ChatGPT is where the 'AI never mentions us' panic usually starts, because it is where buyers actually ask. The mechanics of its citations are knowable enough to work with - and honest enough that anyone promising guarantees is guessing at your expense.
In short
ChatGPT cites sources when it searches the live web to answer - so getting cited means being retrievable (crawler access, search-mode visibility), being the clear answer to the asked question, and being corroborated by other sources the model also sees. No one controls it; all of it is influenceable.
Two modes, two problems
Answer-from-knowledge: the model responds from training, often without citations. Your presence here reflects your corroborated footprint in the broader web over time - slow to move, moved by authority. Answer-with-search: the model fetches live pages and cites them. This is the fast lane: retrievable, quotable pages can enter citations within re-crawl cycles, not training cycles. GEO work targets the search mode first because it responds to effort on measurable timelines.
The mechanics that matter
Access: OpenAI's search and browsing fetchers must be allowed (see crawler policy - training-bot and retrieval-bot rules are separate decisions).
Question coverage: the fetch starts from the user's question. Pages titled and structured as answers to real questions get retrieved; category brochures do not.
Quotability: the model lifts the passage that answers cleanly - summary up top, steps, one claim per passage, dates and methods stated.
Corroboration: when several retrieved sources agree and one of them is you, being the clearest of them wins the citation. When you exist only on your own domain, you are one voice claiming itself.
Freshness: live-search answers favor recently fetched content; stale pages drift out of the pool.
Measuring it, instead of refreshing and hoping
Citation presence is trackable: systematic prompting over a window, logged and scored - our own monitoring runs daily on our topics. Paired with crawler-log watching (OpenAI's fetchers hitting your pages is the leading indicator), you get a real feedback loop: publish, watch retrieval, watch citations, adjust. That loop is the service; everything else on this page is the mechanism it turns.
Common questions
Can we pay OpenAI to be cited?
No - cited organic answers are not a placement product. Treat anyone selling guaranteed ChatGPT citations accordingly.
Why does ChatGPT cite our competitor and not us?
Run the diagnosis: are you retrievable, do you have a page answering that exact question, is it quotable, and do third-party sources corroborate you? In our audits, the answer is usually cause two - the page does not exist.
Does ChatGPT use our structured data?
Structure helps retrieval and comprehension generally; there is no documented schema-to-citation pipeline. Ship clean structure for its proven benefits and skip vendors claiming schema hacks.
How fast can citations change?
Search-mode citations can shift within re-crawl cycles when a clearly better answer appears. Knowledge-mode presence moves on much slower authority timescales. We measure both separately.
Find out what ChatGPT sees when it looks for you.
Your citation baseline on the questions that matter, OpenAI fetcher activity from your own logs, and the gap plan - measured.