How to Cite ChatGPT in APA, MLA and Chicago Style

Most people assume citing ChatGPT works like citing a website: find an author, find a date, paste a URL, done. It doesn't work that way, because a chat session isn't retrievable by your reader, and every major style guide has had to improvise a shape for a source nobody else can look up. The instinct isn't wrong, though. All three systems still build the entry out of author, date, title and location; they just stretch those slots until they creak.

To cite ChatGPT, treat the company as the author and the model as the work. APA lists OpenAI as author with the model version in brackets, MLA puts your prompt in the title slot with ChatGPT as the container, and Chicago handles it in a note and usually skips the bibliography entry altogether.

Why a chat is an awkward thing to cite

Citation systems assume a stable, shared object. Your reader should be able to follow the entry, find the same thing you found, and check that you represented it fairly. A conversation with a language model breaks that assumption in three places at once: the output is not reproducible, the same prompt tomorrow returns something different, and the session usually isn't public.

So the styles fell back on the closest analogous cases they had. APA reached for software and unpublished data. MLA reached for its universal template and treated the tool as a container. Chicago reached for personal communication, which is its traditional home for sources a reader can't go and check.

Those are three different theories of what a chatbot is. Nobody is wrong, and the disagreement is the reason the formats look so unlike each other.

It also explains why the entries feel unsatisfying to write. A normal citation is a set of instructions for finding something. A ChatGPT citation is closer to a sworn statement: here is what I asked, here is when, here is what came back, and you will have to take my word for the rest. The formats below are all attempts to make that statement as specific and falsifiable as possible without pretending the underlying object is public.

How to cite ChatGPT in APA style

APA puts the company in the author slot, the model in the title slot, and a bracketed description after it. The version label matters because it is the only thing distinguishing what you used from what exists now.

Reference list template:

OpenAI. (Year). ChatGPT (Month Day version) [Large language model]. URL

Filled in, with your own details swapped for the placeholders:

OpenAI. (2026). ChatGPT (Apr 3 version) [Large language model]. https://chatgpt.com/

In-text, parenthetical: (OpenAI, 2026). Narrative: OpenAI (2026) generated a first outline of the argument.

Two habits make APA entries hold up. Describe the prompt in your running text rather than burying it, since the reference itself has nowhere to put it. And when the response is long or load-bearing, put the full text in an appendix and point to it, which is what APA recommends for exactly this reason.

How to cite ChatGPT in MLA style

MLA's approach is the most distinctive of the three: your prompt goes in the title-of-source slot, in quotation marks, followed by the word prompt.

Works Cited template:

"Your prompt text" prompt. ChatGPT, Version label, OpenAI, Day Month Year, URL.

A worked example:

"Explain the role of the chorus in Greek tragedy for a first-year seminar" prompt. ChatGPT, 13 Feb. version, OpenAI, 8 Apr. 2026, chatgpt.com.

In-text, use a shortened form of that prompt in quotation marks: ("Explain the role of the chorus"). If your prompt runs long, shorten it in the entry to its first meaningful clause rather than reproducing a paragraph.

MLA's guidance also asks for two things beyond format. Cite the tool whenever you quote, paraphrase or incorporate its output, and acknowledge functional uses (translation, outlining, code) in your text or a note even when nothing is quoted. It also warns you to verify any secondary sources the tool names, which is sound advice for the simple reason that fabricated references remain common.

Chicago style: a note, usually no bibliography entry

Chicago treats generated text as closer to personal communication than to publication. The consequence is structural: it goes in a footnote or endnote, and traditionally does not get a bibliography entry, because a bibliography lists things a reader can retrieve.

If your prompt is already stated in your text:

1. Text generated by ChatGPT, OpenAI, April 3, 2026, https://chatgpt.com/.

If the prompt is not in your text, put it in the note:

1. ChatGPT, response to "Summarize the main arguments against mandatory voting," OpenAI, April 3, 2026, https://chatgpt.com/.

Chicago's guidance on shareable links has been moving, and recent editions are more willing to allow a bibliography entry when a stable public link to the conversation exists. If you are writing for a journal or press, ask which edition they follow before you assume. Editors have opinions about this and they are not always the same opinions.

