Getting your site cited in AI search

Services called GEO and AEO have already appeared, priced like full search marketing. Here is what actually works.

1. Structured data

Schema.org markup is the most direct way to tell a machine what is on a page. The minimum for a company site: Organization with contacts, Service on service pages, FAQPage for questions, BreadcrumbList for navigation, Article in the blog.

None of this is new or hype — markup mattered before AI too. It just pays off twice now.

2. How the text is shaped

Models pull fragments that stand on their own. In practice that means:

• the answer goes in the first paragraph of a section, not the last;
• headings are phrased as a person’s question, not as a category label;
• facts and figures go in a list or a table, not buried in prose;
• a short summary block at the top works better than a long introduction.

Notice that all of this is simply good structure. Human readers like it too.

3. llms.txt

A file at the root of the site describing, in plain language, what the project is and which pages matter. The standard is not formally adopted and there are no guarantees. But it takes half an hour and breaks nothing — we run one and suggest it to clients with large catalogues.

4. The technical basics

The crawler gathering data for a model behaves like any other. If a page takes five seconds, serves content only through JavaScript, or is blocked in robots.txt, it will not appear anywhere. Check that first, then worry about the finer points.

What not to do

Do not stuff the text with “according to experts” hoping to look authoritative, and do not buy “guaranteed placement in AI Overviews”. Nobody controls what a model shows, exactly as nobody guarantees the top spot in Google.