What this article covers: The preparation: a 10-point checklist you can run yourself.
You prepare a website for AI search by taking care of three things: that AI crawlers can reach it, that they understand the content without guessing, and that they judge it reliable enough to cite. The ten points below cover all three steps. You can check each one yourself, most in under a minute.
The list is deliberately practical. If you want the background — what GEO and AEO actually are and whether SEO is really dying — it is in What are GEO and AEO. Here we go straight to work.
1. Does robots.txt allow the crawlers you want?
Why: crawlers have different roles, so blocking one doesn't have the same effect as blocking another:
- Googlebot indexes your site for Google Search. AI Overviews and AI Mode are built on that same index, and Google requires no extra settings for them. If Googlebot is blocked, you appear neither in Search nor in Google's AI features.
- Google-Extended is not a separate crawler but a robots.txt token. It controls whether Google may use your content to train and improve its Gemini models. According to Google, it does not affect your ranking in Search or your appearance in AI Overviews.
- Training crawlers, such as OpenAI's GPTBot and Anthropic's ClaudeBot, collect text for training future models.
- Search crawlers, such as OpenAI's OAI-SearchBot, Claude-SearchBot and PerplexityBot, build the index these tools use to cite and link sources in their answers.
- Blocking a training crawler does not stop the same provider's search crawler from citing you. If you block a search crawler, that tool generally won't cite you with a link, but it can still mention you based on other sources. So it's your decision, crawler by crawler.
Check: open yourdomain.com/robots.txt in a browser. Look for the crawler names. If they are listed with Disallow: /, they are blocked. If they aren't listed, the general rules (User-agent: *) apply to them — check what those allow. Note: a group for a named crawler inherits nothing from the general group. Source: Google Search Central, AI features and your website.
2. Is the hosting security system silently blocking them?
Why: this is the sneakiest point. robots.txt can be correct and the site opens fine for you, yet the hosting firewall returns a 403 to the crawler because it does not recognise it. No warning appears anywhere.
Check: the only reliable proof is the server access log — ask your hosting provider for the last month and search it for crawler names and response codes. If you see 403s, or no crawlers at all, the problem is here. We check this point directly in the free quick check.
3. Is the content in the source code or built in the browser?
Why: many AI crawlers do not execute JavaScript. If the page assembles its text in the browser, the crawler gets an empty shell.
Check: on the page press Ctrl+U (Mac: Cmd+Alt+U) and search the source for a distinctive sentence from the page. If it is not there, an AI cannot see it either. The same goes for content in tabs and behind a "read more" button if it loads only on click.
4. Does the site answer the questions customers actually ask?
Why: an AI system extracts an answer from your site only if the answer exists there — written in the words customers use, not your internal terminology. "How much does it cost", "how long does it take", "where do you operate", "what is included" are questions every business gets.
Check: take the five questions customers ask most often on the phone and try to find the answer to each on your site within 10 seconds. What you cannot find, nobody can.
Tip: The best form is a subheading phrased as a question with the answer in the first two sentences beneath it. Explanation can follow — but the answer comes first.
5. Does the site have an FAQ section — and is it marked up?
Why: an FAQ section is the easiest format to extract an answer from. If it is also marked up with the FAQPage schema, the system receives the questions and answers as data, not as text it has to parse. Note: FAQ schema does not guarantee rich results or appearing in AI answers; Google no longer shows FAQ rich results for most sites. Its value is as a tidy, machine-readable record of answers that are already visible on the page.
Check: open validator.schema.org, enter the page address and see whether FAQPage appears among the detected types. Important: the marked-up questions must also be visible on the page — hidden schema violates Google's guidelines.
6. Are the facts about your business machine-readable?
Why: who you are, where you operate, how to contact you, what you offer. If this exists only in prose, the system guesses — and that guessing is why an AI sometimes gets a business wrong. schema.org markup (LocalBusiness or Organisation with address, contact, service area and links to your profiles) removes the guessing.
Check: the same validator as in point 5. You are looking for Organisation, LocalBusiness or a related type, with address, telephone or email and sameAs (links to your Google Business Profile, directories, social profiles) filled in.
7. Are the facts the same everywhere?
Why: AI systems assemble your business from several sources — the site, your Google profile, directories, social networks. If the price, phone number or address differs between two of them, the system picks one. Not necessarily the right one.
Check: ask ChatGPT or Perplexity "What does [your company] offer and how do I contact them?" Every wrong fact in the answer has a source somewhere — find it and fix it at the source, not just on the site.
8. Is the page structure clear — headings, short paragraphs, lists?
Why: it is hard to extract an answer to a specific question from a page that is one solid block of text. Subheadings that say what a section is about, short paragraphs, lists and tables are the form that hurried readers and machines both understand.
Check: skim the page for five seconds and try to tell what it is about from the subheadings alone. If you cannot, the structure is weak.
9. Does the site contain anything that exists nowhere else?
Why: this is the point most checklists skip, yet Google's guide to AI features puts it front and centre: the systems look for unique, first-hand content — your own experience, concrete examples, images and data that cannot be obtained anywhere else. Generic text that could sit on any competitor's site is not worth citing.
Check: for every important page, ask whether a competitor could publish it unchanged. If yes, it lacks what is yours alone: a case from practice, a photo, a number from your work, an opinion.
10. Is the site fast and free of obstacles?
Why: crawlers have limited time per visit. Slow pages, interrupted loading and errors mean less content gets collected. For Google, speed remains part of how page experience is assessed in Search, and AI Overviews and AI Mode draw on the same index.
Check: pagespeed.web.dev — enter the address and look at the mobile score. More on why speed counts twice in Page speed and AI search.
What about llms.txt?
Deliberately not on the list. llms.txt is a proposed convention that no major provider officially honours. Since it is short and cannot do harm it is worth having — but if you have sorted the ten points above, you have done what genuinely counts. We prepare it for clients as part of the setup, not as a separate service.
What to do with the list
Points 1 to 3 decide whether an AI sees you at all — if any of them fails, everything else is moot. Points 4 to 8 decide whether it understands you. Points 9 and 10 decide whether it prefers to cite you over a competitor.
If you run out of time or access (logs, schema) on any point, we check them free of charge within one working day — including what you cannot see yourself. What is fine, we tell you is fine.
Frequently asked questions
How long does this check take?
Points 1, 3, 4, 5, 6, 7, 8 and 10 together take about fifteen minutes. Point 2 (server logs) needs your hosting provider; point 9 is a reflection, not a check.
Does blocking Google-Extended hide my site from AI Overviews?
No. Google-Extended only controls whether Google uses your content to train and improve its Gemini models. Appearing in Google Search and AI Overviews depends on Googlebot and the usual Search rules.
Do I have to allow AI crawlers at all?
It's your decision, crawler by crawler. You need Googlebot for Google Search and Google's AI features. You need other providers' search crawlers (e.g. OAI-SearchBot, PerplexityBot) if you want their tools to cite you with a link. You can block model-training crawlers (e.g. GPTBot, Google-Extended) without affecting search. What matters is that the setting is deliberate, not accidental.
Which points can I fix myself without a developer?
Points 4, 7, 8 and 9 are content matters you can handle in your site editor. Points 1, 2, 3, 5, 6 and 10 are technical — depending on how the site is built, you may need a developer.
How do I know it worked?
Repeat the questions from point 7 roughly 40 days after the work — earlier you mostly measure chance, since crawlers need to revisit the site. Compare whether you are cited and whether the facts are correct.
Does this require a redesign?
Usually not. Most points are fixed on the existing site. A redesign comes into play only if the site builds its content in the browser (point 3) or has no structure at all (point 8).