The three formats side by side

Read the Version row and the underlying logic separates cleanly. APA cares which model produced the text, MLA cares what you asked it, and Chicago cares that a reader knows the passage came from a machine on a particular day. Each guide has a coherent answer to a genuinely hard question, and none of them can give you the thing citation normally promises, which is verifiability.

Preserving the transcript, and why it matters twice

Every style assumes you keep a copy. Do it properly and it pays off in two different arguments.

The first is scholarly: an appendix containing the exact prompt and the full response lets a reader evaluate what you did with the output. In a methods-heavy discipline, this is not optional politeness. It is the only way your use of the tool is auditable at all.

The second is defensive. A student who can produce a dated transcript showing that they asked for an outline and then wrote 2,000 words themselves is in a materially better position than one who can only assert it. That folder pairs naturally with drafting history, and the case for building it before you need it is laid out in our guide to appealing an AI accusation.

Save four things at the time of use: the prompt verbatim, the response verbatim, the date, and the version label. Screenshot or plain text, either is fine. What is not fine is trusting that the session will still be in your history next semester.

Citation is not disclosure, and you may owe both

These get conflated constantly, and the difference is worth holding onto. A citation attributes specific content. A disclosure describes your process. You can have one obligation without the other, and increasingly you have both.

Here is the shape of the problem in practice. A master's student in public health asks a model for three counter-arguments to her policy thesis, discards two, and rewrites the third so thoroughly that not one generated sentence survives into her final draft. She cites nothing, reasonably, because nothing was quoted. Her department's policy asks students to disclose substantive assistance with the development of an argument. She has an obligation she doesn't know about, she has met none of it, and the failure is invisible until someone asks how she arrived at that particular counter-argument. Nothing she did was dishonest. She simply used the wrong mental model of what the rule was for.

That is the case worth designing your habits around, because it is far more common than outright copying and far easier to fall into.

The pressure here isn't only academic. The EU AI Act's Article 50 transparency obligations became applicable on August 2, 2026, requiring machine-readable marking of AI-generated content, disclosure of deepfakes, and notice when people are interacting with an AI system. That regime governs providers and deployers rather than students, but it has pulled the whole culture toward explicit labeling, and institutional policies have moved with it.

A disclosure sentence that works in most settings:

Generative AI (ChatGPT, OpenAI) was used to [outline the argument / suggest counter-examples / edit for concision] on [date]. All source verification, analysis and final wording are my own.

Specific verbs, a date, and a clear boundary. Vague disclosures ("AI was used in the preparation of this work") satisfy nobody and invite exactly the follow-up question you were hoping to avoid. If you want to know what the other side of that conversation looks like, our pieces on what professors can actually tell and what Turnitin does and doesn't catch are the companion reading.

What the style guides still haven't settled

Several genuinely open questions remain, and pretending otherwise would be a disservice.

How to cite output from a tool with no public version label. How to cite a model accessed through an API with sampling parameters that materially change the result. How to handle text you rewrote so heavily that the generated version is no longer recognizable in it, which is the case most students actually face and the one no guide answers cleanly. And whether shared conversation links count as retrievable sources, given that they can be revoked.

Non-text output is thinner still. Generated images, audio and code sit awkwardly in guides written for prose, and the working consensus as of mid-2026 is to adapt the same slots: developing organization as author, tool and version as title, date of generation, and the prompt recorded wherever the format has room for it. For images especially, note whether embedded Content Credentials survived, since provenance metadata is stronger evidence of origin than any citation you write by hand and platforms strip it constantly. AI Detector 360 inspects C2PA and EXIF provenance on uploaded images for exactly this reason, and an intact manifest is worth more than a paragraph of description.

The practical rule while those get worked out: describe what you did in enough detail that a skeptical reader could reconstruct your process, and choose transparency over elegance when the two conflict.

One last thing worth separating out. Citing correctly does not change a detection score, because detectors read the prose and never see your reference list. If you want to know how your finished text reads to a classifier before you submit it, AI Detector 360's AI essay checker returns a sentence-level heatmap with an explicit confidence level rather than a bare percentage, and the ChatGPT detector view breaks the result down passage by passage. Those are two separate problems, and solving one has never solved the other.