How AI citation works
Citations come from retrieval. When an assistant searches the web to answer, it fetches candidate pages, extracts passages and attributes parts of the answer to them. A page is more likely to be cited when it is reachable by the assistant's crawler, contains a passage that directly supports a specific claim, and comes from a source the system considers reliable for that topic.
Access is the first gate. Each AI search product uses its own crawler or a partner index, and robots.txt rules aimed at AI training bots sometimes block the search crawlers too. Separating the two is worth doing deliberately: ChatGPT's maker, for instance, publishes different user agents for its search feature and for model training, so a site can allow one and refuse the other.
The misconception to avoid is that citations are stable or that being cited means being recommended. The same question can produce different sources an hour later, and an assistant may cite your statistics page while recommending a competitor. Count citations over many runs, and read what the answer actually says about you.
How to improve it
- 01Review robots.txt and firewall rules so AI search crawlers can reach the pages you want cited.
- 02Publish original, specific material, such as your own data, method notes or definitions, that others cannot easily repeat.
- 03Make each key claim a short, self-contained sentence with a clear subject, so it can support an answer alone.
- 04Show authorship, dates and sources on the page to make the content easier to trust.
- 05Watch referral traffic from AI assistants in analytics to see which pages earn clicks.
A commercial cleaning company publishes a page explaining how often office kitchens should be deep cleaned, with a clear schedule and reasoning. Its old robots.txt blocked every bot containing "GPT". After allowing the search crawler while still refusing training, the page becomes reachable for assistants answering facilities managers' questions.