How ChatGPT Cites Sources
When ChatGPT answers a question using live web search, it often shows the links it drew from. Understanding roughly how that works — and being honest about what is not public — helps you make your content easier to retrieve and reference.
A grounded answer surfaces numbered links back to the pages it retrieved.
The general mechanic
It helps to separate two modes. In one, ChatGPT answers from the knowledge baked into the model during training — no live sources, no citations. In the other, the query triggers a web search, and the answer is grounded in pages retrieved in real time. When that happens, ChatGPT search surfaces cited links so the reader can follow them. The three steps below describe the well-understood shape of that process. The precise ranking signals that decide which pages get retrieved and cited are not published, and the behavior changes frequently.
Retrieval
When a query needs current or specific information, the system searches the web and pulls a set of candidate pages relevant to the question.
Grounding
It reads those retrieved pages and composes an answer anchored in their content rather than only from model memory.
Citation
The answer surfaces links to the pages it drew from, so a reader can click through and verify the claim.
What we genuinely don't know
Be wary of confident claims about “the algorithm.” OpenAI has not published the exact criteria that decide which retrieved pages become citations, how many sources an answer draws from, or how freshness, authority, and relevance are weighted against each other. Anyone stating precise percentages or a fixed ranking formula is guessing. What is reasonable to assume — based on how retrieval-grounded systems generally work — is that content which is accessible, clearly on-topic, and unambiguous is easier to retrieve and quote than content that is blocked, buried, or vague.
How to make your content easier to cite
Stay retrievable
Don't block GPTBot in robots.txt unless you intend to. If a crawler can't reach you, you can't be retrieved.
Answer questions directly
Lead with a clear, self-contained answer near the top of the page. Grounding favors passages that state the point plainly.
Be specific and factual
Concrete details — names, numbers, procedures — give a system something quotable to anchor a claim to.
Structure for machines
Clean headings, a curated llms.txt, and accurate structured data all reduce ambiguity about what your page is about.
For a step-by-step version aimed at ChatGPT specifically, see how to get found on ChatGPT.
Frequently Asked Questions
Does ChatGPT always cite sources?
No. When a query triggers web search, cited links are commonly shown. For questions answered from the model's built-in knowledge, there may be no live citation at all. Whether search is triggered depends on the query and the product surface, and that behavior changes over time.
Can I pay to be cited?
There is no advertised pay-to-cite mechanism for organic answer citations as of mid-2026. Citations reflect what the system retrieved and found relevant. Be skeptical of anyone claiming a paid shortcut into AI answer citations.
What is GPTBot?
GPTBot is OpenAI's web crawler. Reputable crawlers document that they respect robots.txt directives, so if you want your content available for retrieval, make sure you are not inadvertently blocking it. Compliance is not guaranteed and is ultimately advisory, but blocking a crawler is a reliable way to reduce your chances of being retrieved.
How do I know if I'm being cited?
Ask AI systems the kinds of questions your customers ask and see whether your pages appear as sources. Tools that track AI visibility can automate this sampling, though no tool can see the full, private selection process.
Will an llms.txt file guarantee citations?
No. There is no guarantee an AI system will use your file. llms.txt is advisory — a curated map of your content. It can make your site easier to understand, but retrieval and citation decisions are made by the AI system, not by your file.
Are you showing up in ChatGPT?
Check whether your site is retrievable and generate an llms.txt to make your content easier to understand.