You spent years optimizing for Google's ten blue links. Today your most consequential visitor is often an AI agent deciding whether to read, trust, and recommend you. If ChatGPT crawls your site right now, can it reach your content, parse your pricing, and cite you in an answer? A traditional SEO audit will not tell you. A GEO audit will. This guide shows you how to run one, what to check, how to fix what you find, and how GEOly's 29-check GEO Audit scores your AI-readiness in minutes.
By the end you will have checked crawler access, structured data, and content readiness, and produced a prioritized fix list.
Key takeaways
- A GEO audit measures whether AI crawlers can access, parse, and cite your site, which is a different question from keyword rankings and backlinks. - The four things that most often block AI visibility are crawler access in `robots.txt`, missing or thin structured data, low semantic density, and content buried behind scripts. - Check for the real AI user agents by name, including `GPTBot`, `OAI-SearchBot`, `ClaudeBot`, `PerplexityBot`, and `Google-Extended`, and decide access deliberately rather than by accident. - An `llms.txt` file gives models a clean map to your most important content, complementing rather than replacing `robots.txt`. - GEOly runs a 29-check GEO Audit that scores site and page AI-readiness on a 0 to 100 / A to F scale, with a free 3-day trial at app.geoly.ai.
Step 1: Understand why an SEO audit is not enough
SEMrush and Ahrefs are excellent at backlinks and keywords, but AI models weigh different signals. They care whether your core content fits and reads cleanly in a context window, whether it is dense enough in verifiable facts to be worth citing, whether your crawler rules accidentally shut them out, and whether you expose the structured data that helps them identify your entity.
A GEO audit checks those signals directly. Think of it as asking, for every page, "if an AI read only this, would it understand what you sell, trust it, and be able to cite it?"
Gotcha: passing a Lighthouse or SEO audit does not mean you pass a GEO audit. A fast, well-ranked page can still block `GPTBot` or ship its content only after JavaScript runs.
Step 2: Audit AI crawler access in robots.txt
Start where visibility most often dies: `robots.txt`. Many sites unintentionally block the very agents that feed AI answers. Open `yourdomain.com/robots.txt` and look for the AI user agents by name.
The ones that matter include `GPTBot` and `OAI-SearchBot` (OpenAI), `ClaudeBot` (Anthropic), `PerplexityBot` (Perplexity), `Google-Extended` (Google's AI training control), and `CCBot` (Common Crawl, which feeds many models). A blanket `Disallow: /` under one of these, or a broad rule you forgot about, quietly removes you from those systems.
Decide access on purpose. If you want AI visibility, allow the search and answer crawlers. You may still choose to disallow training-only bots; that is a policy call, not a default. For the reasoning behind auditing this now, read [why brands should audit AI crawler access now](/blog/why-brands-audit-ai-crawler-access-now).
Gotcha: `Google-Extended` controls AI use, not classic Google Search indexing. Blocking it does not hurt your normal search rankings, but it does remove you from some Google AI surfaces.
Step 3: Give models a map with llms.txt
`robots.txt` says what a crawler may not touch. An `llms.txt` file at your root does the opposite: it points models to your most important, cleanest content, in Markdown, so they read the right pages without wading through navigation and boilerplate.
List your core product, pricing, docs, and key articles with short descriptions. Keep it current, because a stale map is worse than none. For a full walkthrough, see [how to create an llms.txt file for GEO](/blog/how-to-create-llms-txt-file-for-geo).
Gotcha: `llms.txt` is an emerging convention, not a guarantee every model reads it. Treat it as one helpful signal alongside clean HTML and good structured data, not a silver bullet.
Step 4: Check structured data and semantic clarity
AI models identify entities, not just keywords. Confirm you ship valid JSON-LD for the schema types that describe you, typically `Organization`, `Product`, `FAQPage`, and `Article`. These give a model unambiguous facts: who you are, what a product costs, what questions you answer.
Then check semantic density. State plainly, in your own words, what you are and who you serve, especially on your homepage and about page. A sentence like "[Brand] is a [category] for [audience] that [core benefit]" is easy for a model to lift verbatim into an answer.
Gotcha: schema that contradicts your visible page, for example a `Product` price that no longer matches, actively hurts trust. Validate it and keep it in sync with what users see.
Step 5: Confirm content is readable without scripts
Models are far better at reading server-rendered HTML than content that only appears after client-side JavaScript. View source, or fetch the page as plain HTML, and confirm your key copy, headings, and prices are present in the markup, not injected later. If your product details only exist post-render, many crawlers see an empty shell.
Gotcha: single-page apps are the classic offender. If your framework renders client-side, use server-side rendering or prerendering for the pages you most want cited.
Step 6: Run GEOly's 29-check GEO Audit and prioritize
Doing all of the above by hand is thorough but slow. GEOly's GEO Audit runs 29 AI-readiness checks across crawler access, structured data, content, and technical setup, and returns a site- and page-level score on a 0 to 100 / A to F scale, so you get a ranked fix list instead of a pile of raw findings. It is part of the broader platform that also tracks your AIGVR, Share of Model, citations, and competitors, so the audit connects to whether the fixes actually move your visibility. Start with a free 3-day trial at app.geoly.ai; see [what GEOly AI is](/blog/what-is-geoly-ai) for how the audit fits the rest.
Work the list by impact: unblock crawlers first, then fix structured data and rendering, then deepen thin content. For content-level tactics, [best practices to increase AI crawling and citations](/blog/best-practices-increase-ai-crawling-citations) pairs well with your fix list.
How to verify your fixes landed
Re-fetch `robots.txt` and confirm the AI agents you intend to allow are no longer disallowed. Validate your JSON-LD in a schema testing tool and confirm no errors. Fetch a key page as raw HTML and confirm the copy and price are in the source. Then re-run the GEO Audit and compare the new score to your baseline; a rising grade with the same fix list cleared is your proof.
FAQ
Is a GEO audit different from an SEO audit? Yes. SEO audits weigh rankings and backlinks; a GEO audit weighs whether AI crawlers can access, parse, and cite you.
Which crawlers should I allow? At minimum the search and answer bots like `OAI-SearchBot`, `PerplexityBot`, and `ClaudeBot` if you want AI visibility. Training-only bots are a separate policy choice.
Do I need llms.txt? It is optional but helpful. It guides models to your best content and complements, not replaces, `robots.txt` and clean HTML.
How many checks does GEOly's audit run? 29 AI-readiness checks, scored at the site and page level from 0 to 100 / A to F.
How often should I re-audit? Re-run after any major site change, and on a regular cadence otherwise, since crawler rules and rendering can regress silently. See the guides from [GEOly AI](/blog/author/geoly-ai).

