GEO audit: can AI crawlers read, understand and cite your website?
Two analyses of your own site: whether AI crawlers can reach your pages and extract their content, and whether that content is worth citing.
Short answer
A GEO audit checks whether your website is readable by AI engines. In Refractia it has two parts: GEO Optimization, which scores every page with a 6-pillar GEO Score and builds a content plan, and the Technical Workshop, 56 crawlability and extraction checks, each graded A, B or C by the evidence behind it. The Technical Workshop also detects when your CDN or firewall blocks AI bots even though your robots.txt allows them.
What a GEO audit checks
A GEO audit answers two different questions, and it pays to keep them apart. The first is technical: can an AI crawler reach your page and extract the text? The second is about content: does what it finds answer the questions in your category clearly enough for an engine to use it, cite it and recommend you?
A classic SEO audit looks at how you rank in a list of links. AI engines read passages, summarize them and cite a few sources. A page can rank well and still be blocked for AI bots without anyone noticing. That is why the analysis is split into two modules:
| Question | Module | What you get |
|---|---|---|
| Can AI reach your content and extract it? | Technical Workshop | 56 crawlability and extraction checks, each with its evidence grade, organized as a backlog. |
| Does your content answer the question, and is it worth citing? | GEO Optimization | A 6-pillar GEO Score, page-by-page scores and a content plan. |
How the GEO Score is calculated
The GEO Score sums up in one number how AI engines read your site. It is built from six pillars, and each one answers a specific question:
| Pillar | The question it answers |
|---|---|
| Understanding | Can AI understand what your content is about? |
| Retrieval | Can AI find your content when it looks for an answer? |
| Citation | Does your page have what an engine needs to cite it as a source? |
| Recommendation | Does the content give AI reasons to recommend your brand? |
| Semantic coverage | Is the content complete for the questions in your topic? |
| Coherence across engines | Do the different engines read your content the same way? |
Every page analyzed gets its own score, so you can see which pages pull the result down and on which pillar. The standard analysis works on a selection of pages; full-site mode analyzes up to 500, for large sites and catalogs.
The analysis produces a content plan: what to create or improve. The plan does not stay in a report. It becomes briefs tracked as tasks in the Content Workshop.
Technical Workshop: 56 checks graded by evidence
The Technical Workshop runs 56 crawlability and extraction checks for AI. The difference is that every check declares the evidence behind it, and that evidence decides how much it weighs in the score:
| Grade | What backs it | Effect on the score |
|---|---|---|
| A | Documented by Google, or mechanically verifiable on your site. | Counts toward the score. |
| B | An independent study. | Counts toward the score. |
| C | An industry convention with no measured effect. | Shown as an observation, with zero weight. |
Why it matters: a lot of "SEO for AI" advice is convention repeated without ever being measured. If a convention weighs as much as a documented defect, your score drops for things that don't change what AI cites, and your team spends weeks fixing them. A grade C finding stays visible so you can decide, but it doesn't move the number.
A concrete example: Google documents that no llms.txt file or special markup is needed to appear in its AI features. An audit that penalizes you for not having llms.txt puts your team to work on something Google says you don't need.
The block robots.txt doesn't show
There is one problem a robots.txt check cannot see: your robots.txt lets AI bots in, but the bots never receive your content. Here is how it happens:
- 01
The file says yes
Your robots.txt allows AI bots to crawl. A check that only reads that file concludes everything is fine.
- 02
The edge decides otherwise
Between the bot and your server there is often a CDN or a web application firewall (WAF), such as Cloudflare or Imperva, with bot rules. Those rules don't read robots.txt: they decide by user-agent, IP reputation or the behavior of the visit.
- 03
The bot gets a challenge
If the rule flags the visit as suspicious, the CDN doesn't serve your page. It serves a challenge page ("checking that you are human") that requires running JavaScript. A browser gets through; an AI crawler doesn't.
- 04
AI is left without your content
Sometimes that challenge page returns a 200 status code, so basic uptime monitoring sees a healthy site. The AI engine, meanwhile, is left with the challenge text or with nothing.
The Technical Workshop detects when your CDN or WAF blocks AI crawlers even though robots.txt allows them: it looks at what the bot receives, not at what the file declares. The fix isn't in robots.txt either. It lives in your CDN's bot rules.
From finding to work: backlog and exports
Findings don't sit in a static report. The Technical Workshop organizes them as a backlog: it separates defects from observations, flags quick wins (low-effort fixes) and lets you track each item. It exports to PDF and Excel, so you can hand it to your development team or to the agency that maintains the site.
For brands with several product lines on one website, the technical audit and the GEO analysis are shared by the group; content work runs per line.
What a site audit doesn't solve
A readable site is a precondition, not a guarantee. In Refractia's data, a brand's own site accounts for 0.5% to 2.8% of the sources AI cites; the rest are media and third-party sites. That is why the audit should be read together with Influence Sources. It is also a snapshot: the analysis is started by a person on your team or at Refractia, and it does not run continuously.
Frequently asked questions
What is a GEO audit?
It is a review of your website built for AI engines: whether crawlers can reach your pages and extract the text, and whether the content answers the questions in your category in a way an engine will cite. In Refractia it combines the Technical Workshop (56 checks) and GEO Optimization (a 6-pillar GEO Score).
How do I know if ChatGPT can read my website?
Checking robots.txt is not enough. You need to confirm that no CDN or firewall serves the bot a challenge page and that the text on your pages can be extracted. The Technical Workshop checks both.
Can Cloudflare block AI crawlers even if robots.txt allows them?
Yes. CDN and firewall bot rules decide by user-agent, IP reputation or behavior, not by robots.txt, so an AI crawler can receive a challenge page instead of your content. The Technical Workshop looks at what the bot actually receives.
Do I need an llms.txt file to show up in AI answers?
Google documents that no llms.txt or special markup is needed to appear in its AI features. In an audit graded by evidence, not having one should not lower your score.
How is a GEO audit different from an SEO audit?
An SEO audit looks at how you rank in a list of results. A GEO audit looks at whether an AI engine can extract, understand, cite and recommend your content, and whether something blocks only the AI bots.
What does a grade C check mean?
It means the check rests on an industry convention with no measured effect. It is shown as an observation so you can take it into account, but it carries zero weight in the score.
Keep reading
Start by finding out what AI says about your brand
Your customers already ask AI what to buy; today you don't know what it tells them. In the free diagnostic we show you what ChatGPT, Claude, Gemini, Google and Meta AI say about your brand and your competitors. No commitment: we reply within one business day.