When marketers first hear “AI citations,” the instinct is to map them onto backlinks. They’re not. Backlinks are static, durable graph edges — link from A to B exists until someone deletes it. Citations from an AI model are something different: dynamic, probabilistic, and usually invisible to you unless you’re tracking them.
What a citation actually is
When a model decides to cite a URL alongside its answer, it’s not voting for that URL the way a backlink does. It’s saying: of the documents I retrieved while answering, this one was the highest-confidence support for the claim I’m about to make. That decision has three layers:
- Retrieval. The model (or its retrieval layer) pulls a candidate set of documents based on the prompt. You can’t be cited if you’re not in that candidate set.
- Selection. From the candidates, the model picks the ones whose content most directly supports the answer it’s generating. Authoritativeness, recency, and clarity all play in.
- Display. The frontend chooses how many citations to show — Perplexity shows everything, ChatGPT shows nothing in plain chat but shows sources in browse mode, Gemini shows a small inline link.
Why the SEO mental model breaks
The thing that makes an AI model cite you is not the same thing that makes Google rank you. Google rewards link equity and on-page optimisation. Models reward being the cleanest, most direct answer to the specific question, in a corpus the model has high confidence in.
That’s why Reddit and YouTube punch so far above their domain authority — they have the answer, in plain language, without the marketing varnish.