How to Cite ChatGPT

Why Do You Need to Cite ChatGPT?

You cite ChatGPT for the same reason you cite a book or a website: to show your reader which words and ideas came from somewhere other than your own head. When a passage in your paper was produced by a model rather than by you, leaving that unmarked misrepresents who did the work, and that is the objection your instructor actually has, whatever name the syllabus files it under. Citing it is the honest fix, and where your course permits AI, it is also the fix that keeps you inside the rules. The University of Missouri’s Office of Academic Integrity draws the line at permission rather than at any percentage: a student who uses ChatGPT with the instructor’s approval is not committing dishonesty, while using it without approval breaks the rules (EV-good-ai-score-02). The MLA Style Center is more specific about the trigger, telling writers to cite generative AI whenever they quote it, paraphrase it, or build their own text on its output (EV-is-ai-writing-plagiarism-04).

There is a catch worth naming before you relax, because it is the reason this page exists rather than just a format chart. A correct citation clears the authorship question, not the detection question. Even a passage you cited perfectly and disclosed honestly can read as more machine-written than you expect, because a detector reads the texture of the prose, not your reference list. That is why the honest first move, before you submit anything AI-assisted, is to see how the text itself reads. You can check a paragraph on MeteGPT’s free detector and treat the number as one signal rather than a verdict.

Do You Have to Cite ChatGPT If Your Course Allows It?

If your course allows AI and you used ChatGPT to produce text you are handing in, then yes, you almost always have to disclose it, and the format below is how. Permission to use a tool is not permission to hide that you used it. The two are separate grants, and most policies that open the door to AI attach a disclosure condition on the way through. Where they do not spell it out, disclosing is the safe default, because the cost of an unrequested acknowledgment is nothing and the cost of a missing one can be a misconduct case. Whether an undisclosed use crosses from sloppy into an actual violation depends on your specific institution, which is a policy question rather than a formatting one; the fuller answer to whether AI writing counts as plagiarism works through how different schools draw that line.

This is also the one place a rewriting tool has a legitimate role, and I will be plain about both the use and its edge. If your policy permits AI-assisted drafting and you want the assisted parts to read in your own voice rather than in a flat model register, you can revise those parts into your own register with MeteGPT and see how they score in the same pass. Humanizing where it is permitted and disclosing the help are two separate steps, and doing one never lets you skip the other; a passage you rewrote still needs the same citation a quoted one would. What this is not is a way to launder a prohibited use into an allowed one. If your course bans AI outright, no citation and no rewrite changes that, and I would rather turn that reader away here than pretend otherwise.

How to Cite ChatGPT in APA Format

APA treats ChatGPT as software, crediting OpenAI as the author of the algorithm, because a private chat session is not something your reader can retrieve and check the way they could a published article. That single decision drives the whole format.

In-Text Citation

The in-text citation is a standard author-date parenthetical naming OpenAI and the year:

(OpenAI, 2023)

If your chat has a shareable link, use the year of that shared conversation instead, for example (OpenAI, 2026).

Reference List Entry

The APA Style blog’s original ChatGPT guidance gives this reference-list form, treating the model as software and adding the version and a bracketed description (APA Style blog, “How to Cite ChatGPT,” last updated September 15, 2025; EV-how-to-cite-chatgpt-01):

OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat

APA’s newer guidance, “Citing generative AI in APA Style,” updated the form on September 9, 2025 (EV-how-to-cite-chatgpt-02). Because most chat tools, ChatGPT included, now offer a shareable link to a specific conversation, APA recommends citing that specific chat when a stable URL exists, and it dropped the mandatory version-number field, keeping only the model name where you have it. The worked example APA prints under this newer template is for Claude, so here is the same template with ChatGPT substituted in, element for element:

OpenAI. (2026, September 12). Comparing single-stage and two-stage thesis structures [Generative AI chat]. ChatGPT. https://chatgpt.com/share/1a2b3c4d

Read the elements left to right: the AI company as author, the date of the chat, a short description of the exchange in place of a title, the bracketed content type, the tool or model name, and the shareable link. Use the older version-number form when your chat has no shareable URL, and the newer shared-chat form when it does.

How to Cite ChatGPT in MLA Format

MLA does not treat the AI as an author. It runs ChatGPT through its core-elements template, building the entry around the prompt you typed, with the model name in the Version slot and a shareable conversation link as the Location when one exists.

Works-Cited Entry

The MLA Style Center’s updated guidance, revised August 13, 2025, gives this worked ChatGPT entry (EV-how-to-cite-chatgpt-03):

"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.

The prompt, in quotation marks and labeled as a prompt, opens the entry. Then the tool, the specific model, the company, the date of the exchange, and the shareable link.

In-Text Citation

MLA in-text citations point to the first element of the works-cited entry, and here that first element is the prompt rather than an author. So the in-text mention uses a shortened form of the prompt in parentheses:

("Describe the theme of nature")

Paraphrase vs. Direct Quote Example

Signal in your sentence whether you are quoting the model or summarizing it, the same way you would with a human source. A direct quote sits in quotation marks and points to the entry: When prompted about Austen, the tool replied that nature “frames the moral education of the characters” (“Describe the theme of nature”). A paraphrase drops the quotation marks but still names the tool and still carries the citation, so your reader can see the idea originated with the model and not with you.

