GEO basics: what it is and the five things to check first

Generative engine optimization (GEO) is the work of making your pages easy for AI search engines to fetch, read and quote. If ChatGPT, Perplexity, Claude or Google's AI features can't get your page, or can't make sense of it once they have it, they can't cite it.

Most of that work is unglamorous. It comes down to whether a crawler is allowed in, whether your text is in the HTML, and whether a passage still makes sense when it's pulled out of the page. This post covers what GEO includes, what it can't promise, and the five checks to run first.

What is GEO?

GEO is a set of practices that raise the chance an AI answer engine can find your page, understand it and attribute a statement to it. The name comes from "generative engine", the term for a search product that writes an answer instead of listing ten links.

It overlaps with SEO but isn't the same thing. The short version: SEO mostly decides whether a page ranks in a list, while GEO is about whether a page can be fetched, parsed and quoted by a system that writes an answer.

GEO is not a way to buy citations. Nobody outside the AI companies knows exactly how their engines pick sources, and anyone who guarantees you a mention in ChatGPT is guessing. What you can control is whether you're eligible.

The five things to check first

We built Georate's checks around the public documentation from OpenAI, Perplexity, Anthropic, Google and Microsoft. The two heaviest categories in the score are crawler access and content structure, at 20 points each out of 97, plus up to 3 bonus points. Here are the five areas that explain most failures, in the order to check them.

1. Can AI crawlers fetch the page?

Each AI company runs more than one crawler, and they do different jobs. As of October 2026, the ones that matter for search are:

The distinction matters because you can block training crawlers and still be eligible for AI search. OpenAI says so in its own documentation: its bots follow separate settings, so a site can allow OAI-SearchBot to appear in ChatGPT search while disallowing GPTBot. A blanket Disallow: / for everything blocks all three groups, which is how many sites end up invisible to AI search without anyone deciding that.

One caveat: fetchers that act on a user's request may not follow robots.txt. OpenAI notes that robots.txt rules may not apply to ChatGPT-User, because a person initiated the visit.

Look at your robots.txt first:

curl -s https://example.com/robots.txt

Then check for a firewall or CDN rule that returns 403 to bots even when robots.txt allows them. Robots.txt is only a request, and a bot-protection rule can still block the visit without showing up in the file. Our AI Crawler Checker sends a request as each search crawler and reports what came back, and the Robots.txt Checker parses the file and points to the line that matches.

We treat this as a gate in the score. If a search crawler for an engine is blocked, we cap that engine's score at 10, because nothing else on the page matters if the crawler can't get in.

2. Is the content in the HTML, without JavaScript?

Don't assume an AI crawler runs your JavaScript. Vendors rarely document it, and a page that builds its text in the browser can look empty to a crawler that only reads the HTML. You can see roughly what such a crawler gets:

curl -s https://example.com/ | grep -i "<h1"

If there's no <h1>, or the body is a single <div id="root"></div>, the content is rendered client-side. Server-side rendering or static generation fixes it. We check that the main content, the <h1> with its opening paragraph, and your structured data are all in the raw HTML.

Crawler behaviour changes, so treat "it's in the raw HTML" as the safe choice and not as a hard requirement for every bot.

3. Does each section answer something on its own?

AI engines quote passages, not whole pages. A passage that begins with "As mentioned above" and depends on the previous section can't be quoted cleanly.

What works is a heading phrased the way a person would ask the question, followed by a direct answer in the first sentence or two, then the detail. A short definition of the main subject near the top helps too. We check for this in the content category, along with specific facts such as numbers and dates, since a concrete statement is easier to attribute than a vague one.

4. Is the page labelled so machines can identify it?

Structured data (JSON-LD) tells a machine what a page is: an article, a product, an organization, an FAQ. It also gives a way to say who published it and when. Three details we check:

• The JSON-LD is valid and its type matches the page.
• Articles carry a headline, dates and an author, and the headline matches the visible heading.
• Any FAQ markup matches the FAQ text visible on the page.

None of this guarantees a citation. It removes ambiguity, which is the most you can ask structured data to do.

5. Does it look current and trustworthy?

Dates help a reader or a crawler judge whether a page is current. Check that the page shows a real updated date, that the dates in the HTML, the sitemap and the headers agree, and that the copyright year isn't three years old.

For trust, give the site an organization with a clear name, a logo and links to real public profiles, so a machine has something to match you against. Only link profiles that exist and that you control.

What GEO can't tell you

• Whether you'll be cited. A high score means the page is eligible and readable. It does not measure real citations. Georate is a diagnostic, not a ranking prediction.
• Whether you need llms.txt. As of October 2026 no major AI engine has confirmed using it, so we count it as a small bonus and never as a penalty.
• How a crawler will behave next month. Vendors add and rename bots. Check their documentation when something changes, for example OpenAI's crawler documentation, Perplexity's crawler page and Anthropic's help article on its crawlers.

Next step

Start with the first check, since it's the one that can make everything else irrelevant. Paste your URL into the AI Crawler Checker, and if something is blocked, the result includes a robots.txt snippet you can adapt. To check the other four areas on one page, run the full Georate check.

If you disagree with any of this or have seen a crawler behave differently, write to contact@georate.me.