Citing ChatGPT or another generative AI tool is not the same task as disclosing that you used one. A citation tells a reader where a specific piece of text or output came from, in whatever format your target style guide requires. A disclosure statement is a separate, often journal-mandated declaration of how AI was used in producing the manuscript at all — drafting, editing, translation, code, and so on. Most manuscripts that use AI need both, and they are not interchangeable. If you’re looking specifically for disclosure wording, see AI Disclosure Statement Examples and the underlying generative-AI disclosure statement concept. This page covers the citation mechanics: how APA, MLA, and Chicago each want you to format a reference to AI-generated text, why all three treat that output as fundamentally different from citing a normal source, and the research-integrity obligation that comes before any of it — verifying that anything ChatGPT tells you to cite actually exists.
Quick answer
| Style | Reference-list / Works-cited entry | In-text citation |
|---|---|---|
| APA (7th ed.) | OpenAI. (2025). ChatGPT (Aug 2025 version) [Large language model]. https://chatgpt.com/ | (OpenAI, 2025) — parenthetical, or OpenAI (2025) narrative |
| MLA (9th ed.) | “Describe the symbolism of the green light in The Great Gatsby” prompt. ChatGPT, 13 Feb. version, OpenAI, 8 Mar. 2025, chatgpt.com/. | No separate parenthetical is generally needed if the tool is named in the text; otherwise cite by the first element of the works-cited entry |
| Chicago (18th ed.) | Not included in the bibliography/reference list at all — cited only in a footnote or in-text parenthetical, unless you provide a publicly shared, login-free link to the exact exchange | Footnote: “Text generated by ChatGPT, OpenAI, March 7, 2025.” / Author-date: (ChatGPT, March 7, 2025) |
Each format is explained in full, with the reasoning behind it, below. All three are pulled directly from each style body’s own published guidance and re-verified for this page (see the last-verified date at the bottom) — not inferred or reconstructed from memory, since all three bodies have revised this guidance since generative AI chat tools launched and will likely revise it again.
APA style: how to cite ChatGPT
APA Style’s official position, published on the APA Style blog, is to treat ChatGPT (and comparable tools) as an algorithm, with OpenAI credited as the author/publisher — not as a personal communication and not as an anonymous work.
APA reference-list entry
OpenAI. (2025). ChatGPT (Aug 2025 version) [Large language model]. https://chatgpt.com/
Structure: Author (the company) → Year → Title of the tool, italicized, with the specific version in parentheses → bracketed description of what it is → URL. Swap in whichever tool and company actually generated your output (e.g., Google for Gemini, Anthropic for Claude) and the version/date you actually used.
APA in-text citation
- Parenthetical:
(OpenAI, 2025) - Narrative:
OpenAI (2025) generated the following summary…
APA’s guidance on the prompt and the output
Because a ChatGPT conversation is not a stable, publicly retrievable source — nobody else can go to that URL and see what you saw — APA’s own guidance is explicit: quote or reproduce the prompt you used, then quote the relevant portion of the generated text immediately after it, in the body of your paper. For longer outputs, APA recommends placing the full response in an appendix and citing it in text as, for example, (OpenAI, 2025; see Appendix A for the full transcript), with a note that the same prompt can produce a different response in a different session, so the appendix is the actual record of what you used, not just a courtesy.
MLA style: how to cite ChatGPT
The MLA Style Center’s guidance, published by the Modern Language Association, deliberately does not treat the AI tool as an author. Instead it builds the works-cited entry around MLA’s standard container model, with the prompt itself as the “title of source.”
MLA works-cited entry
“Describe the symbolism of the green light in The Great Gatsby” prompt. ChatGPT, 13 Feb. version, OpenAI, 8 Mar. 2025, chatgpt.com/.
Structure, in MLA’s core-elements order: a quoted, and if needed shortened or generalized, description of your prompt as the “title of source,” followed by ChatGPT (the container, italicized) → the specific version → OpenAI as publisher → the date you generated the content → the URL. MLA is explicit that you don’t need to quote your entire prompt verbatim if it was long or conversational — a general, accurate description in quotation marks is acceptable, since these exchanges are iterative and rarely map cleanly onto a single “title.”
MLA in-text citation
If you name the tool directly in your sentence (“When prompted to explain X, ChatGPT responded that…”), MLA’s guidance says a separate parenthetical citation is often unnecessary, since the works-cited entry supplies the detail. Where an in-text citation is needed, cite by the first element of your works-cited entry, per standard MLA parenthetical practice.