When ChatGPT Points You to a Secondary Source

One trap worth its own line: when the model attributes a claim to a real study, article, or author, do not cite ChatGPT as the source of that work. Find the original, read it, and cite that directly. The model was your route to the source, not the source itself, and citing the route as if it were the origin both misstates where the evidence came from and inherits any error the model introduced. If you cannot locate the original, that is a strong sign the claim needs to come out rather than go in.

How to Cite ChatGPT in Chicago Style

Chicago is the outlier of the three, and knowing that saves graduate students in particular a lot of second-guessing. It handles ChatGPT closer to a personal communication than to a published work: you credit it in a note, and you generally do not build a separate bibliography entry.

Note or In-Text Mention

The Chicago Manual of Style’s own Q&A states that you do need to credit ChatGPT and similar tools whenever you use text they generate in your own work, and it offers a numbered footnote or endnote as the formal option, with or without the prompt quoted inside it (EV-how-to-cite-chatgpt-04). Its worked note reads:

ChatGPT, response to "Explain how to make pizza dough from common household ingredients," OpenAI, January 10, 2026, URL.

Chicago also accepts crediting the tool directly in your running text, for example noting in the sentence itself that the following summary was generated by ChatGPT in response to your prompt, in which case a separate note may not be needed at all.

Do You Need a Bibliography Entry?

Usually not. Chicago’s guidance is that it is not necessary to list generative AI tools in a bibliography or reference list, particularly when the generated content has no publicly available link for a reader to follow. The note, or the in-text credit, carries the acknowledgment on its own. If your instructor specifically asks for a bibliography line anyway, follow that instruction over the general rule; a stated course requirement always outranks a default.

A quick caution that applies to all three styles: a third-party “ChatGPT citation generator” can steer you wrong, because these tools tend to lag behind the style authorities. APA’s September 2025 revision dropped the version-number field that older generators still insert, and Chicago moved into its 18th edition, so a generator that has not caught up will hand you a form that looks official and is a revision or two out of date. When in doubt, match your output against the primary-source examples above rather than trusting a generator field.

Here is how the three formats line up side by side, so you can see at a glance which one your course expects and how differently each treats the same chat.

Is It Plagiarism If You Don’t Cite ChatGPT?

Not in the classic sense, and yet it can still land you in the same disciplinary place, which is exactly why the question confuses people. Classic plagiarism means passing off another person’s work as your own, and a model is not a person whose credit you withheld, so a strict reading of the old definition does not catch AI text on its own terms. That is a technicality, not a green light. Handing in undisclosed ChatGPT output still misrepresents who did the work, and some institutions choose to file that directly under plagiarism regardless of the wording gap. Touro University’s AI addendum, effective January 2026, does exactly that, treating undisclosed AI use as plagiarism no matter what a syllabus does or does not say (EV-is-ai-writing-plagiarism-03). Others call the same act unauthorized assistance. The label moves from campus to campus, but the underlying wrong, and the consequence, does not.

Because the answer genuinely turns on your institution’s own definitions, I am not going to flatten it into a universal yes or no here. If that is the question actually keeping you up, the fuller answer to whether AI writing counts as plagiarism lays out how real schools split on it and how to read your own policy against those examples. The practical takeaway for this page is narrower and firmer: citing removes the ambiguity, and it costs you nothing to do it where AI use is permitted.

Does Citing ChatGPT Stop Turnitin From Flagging It as AI-Written?

No. A citation and an AI detector answer two unrelated questions, and this is the single most useful thing on this page to understand before you submit. Your citation tells a human reader where the words came from. A detector like Turnitin’s never reads your reference list or your footnote; it scores the texture of the prose itself, the evenness and predictability that models tend to produce, and returns a probability. So a passage you cited flawlessly and disclosed honestly can still come back flagged, not because you did anything dishonest, but because the words carry a machine-like rhythm the tool was built to distrust. If you want the mechanics under that number, why a detector flags text at all walks through what the score is actually measuring.

I watch this collision land on honest students more than any other. Someone does everything right, permitted use, correct APA entry, a clean disclosure note, and the detector reads high anyway, and the citation they were counting on to protect them turns out to have been answering a different question the whole time. The citation was never going to move that number, because the number is not about attribution. Naming this openly is the point: a correct citation is necessary, and it is not sufficient. This is a reason to look before you submit, not a reason to hide anything, and certainly not an instruction to dodge a tool. Knowing what a detector sees is reader protection.

So work it in order. If you wrote the passage yourself and it got flagged, do not rewrite it; document your drafts and appeal, and run the whole draft through a checker first so you know what an instructor is looking at. If instead you drafted with AI where it is permitted and want to see how the text reads before it counts, check that passage on MeteGPT’s free detector. The free detector reads 125 words at a time with no account and no signup, four scans a day, which covers a paragraph rather than a full essay.

And if your course permits AI and you want a disclosed, assisted draft to read in your own voice, MeteGPT revises the text and scores every rewrite in the same screen, so you watch the number move instead of pasting into a second site and hoping. The free tier is deliberately small, and I would rather you hear the ceiling from me now than hit it mid-assignment: a free account includes four humanizer runs in total, each capped at 250 words, and those four do not refill the next day. That is enough for one disclosed paragraph in your own register, not a chapter of them. If you are working across a long essay or a thesis section, see what a full account unlocks before you start, so the cap is a decision you made rather than a wall you hit.