When ChatGPT answers a question with citations, it has run a live web search, read a set of pages, and chosen which ones to quote. Each step is something a publisher can influence. This guide walks through the pipeline and what to do at each stage.
Step one: be crawlable by the right bots
OpenAI operates separate crawlers for separate jobs. OAI-SearchBot fetches pages for ChatGPT's search feature, while GPTBot collects training data. If your robots.txt blocks OAI-SearchBot, your pages cannot appear as citations no matter how good they are. Check your robots.txt and your CDN's bot rules, since some firewalls block AI crawlers by default without telling you.
Step two: rank in the underlying index
ChatGPT does not search the whole web from scratch. It retrieves candidates from a search index, so pages that perform well in conventional search are the pool it picks from. The boring fundamentals carry over directly: fast pages, clean titles and meta descriptions, a sitemap, and internal links that make every article reachable. If a page is invisible to ordinary search, it is invisible to ChatGPT.
Step three: write passages worth quoting
Once candidate pages are retrieved, the model selects passages. The pattern in cited pages is consistent enough to treat as guidance:
- The heading states the question and the first sentence under it gives the answer.
- Facts come with dates, numbers, and named sources rather than vague claims.
- Comparisons live in tables or tight lists the model can lift cleanly.
- The page commits to a position. Hedged content gives the model nothing quotable.
Step four: build entity consistency
Models resolve brands and authors as entities. A site whose name, author bios, and structured data all agree, and which is described the same way on other sites, is easier to attribute and safer to cite. Inconsistent naming across your own pages works against you in a way classic SEO never punished this hard.
Step five: measure and iterate
Watch for chatgpt.com referrals in your analytics, and sample the questions you care about by asking them in ChatGPT on a recurring schedule. Track whether you are cited, which page gets the citation, and what wording the answer borrows. That last detail tells you which passages the model finds quotable, and it is the most direct feedback loop AEO currently offers.
For the wider context on how answer engines choose sources, start with our plain-English guide to AEO, or browse the full blog for per-engine breakdowns.