MLA’s core principle
MLA states plainly: do not treat the AI tool as an author. That single line is the throughline of its whole approach — the tool is a container/publisher-credited source you consulted, not a contributor with intellectual accountability for the work, which is the same line ICMJE, COPE, and every major publisher draw when the question shifts from citation to authorship. See Can AI Be Listed as an Author? for that separate question in full.
Chicago style: how to cite ChatGPT
The Chicago Manual of Style’s Q&A guidance takes the most restrictive position of the three: it recommends treating AI-generated content similarly to a personal communication, and explicitly advises against including it in a formal bibliography or reference list at all, unless you can link to a publicly available, login-free copy of the exchange (for example, via a shared-conversation tool). Because a standard chatgpt.com session URL requires the reader to be logged into your account to view it, Chicago’s default is: cite AI output only in the text or in a note, not in the bibliography.
Chicago notes-bibliography system
Footnote or endnote: Text generated by ChatGPT, OpenAI, March 7, 2025, https://chatgpt.com/.
If your prompt isn’t already stated in your running text, Chicago suggests folding it into the note itself: ChatGPT, response to “[your prompt],” OpenAI, March 7, 2025.
Chicago author-date system
Parenthetical, placing whatever information isn’t already stated in the surrounding text: (ChatGPT, March 7, 2025)
What Chicago expects beyond the citation itself
Chicago’s guidance also asks writers to: always credit the AI tool whenever its text is reproduced (even loosely); note explicitly if you’ve edited the AI’s output beyond basic formatting; and understand that in Chicago’s framing, ChatGPT effectively functions as the “author” of the quoted text with OpenAI as its publisher/sponsor — language, not a claim about legal or intellectual authorship.
Style comparison at a glance
| Question | APA | MLA | Chicago |
|---|---|---|---|
| Goes in the formal reference list / bibliography? | Yes, always | Yes, always (works-cited) | Generally no — text/note only, unless a public link exists |
| Who/what is credited as “author”? | OpenAI (the company) | No author credited — prompt is the title, tool is the container | ChatGPT functions as author-in-name, OpenAI as publisher |
| Must you quote or describe your prompt? | Yes, in text, plus the generated excerpt | Yes, as (or within) the works-cited “title of source” | Yes, in the note if not already in your text |
| Recommended for long output | Full transcript in an appendix, cited in text | Not specifically addressed — general prompt description is enough | Not specifically addressed beyond the no-bibliography default |
All three converge on the same underlying logic even though the mechanics differ: name the specific tool and its maker, record when you used it, and preserve or point to what it actually said — because unlike a journal article or a book, nobody else can independently retrieve your exact output later.
Why AI output can’t be cited like a normal source
A journal article has a DOI. A book has an ISBN and library holdings. A webpage has a URL that (usually) resolves to the same content for every reader. A ChatGPT conversation has none of that: the same prompt, run again, can produce a materially different response, the underlying model itself is periodically updated or retired, and a shared session link is typically tied to your account. This is the reason every major style guide places AI output in the same conceptual bucket as a personal communication or an unrecoverable interview — a real exchange happened, but it isn’t independently verifiable or retrievable by your reader the way a published source is.
That’s also why all three style guides converge on the same two practical requirements regardless of formatting differences: record the exact prompt (or an accurate, if abbreviated, description of it), and preserve or reproduce enough of the actual output that a reader can see what you saw, rather than trusting the citation alone. When output is long, an appendix with the full transcript, clearly labeled with the date, model version, and exact prompt, is the most defensible practice across all three styles.
Citing AI is not the same as disclosing AI use
It’s easy to conflate these two obligations because they both involve acknowledging AI, but they serve different functions and usually have different rules attached:
- Citation is a source-attribution mechanic: you generated or quoted a specific piece of text or output and you’re crediting where it came from, following your target style guide’s format.
- Disclosure is a research-integrity and editorial-policy requirement: most major publishers (ICMJE-aligned journals, Elsevier, Springer Nature, and others) now require a separate statement, usually in the methods or acknowledgments section, describing whether and how generative AI was used anywhere in preparing the manuscript — drafting text, editing, generating code, translating, or summarizing literature — independent of whether any AI output is directly quoted or cited anywhere in the paper.
