When you ask ChatGPT or Perplexity for a recommendation, you get a finished answer with a few sources. But why exactly these sources? Why do some companies show up and others never do? It's worth a look behind the scenes.
The process in two steps
Roughly, this is what happens:
1. Fetch sources. The AI searches for matching pages. Where from varies: ChatGPT and Copilot use a lot of the Bing index, Perplexity crawls on its own, Google AI Overviews draw on the Google index.
2. Build the answer. From these sources, the AI writes an answer and cites the pages it relies on.
Step one is what matters. If you're not in the index, you can't show up in step two. It's that simple.
What makes a page citable
The AI prefers content it can use easily:
- Self-contained paragraphs that make sense even without the rest of the page.
- Factual and concrete. Clear statements instead of vague advertising.
- Up to date. Old, dead content is used less often.
- Matching the question. The page has to cover exactly what's being asked.
- A trustworthy domain and a clear entity: Who you are, what you do.
Why some never show up
- They're not in any of the indexes the AI draws from.
- The content is buried in JavaScript or under so much clutter that the AI can't read it cleanly.
- The page stays vague and doesn't answer any concrete question.
- It lacks authority, nobody links to it.
What you can do
Readable, clearly structured content. Concrete answers to real questions. An llms.txt and structured data. And very practically: Register your site with Bing Webmaster Tools, because that's where ChatGPT gets many of its sources.
The big picture is in SEO is no longer enough: Now GEO counts too. If you want to know whether AI systems can use your site at all, take a look at my SEO and GEO analysis. For a first impression in seconds, try the free AI check.