
How Do You Cite ChatGPT?
Carlos Garcia10/6/2026Almost every "how to cite ChatGPT" page you will find shows the same reference: `OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/`.
That format is three years old, and the reasoning behind it has since changed. The original problem was that nobody else could retrieve your conversation, so the style guides treated the tool itself as the thing being cited. Shared links solved that problem, and the guidance moved.
So the short version is: cite the specific conversation, not the chatbot. Below are the current formats for APA, MLA and Chicago, where each one differs, and the two practical traps that break an AI citation after you have submitted the work.
How Do You Cite ChatGPT?
Generate a shareable link to the exact conversation, then build the citation around that link rather than around ChatGPT in general. All three major style guides now point the same way, and all three decline to treat the AI as an author.
In APA, the company is the author and the specific chat is the work:
OpenAI. (2025, August 21). *High school grammar concepts* [Generative AI chat]. ChatGPT. https://chatgpt.com/share/68a77b60-0ee4-800c-9acc-cd3fd573c311
In MLA, the prompt is the title and the tool is the container:
"Describe the theme of nature in Jane Austen's *Mansfield Park*" prompt. *ChatGPT*, model GPT-4o, OpenAI, 23 Sept. 2024, chatgpt.com/share/66f1b0a0-d704-8000-be9a-85f53c850607.
In Chicago, it goes in a note rather than a bibliography entry:
Text generated by ChatGPT, OpenAI, January 10, 2026, URL.
Most credibility problems are citation problems wearing a different hat. Get a free audit and find out whether your content stands up to scrutiny.
Why the Old Format Was Replaced
The 2023 advice rested on a specific technical claim: your conversation with ChatGPT was not retrievable by anyone else, so it behaved like a personal communication — closer to an unrecorded interview than to a published source.
Under that reasoning, citing the chat was pointless because no reader could check it. The best available option was to cite the model as a piece of software and paste the transcript into an appendix.
Then OpenAI shipped shared links, and the premise stopped being true. A reader handed a share URL can now open the conversation and read it, which makes it a retrievable source and puts it back inside the ordinary citation machinery.
What APA Changed, and When
APA Style published new guidance on citing generative AI on 9 September 2025, moving from citing an AI tool generally to citing specific chat instances wherever sharing is available from the platform.
The practical result is that APA now has two formats, and which one you need depends on what you are referring to.
Citing a specific conversation:
`AI Company Name. (year, month day). *Title of chat* [Description, such as Generative AI chat]. Tool Name/Model. URL of the chat`
Citing the tool itself — appropriate when you are discussing ChatGPT as a product rather than quoting a particular exchange:
`AI Company Name. (year). *Tool Name/Model* [Description; e.g., Large language model]. URL of the tool`
Use the first for anything you quote, paraphrase or built on. Use the second only for statements about the software.
What MLA Says
MLA's guidance on generative AI was published in August 2025 and revised on 23 October 2025, and its headline position is explicit: "We do not recommend treating the AI tool as an author."
Instead the elements map onto MLA's standard template. The description of what you asked for becomes the title of the source. The tool — *ChatGPT* — becomes the container. The model goes in the version slot, with MLA asking you to "name the specific AI model or model version as specifically as possible." OpenAI is the publisher, the generation date is the date, and the location is, in MLA's words, "the stable, shareable URL for accessing the generated content."
If no stable link exists, MLA allows the general tool URL in the location slot instead.
What Chicago Says
Chicago keeps AI out of the bibliography and handles it in a note. The basic form is "Text generated by ChatGPT, OpenAI, January 10, 2026, URL," where the URL is the one copied from the Share button.
If the prompt is not already stated in your text, Chicago folds it into the note: "ChatGPT, response to [prompt], OpenAI, January 10, 2026, URL."
There is one more Chicago instruction worth following regardless of your style guide: if you edited the AI's output, say so — in the text or at the end of the note, for example "edited for style and content." That disclosure is what separates using a tool from quietly passing its work off as yours.
Guidance moves, and content that was right in 2023 is quietly wrong now. Get a free audit and find out which of your pages have gone stale.
How to Get the Link You Need
Open the conversation, use the share control on the chat, preview what it will show, and copy the link. That URL is what goes in the citation.
Two properties of that link decide whether your citation survives, and neither is obvious.
It Is a Snapshot, Not a Live Feed
A shared link captures the conversation up to the moment you created it. Messages you add afterwards do not appear in it.
This is the trap that catches people mid-project. You share a conversation early, keep working in the same thread for another hour, then quote something from the end of it. The link in your reference list does not contain the passage you quoted, and anyone who checks will find a conversation that stops short of your evidence.
