How to Cite ChatGPT: APA, MLA, Chicago, IEEE and Disclosure Rules
APA, MLA, Chicago and IEEE each answer a ChatGPT citation differently: as an author, a container, a note, or no citation at all. Here is what each one actually requires, and where citation stops and disclosure begins.
Both of these references are valid, but they refer to exactly the same conversation. The first one is a footnote with the text "ChatGPT, response to prompt, OpenAI, 14 March 2026". The second is a line in the reference list "(OpenAI, 2026)". Both citations look completely different. Both come from different style guides, which do not agree about what kind of object a ChatGPT exchange is (author, container, note, or no citation covers it). It depends on your style guide.
Search how to cite ChatGPT and the answer depends entirely on which manual is asking. APA treats the company behind the model as an author. MLA refuses to call the tool an author at all and builds the entry around your prompt instead. Chicago keeps the whole exchange out of the bibliography. IEEE hasn't published a citation format for it in the first place, only a rule about disclosing that you used one.
That question contains two different things you need to do. A style guide only tells you one of those. When you're done writing with a source, whether it's word for word, an idea, or a bit of output from the source, then you cite the source. The other thing you can be asked to do is to say that you used a tool to create the work, and that you did so with disclosure. If you skip over this distinction, you'll either cite something that didn't need a disclosure line, or you won't disclose anything when you should've cited. Rather than one being tacked on to the other, both are walked through, style by style.
ChatGPT is the example every major manual reached for first. In fact, when APA, MLA and Chicago each wrote about generative AI for the first time, they all used ChatGPT as the example. This is why its citation style is the most well-established of any AI tool's. If you change the company name to Claude or Gemini or Copilot or whatever launches next, it's still the same underlying rules.
| Style | How it frames the exchange | Where the citation lives | In-text form |
|---|---|---|---|
| APA 7 | Algorithm output, credited to the company that built it | Full reference list entry | (OpenAI, 2023) |
| MLA 9 | A titled work inside a container, built from your prompt | Works Cited entry | A shortened prompt in quotation marks |
| Chicago 18 | Content generated on request, named in a note | Footnote or endnote only, not the bibliography | No parenthetical form; the note carries it |
| IEEE | Not addressed by an official citation template | Acknowledgments disclosure, not a reference entry | None published |
Everything else in your reference list, the books, journal articles and reports that are not an AI exchange at all, still follows one fixed rule per style rather than four competing ones. A citation generator built from real publisher metadata handles that ordinary case automatically; a ChatGPT exchange is the one entry in the list no database can look up for you, which is exactly why it needs its own set of rules.
Do You Need to Cite ChatGPT, or Just Disclose It?
Cite ChatGPT if any part of the text that it produced (including exactly-worded ideas) appears anywhere in your writing, whether quoted or paraphrased. Mention your use of it separately if it helped you think, translate, or edit. But none of its own words ended up in the final version. Plenty of assignments need both at once: a disclosure statement covering the help you had, and a citation for the one paragraph you quoted directly.
I don't think anyone would dispute this claim, but just in case you need evidence, here's what APA says about this kind of reference (see the second sentence): You should provide details of how you used the tool in your method section or introduction. This might include describing the prompt you gave to the tool and any part of the output from the tool that you included in your research article. APA doesn't say anything like this in a reference entry. It also doesn't say anything about the right way to format a citation. This is necessary because you need more information than a reference entry provides if you want someone to know how much of the thought in your paper came from you.
Why the Style Guides Don't Agree on One Format
Four manuals looking at the same technology came to four different conclusions about what it basically is. APA decided that a ChatGPT response is algorithmic output; i.e., it's closer to a dataset than to a conversation, and therefore gets a full reference entry (instead of the personal communication treatment APA gives to an email or an interview). MLA decided that the tool can't play the author role at all. So its entry focuses on what you actually wrote rather than what answered you.
Chicago and IEEE go even further. Chicago says that a ChatGPT exchange is more like an interview transcript, worthy of a note but not of a permanent spot in the bibliography because another reader can't get the exact exchange you had. IEEE hasn't weighed in on a citation format at all. It says what matters is telling the world that your work involved the use of a tool, not formatting an entry for it. That makes IEEE the clearest example of a style body simply declining to answer this question.
How to Cite ChatGPT in APA and MLA
APA's reference entry opens with OpenAI as the author, followed by the full date of the conversation, the chat's own title italicized, a bracketed [Generative AI chat] label, then the model name and a link to the shared chat. The in-text form stays parenthetical or narrative like any other APA citation: (OpenAI, 2026), with ChatGPT named in your sentence rather than in the brackets. APA rewrote this in September 2025, moving from citing the tool to citing the conversation, and marked its 2023 post archival, so guides still showing ChatGPT in italics with a version in parentheses are teaching the retired form. A ChatGPT citation generator built on the current guidance will ask for the chat title and the share link, which is the quickest way to tell whether the one you are using has kept up.
