description of what’s on a page — it helps search engines and AI systems recognize the page type, author, organization, product, price, availability, or frequently asked questions (though it doesn’t replace basic HTML structure). The impact isn’t cosmetic: it reduces the risk that machines misinterpret the page.
In the audit, check whether you have at least the basic types deployed site-wide and whether they’re valid:
Structured data to verify
- Article + breadcrumb navigation (BreadcrumbList) on content and article pages.
- Organization with contact details — name, address, phone.
- FAQPage wherever you genuinely answer common questions; for most sites it no longer guarantees a rich result in Google, but it does help machine understanding.
- Product (for e-commerce sites) product price and availability.
Verify it in the Rich Results Test and the Schema.org structured data validator. Why this matters so much for AI, and how to build the content structure around it, is covered in the guide to structuring pillar content and the section on generative engine optimization (GEO).
Step 4: Is your content written so it can be cited?
AI will skip even well-accessible, well-tagged content if there’s no easy answer to pull out of it. Go through your top pages and fix these:
Rules for citable content
- A short answer up top Right after the heading, 40–60 words that answer the page's main query on their own — it raises the odds AI uses them as a source.
- The answer within the first 100–200 words The intro sets the topic and the main answer; bury the answer further down and you reduce the chance the page gets judged relevant.
- The standalone test Every paragraph makes sense pulled out of context; rewrite vague sentences starting with It / This / That so they name a specific subject.
- Tables and lists Structured content gets cited more readily than a solid block of prose.
- Freshness Update content tied to the current year (2026, say) on a regular basis — stale numbers undercut both trust and citability.
Step 5: Measure and repeat
An AI SEO audit isn’t a one-off. Set up basic measurement — how many of your test queries mention you (your share of mentions, or share of voice) and which topics you’re losing citations on to competitors. Keep a simple update calendar for your key pages.
Then repeat the whole audit once a quarter. The landscape moves fast: new crawlers appear, and platforms change both the rules and the way citations work. If you’re actively working on AI visibility, an annual check tends to be too slow — a basic comparison of queries and citations is worth repeating more often.
Common mistakes to avoid
Treating Google rankings as AI visibility
According to 2025 analyses, plenty of pages cited in ChatGPT have weak to nonexistent organic visibility in Google.
Fix: Measure AI visibility separately — run manual queries on ChatGPT and Perplexity.
Accidentally blocked AI crawlers
A block inherited from a template reduces the chance AI systems find your content, understand it, and use it as a source.
Fix: Check robots.txt — GPTBot, ClaudeBot, PerplexityBot.
Relying on llms.txt
The llms.txt file is still a nice-to-have with no demonstrated widespread use by the major AI platforms.
Fix: Structured data and citable content are the priority.
One audit and done
Without repeating it regularly, you’ll fall behind the fast pace of change on AI platforms.
Fix: Repeat it once a quarter.
What’s next: a framework, or a custom audit
An AI SEO audit shows you what needs fixing on your site. When you want to apply that to your own site step by step — from the homepage through product and category pages to the blog — the AI SEO playbook lays out the framework: copy templates, structured data, and an implementation order you can start on in a single afternoon.
If you’d rather not handle the implementation yourself, this site is run by Sniper Design — we run custom AI SEO audits for e-commerce (WooCommerce, Shopify) and B2B sites. The playbook gives you the framework; a custom audit gives you the exact plan for your site.