The fix is simple once you know: generate the link after you have finished with the conversation, not when you first think of citing it.
You Can Break It Yourself
Shared links can be deleted — individually from the conversation, or all at once from settings. A deleted link is a dead citation.
Treat a link you have cited as part of the submitted work. If you do a periodic clean-out of shared links, exclude anything that appears in a reference list.
Links that quietly stop working are the most common form of content decay. Get a free audit and find out what is breaking on your site.
When Do You Actually Need to Cite It?
Cite it when the AI produced something that ends up in your work: a quoted passage, a paraphrased explanation, generated code, a summary you relied on, an outline you built from.
The harder cases are the ones people get wrong in both directions.
Used for mechanics only — spelling, grammar, reformatting a reference list — generally does not need a citation, in the same way a spellchecker does not. Check your institution's policy, because some ask for disclosure of any AI use at all.
Used to find sources that you then read and cited yourself: cite the sources, not the chatbot. The AI was a search step, and search steps are not citable. This one matters because language models produce plausible citations to papers that do not exist, so the rule has a second purpose — if you cannot find the source, you cannot cite it, and you have just caught a fabrication.
Used to generate something you then rewrote entirely: this is the genuinely ambiguous one. If the final sentences are yours and the AI supplied only a starting shape, most guidance treats that as process rather than source — but it is exactly the case where a disclosure sentence is worth more than a reference entry. See the last section.
Used as the subject of your discussion — writing about ChatGPT rather than with it: that is the APA general-tool format, and the one case where citing the product instead of a conversation is right.
Where This Still Does Not Work Cleanly
Style guidance is ahead of where it was in 2023, but there are real gaps.
Reproducibility is still weak. Your shared link proves what the model said to you on that date; it does not let anyone reproduce it. Run the same prompt tomorrow and you will get different text. A citation to a conversation is closer to citing a one-off event than citing a stable document.
Model versions are hard to pin down. MLA wants the specific model named, but the picker labels in consumer apps change, routing is not always visible, and the exact version serving a given message may not be stated anywhere. Record what the interface tells you at the time, and do not infer more precision than it gives you.
There is no agreed way to cite a conversation that spans models either. A long thread can be served by more than one model, and if you quote two passages produced under different versions, neither MLA's single version slot nor APA's single chat reference really accommodates that. The pragmatic answer is to cite the conversation once and name the model shown at the point you quoted from.
Institutional policy outranks all of it. Plenty of courses and journals have their own disclosure requirements, and some prohibit AI assistance outright. The citation format is the easy question; whether you were permitted to use the tool is the one that carries consequences.
Shared links are also not private by default in the way people assume. They are not designed to surface in public search results, but anyone holding the link can read the conversation. Before sharing one, check that nothing in the thread is confidential.
Citing It vs Disclosing It
These are two separate obligations and satisfying one does not satisfy the other.
A citation is a pointer. It tells a reader where a specific piece of material came from so they can go and look.
A disclosure is a statement of method. It tells a reader how the work was produced — that you used an AI assistant to draft, summarise or revise — and it usually belongs in a methods section, a cover note, or wherever your institution specifies.
The reason they get conflated is that both feel like admitting something, so people do one and consider the obligation met. In practice they are checked by different readers for different purposes: a marker follows your citations to verify a claim, and an integrity policy asks about your method.
A paper can be fully cited and still fall short on disclosure, and that is the version that tends to cause problems. If your use was broad enough to shape the work rather than supply one quotation, a sentence describing it is worth more than another entry in the reference list.
The fastest way to lose authority is to be caught claiming something you cannot show. Get a free audit and see how your content holds up.
Final Thoughts
The current answer is narrower than the question suggests: cite the conversation, give it a shareable link, do not list the AI as an author, and generate that link only when you are finished with the thread.
The reason this question is harder than it looks is not the formatting. It is that a chat is a strange kind of source: generated on demand, for one reader, with no edition and no fixed text. The style guides have settled on the closest available analogue, which is a dated, linkable document, and that works well enough as long as you remember what it is standing in for.
Most of the examples circulating online still show APA's 2023 format, which was built on the assumption that nobody could ever read your chat. That assumption no longer holds, APA revised its guidance in September 2025, and MLA revised its own in October of the same year. If a page you are copying from does not mention a share URL anywhere, it predates the change.
And keep the two obligations separate. The format question has a clean answer. Whether your use of the tool needed declaring — and whether it was allowed at all — is the one worth checking first.
If the wider question behind this is how AI-generated text gets identified in the first place, our guide to whether ChatGPT checkers actually work covers what AI detectors can and cannot reliably tell.