You can have a paper that requires a disclosure statement but contains zero AI citations (because you used AI only to polish prose you wrote yourself, with no quoted output), and, less commonly, a paper that quotes and cites an AI tool’s output without necessarily triggering every publisher’s disclosure threshold for lighter uses. Check your target journal’s specific policy rather than assuming one obligation satisfies the other. For ready-to-adapt disclosure language, see AI Disclosure Statement Examples; for the separate, adjacent question of whether an AI tool can ever appear as a listed author (it cannot, under ICMJE, COPE, and every major publisher’s current policy, because authorship requires accountability an AI tool cannot bear), see Can AI Be Listed as an Author?. The broader policy landscape across individual publishers is mapped in Journal and Publisher Policies on Generative AI in Manuscripts, and jurisdiction-level disclosure requirements are covered in AI Disclosure Laws.
The real risk: ChatGPT invents citations
Formatting a citation to ChatGPT correctly solves an attribution problem. It does not solve a much more serious one: generative AI tools are well documented to fabricate plausible-looking references — real-sounding author names, real journal titles, realistic-looking volume/issue/page numbers, and even syntactically valid but nonexistent DOIs — when asked to supply citations or a literature list. This behavior is a direct consequence of how large language models generate text: a model predicts statistically likely sequences of words based on patterns in its training data, it does not look up and verify sources against a live database the way a citation manager or a librarian does, so a fabricated reference is not a rare glitch but a structurally expected failure mode of the technology. CASRAI’s Dictionary has this catalogued as hallucination, and the resulting fake reference itself as a fake citation — worth reading both if you’re building institutional guidance on this.
The practical, non-negotiable obligation that follows: every reference an AI tool suggests must be independently verified against the actual source before it enters a manuscript — confirmed to exist, confirmed to say what the AI claimed it says, and confirmed to carry the exact metadata (authors, year, title, DOI) cited. This is not materially different from an author’s existing duty to have actually read and correctly represented any source they cite; AI-suggested references just make it far easier to skip that step by accident, because the fabricated citations are specifically designed (by the model’s own generation process, not by any intent to deceive) to look indistinguishable from real ones at a glance. Checking a DOI resolves, checking that a journal issue and page range actually exist, and pulling the actual article to confirm it says what was claimed are the minimum verification steps before an AI-suggested citation is used. This is also a distinct issue from whether AI-generated prose counts as plagiarism — see Is ChatGPT (AI-Generated Text) Considered Plagiarism? for that separate question.
Frequently asked questions
How do I cite ChatGPT in APA format?
Reference-list entry: OpenAI. (2025). ChatGPT (Aug 2025 version) [Large language model]. https://chatgpt.com/. In-text: (OpenAI, 2025) or narrative OpenAI (2025). Quote your prompt and the relevant output in text, or place a full transcript in an appendix for longer exchanges.
How do I cite ChatGPT in MLA format?
Build the works-cited entry around a quoted description of your prompt as the title of source, followed by ChatGPT as the container, the version, OpenAI as publisher, the date, and the URL — for example: “[prompt description]” prompt. ChatGPT, [version], OpenAI, [date], chatgpt.com/. MLA explicitly does not treat the tool as an author.
Do I need to cite ChatGPT in Chicago style?
Generally not in a formal bibliography — Chicago recommends citing AI-generated content in a footnote or in-text parenthetical only, similar to a personal communication, unless you can link to a publicly accessible, login-free copy of the exchange.
How do I cite AI in APA when it’s not ChatGPT specifically?
Use the same structure with the actual company and tool name swapped in — for example, Google. (2025). Gemini (version) [Large language model]. URL, or Anthropic. (2025). Claude (version) [Large language model]. URL. The bracketed description and in-text format stay the same.
Is citing ChatGPT the same as disclosing that I used AI?
No. A citation attributes a specific piece of quoted or generated text. A disclosure statement, which many journals now require separately, describes how AI was used anywhere in preparing the manuscript, whether or not any AI output is directly quoted. See the dedicated section above and AI Disclosure Statement Examples.
Why does ChatGPT sometimes invent fake citations?
Because a language model generates statistically plausible text rather than retrieving verified facts from a database, it can produce references with realistic author names, journal titles, and DOIs that do not correspond to any real publication. Always independently verify any AI-suggested citation against the actual source before using it. See hallucination and fake citation.
Formats verified directly against APA Style’s official blog guidance (apastyle.apa.org), the MLA Style Center (style.mla.org), and the Chicago Manual of Style Online Q&A (chicagomanualofstyle.org) as of August 2026. All three bodies have revised this guidance since generative AI tools became widely used and are likely to revise it again as tools and interfaces change — if your institution or target journal specifies a particular edition or an updated version of any of these guides, defer to that current primary source over this summary.
