How to cite ChatGPT
- Share the conversation before you leave it. Use ChatGPT's Share control to generate a stable link. Both style bodies now build the citation around that URL, and it is the only thing that lets a reader check what you were actually shown. Without it you are citing something nobody can open.
- Note the model and the date. MLA asks for the model in the Version element — model GPT-5.6 Sol, not just "ChatGPT" — and the date the content was generated. This matters more than it looks: OpenAI retires models from ChatGPT regularly, so the version is often the only record of what produced the text.
- Format for your style. MLA: the prompt or a description of it as the Title of Source, ChatGPT as the container, the model as Version, OpenAI as publisher, the date, then the share URL. APA: OpenAI as author, the year, the tool with its descriptor in square brackets, and the URL — with in-text citations as
(OpenAI, 2026). - Check whether you should be citing at all. Many institutions and journals want AI use disclosed in a statement rather than, or as well as, cited — and some prohibit it outright. Your assignment brief or the journal's submission policy overrides any style guide.
MLA read at source. APA not, and we would rather tell you.
MLA, verified directly. The MLA Style Center's updated guidance maps ChatGPT onto its template of core elements: do not treat the AI tool as an author; describe what was generated in the Title of Source element, often by giving the prompt; name the tool (ChatGPT) as the container; give the model in the Version element; name the company as publisher; give the date generated; and use the stable, shareable URL as the location, falling back to the tool's general URL only if no share feature exists. Its worked entry takes the form: prompt in quotation marks, then ChatGPT, model [name], OpenAI, [date], [share URL].
APA, second-hand — stated plainly. APA's own pages would not open through either of our fetching tools, so we could not read them the way we read MLA's. From APA's published guidance as it surfaced in search, the reference follows author–date–title–source with OpenAI as author, the tool named with a bracketed descriptor such as [Large language model], and the URL; in-text citations are (OpenAI, 2026) parenthetically or OpenAI (2026) narratively, and newer guidance supports citing a specific shared conversation by its title and share link. Confirm the exact template on APA's page before you submit — we are not going to present a format we could not open as though we had.
The point both bodies agree on. Do not cite ChatGPT for facts it attributes to somebody else. MLA's guidance is direct about why: AI tools can hallucinate or make up sources, or summarise them incorrectly, so the recommendation is to click through, read the original, and cite that. Our page on accuracy covers how routine fabricated citations are.
Style guidance in this area is revised as the tools change — MLA has already updated its post once and says so. Treat any citation format you find online, including this page, as a pointer to the official one rather than a substitute for it.
The parts that go wrong.
The single most common failure is citing a conversation nobody can reach. MLA now explicitly prefers a stable, shareable URL and treats the tool's general address as the fallback — because a link to chatgpt.com tells a reader nothing about what you were shown.
Generate the share link while the conversation is open. If you delete the chat later, you have removed the evidence behind your own citation — and per OpenAI's documentation, deleted chats are not retrievable through the interface, the API or support.
MLA asks for the model as specifically as possible, and this is the element people skip. It matters here more than in most citation work because OpenAI retires models from ChatGPT on a schedule — GPT-4o, GPT-4.1 and o4-mini in February 2026, GPT-5.1 in March, GPT-5.2 in June, o3 on 26 August 2026.
So a text generated by a model that no longer exists cannot be regenerated by anyone, including you. Recording which model produced it is the difference between a citation and a claim. Which model your account uses is on our page about what ChatGPT is.
These are different obligations and they are frequently confused. Citing covers a specific piece of AI-generated content you quoted or drew on. Disclosing covers your use of AI in producing the work at all — drafting, editing, summarising, generating ideas.
MLA points to separate guidance on acknowledging AI use beyond citation, precisely because a works-cited entry does not describe how a tool was used across a whole piece. Many institutions and publishers now require a statement, and some prohibit AI use in assessed work entirely. Your brief beats the style guide; check it before you decide a citation discharges the obligation.
Worth understanding rather than working around: a citation normally points at something fixed. This one does not. OpenAI states there is inherent randomness in generation, so the same prompt to the same model can produce a different answer — which is why the share link, capturing what you actually saw, does the work the page number does in a book citation.
It is also why treating ChatGPT as a source of fact is a poor idea regardless of formatting. A well-formed citation of a fabricated statistic is still a fabricated statistic; the format makes it checkable, not correct.
Better places to point a reader.
In most academic and professional writing, citing a chatbot is the wrong move — you want the thing it was paraphrasing. Each of these gets you to a citable original instead.
Frequently asked.
Quick follow-ups people search after this question.