MLA reverses almost everything about that structure. You start by giving a truncated version of your own prompt (in quotes) as a stand-in for a title, followed by ChatGPT as the italicized container, the version of the model, OpenAI as the publisher, the date, and a link to the actual exchange (where the interface gives you one). MLA's own current guidance recommends naming the model version as specifically as the interface allows, GPT-4o rather than just ChatGPT, since a vaguer label cannot be traced back to a specific exchange.
Humanize your own paper
Transform your AI-assisted text and make it sound human, without touching important words or citations.
How Chicago and IEEE Handle a ChatGPT Citation
In Chicago style, this reference lives only in a footnote or endnote: I credit the content to ChatGPT, describe or quote the prompt, credit OpenAI as the publisher, and give the full date. None of it repeats in the bibliography unless you can point to a publicly available link for that exact exchange. The bibliography remains reserved for sources that another reader could actually go and find.
IEEE bypasses that issue by asking another one: does your acknowledgments section mention the AI system you used, what part of the paper it worked on, and how much it really did? That disclosure is mandated by IEEE's own editorial policy even if you also try to create a citation for the tool, and excludes grammar or copyediting assistance. The same acknowledgments-first logic now shows up in many journals far outside engineering. So perhaps IEEE's disclosure-only approach will prove more lasting even without a citation template to go along with it. But the journal that does provide a citation template goes further: they need that authors use their template for any references to tools used in the research, including ones not owned by them.
How Do You Disclose AI Use Beyond a Citation?
This is most clearly answered by medical and biomedical journals following the ICMJE recommendations, which most Vancouver-style journals do. At submission, authors are asked to suggest if they've used an AI tool. Authors must disclose at submission whether an AI tool was used at all, and where they describe that use depends on what it did: writing assistance goes in the acknowledgments, while AI used for data collection, analysis, or generating a figure gets described in the methods section instead.
The detail every one of these policies shares is what it refuses to allow: none of them will let an AI system stand in as an author. ICMJE puts it as plainly as any body has: authors should not list or cite an AI tool as an author under any circumstances. A citation credits a source, and authorship claims responsibility for the work in a way no model can hold. Working out the line between editing your own argument and submitting someone else's, the harder judgment call disclosure exists to cover, is its own subject.
What a Citation Does Not Verify
A correctly formatted entry tells us nothing about whether the reference it contains refers to something that exists. Predicting text that looks like a reference, including a plausible author, journal and DOI, is what a language model is doing when it generates a reference list. Some of this generated text will refer to things that have never been published. Formatting a reference correctly is distinct from checking that the reference resolves to a real scholarly publication, and it must be done first. An AI citation checker that resolves each entry against a real scholarly database is the only way to catch that before a supervisor or reviewer does.
In an AI citation, the version field is more important than an edition number will ever be in a book reference. When a prompt has been run once before and again after an update, the wording might change. Note down exactly what your interface displayed when it answered, ChatGPT's version picker, Claude's model menu, Gemini's version selector, along with the date. An unversioned citation cannot be tracked back to the exchange it claims to describe. Skip the version field and the reference won't point at anything anymore.
Citing Claude, Gemini and Other AI Tools
None of the four questions above change when the company name does. Work out whether the manual you are using grants the tool authorship, builds the entry around your prompt instead, keeps the citation to a note, or asks only for disclosure, then apply that same pattern to whichever company and product actually answered you. An AI citation generator that presets ten model families, ChatGPT, Claude, Gemini and Copilot among them, saves you the lookup once you already know which pattern your style requires.
A properly cited ChatGPT conversation and a fully annotated one do not answer all our questions. Whether this text in which we've placed those citations sounds like the person who cited those references also wrote the rest of the page remains an open question. That's a different task than formatting citations. This is what an AI humanizer for academic writing does: going through the phrasing and pacing to make sure that everything sounds like it comes from the same person.
Frequently Asked Questions
Usually not as a formal citation. Most instructors and journals still want a disclosure statement even so. Citation covers wording, ideas, or generated text that actually entered your draft. Brainstorming that shaped your thinking without contributing any of its own phrasing sits closer to disclosure territory, the same category as thanking a colleague for a conversation that helped you settle on an argument.
Content planner and copywriter at TextPulse. Sara runs the blog day to day, from planning and drafting through to publishing. She writes the practical guides: clear explanations of academic writing problems, aimed at the person who actually has to hand something in.
