What happens when a student uses ChatGPT to outline a paper, then can’t find a single example of how to cite it? That confusion is now common in classrooms worldwide. Generative AI arrived faster than most style guides could respond, and the rules are still shifting. The good news: every major citation system has now staked out a position, and once you understand the shared logic, citing AI tools becomes far less mysterious. (Citation)
This guide walks through what APA, MLA, Chicago, and IEEE currently expect, where journal policies override style guides, and the practical mistakes that trip people up most often.
Why citing AI tools is harder than citing a book
A book has an author, a publisher, and a stable page you can return to. AI output has none of those things in the traditional sense. The same prompt can produce different text on different days, and readers can’t retrieve your exact output even if they run the identical query.
That instability drives almost every rule that follows. Most style authorities treat AI-generated text as a non-recoverable source, meaning readers cannot independently check it. Instead of a normal reference entry, they ask for disclosure: name the tool, the version, the developer, the date, and the prompt you used.
The Committee on Publication Ethics (COPE) settled one big question early. In its 2023 position statement, COPE ruled that AI tools cannot be listed as authors because they cannot take responsibility for the work or consent to publication. The International Committee of Medical Journal Editors (ICMJE) reached the same conclusion, adding that AI use should be described in the methods section or acknowledgments. If you want to see how quickly tooling around AI has matured, the same pace shows up in writing workflows covered in guides to the best ai writing tools for content creators.
The core elements every citation style wants
Before diving into individual formats, here is the pattern that nearly all major styles converge on. Get these five pieces and you can adapt to almost any format:
- Tool name and version (for example, ChatGPT, GPT-4)
- Developer or publisher (OpenAI, Google, Anthropic)
- Date accessed or date generated
- The prompt or a description of it
- A disclosure statement in your methods, acknowledgments, or notes
What differs between styles is mostly where this information lives: in the reference list, in-text only, or in a note. That distinction matters more than most students realize, because putting AI output in the wrong place can cost marks even when the details are correct.
APA style: in-text citation, usually no reference entry
The APA Publication Manual, 7th edition, was published in 2019, well before ChatGPT launched in November 2022. That means APA’s AI rules come from supplementary guidance on the APA Style blog rather than the manual itself, and the blog has revised its advice more than once.
APA’s general approach treats AI output as a non-recoverable source. You cite it in-text with the model version and developer, something like (OpenAI, 2023), and you generally do not add a reference list entry unless the output is retrievable by readers. If you paste a full transcript into an appendix, that changes things, because now the material is recoverable.
A practical tip that saves headaches: record the exact prompt and the date you ran it the moment you generate the text. Reconstructing that a month later is nearly impossible, and APA guidance asks for the prompt in the reference entry when one is required.
MLA style: the prompt becomes the title
MLA moved faster than most. Its Style Center published formal guidance on citing generative AI in 2023, and the approach is distinctive: the description of the prompt becomes the title of the source, and the AI tool is treated as the author and container.
A typical MLA entry looks roughly like this in structure: a quoted description of the prompt as the title, the tool name as container, the version, the publisher, the date, and the URL if available. MLA also leans toward treating AI output as a source you consulted rather than a traditional authored work.
One thing MLA guidance stresses: you should still verify facts the AI gives you, because citing the tool does not transfer responsibility for accuracy to the tool. That point applies to every style in this article, and it connects to broader questions about how these systems generate text, explored in this explainer on how does ai image generation work.
Chicago style: notes and acknowledgments over bibliography
The Chicago Manual of Style addressed AI citation through its Q&A service in 2023. Chicago generally advises citing AI-generated content in notes or acknowledgments rather than the bibliography, again because readers cannot retrieve the output.
For humanities papers using notes-bibliography format, that means a numbered note describing the tool, version, developer, date, and prompt. For author-date style, an in-text mention plus a note usually suffices. Chicago’s reasoning is consistent: a bibliography entry implies a stable, findable source, and AI output is not that.
IEEE and the sciences: disclosure first
IEEE guidance, developed after the other major styles, generally treats AI-generated text as a non-recoverable source and discourages listing it as a formal reference. In practice, engineering and computer science papers disclose AI use in the methods or acknowledgments and describe the prompt and tool in text.
Scientific publishers often go further than any style guide. The Nature portfolio announced in January 2023 that LLM use must be documented in the methods section and that LLMs cannot be listed as authors. ICMJE recommendations say the same. When a journal’s policy conflicts with a style guide, the journal wins, every time.
Side-by-side comparison
| Style | Where AI citation goes | Reference list entry? | Prompt included? |
|---|---|---|---|
| APA (7th ed. + blog guidance) | In-text, sometimes reference | Usually not, unless retrievable | Yes, when a reference is required |
| MLA | Works cited entry | Yes | Yes, as the title |
| Chicago | Notes or acknowledgments | Generally no | Yes, in the note |
| IEEE | Methods or acknowledgments | Discouraged | Yes, described in text |
| Nature / ICMJE journals | Methods section | No, and no AI authorship | Yes, described |
Where people go wrong most often
A few mistakes show up again and again in real submissions.
Listing the AI as an author. This violates COPE and ICMJE guidance and gets papers rejected outright at many journals. The tool is never an author.
Forgetting the version number. “ChatGPT” alone is not enough. GPT-3.5, GPT-4, and later versions behave differently, and the version is part of what makes the citation meaningful.
Skipping the prompt. The prompt is often the only thing a reader can actually evaluate. Omitting it removes the one reproducible element of the whole citation.
Citing a style guide’s old advice. Guidance from 2023 has been revised repeatedly. Check the official style site for the current wording rather than trusting a screenshot from two years ago.
Assuming one style’s rules apply everywhere. APA’s in-text approach and MLA’s works-cited approach are genuinely different. Mixing them produces citations that satisfy nobody.
Ignoring disclosure requirements. Even when a citation is technically correct, many courses and journals require a separate statement describing how AI was used. A citation is not a disclosure.
Practical workflow that actually works
Experienced writers tend to follow a repeatable routine rather than improvising each time:
- Log everything at the moment of use. Save the tool name, version, date, and exact prompt in a running notes file.
- Check your target style’s official site first. Confirm the current guidance before formatting anything.
- Check your journal or course policy second. It can be stricter than the style guide and always takes precedence.
- Decide whether the output is retrievable. If you include the full transcript in an appendix, you may need a reference entry; if not, in-text or notes usually suffice.
- Add a disclosure statement. Describe which parts of the work used AI and how.
- Verify every fact the tool produced. Citation does not excuse errors.
That last step matters more than it sounds. Researchers have repeatedly found that language models generate confident-sounding citations that do not exist, so treat any reference the AI hands you as unverified until you have seen the actual source. Writers who lean on tools for drafting often pair them with checkers, similar to how people use the best ai paraphrasing tools for writers to rework text while keeping the meaning intact.
What about citing AI in group projects and theses?
The same rules apply, with one addition: scope. In a thesis, examiners usually want to know exactly which chapters involved AI assistance and for what purpose, such as brainstorming, editing, or code generation. A single vague sentence in the acknowledgments often is not enough. Be specific about the task, the tool, and the extent of use. (How to cite ChatGPT)
For collaborative work, keep a shared log so every contributor cites consistently. Mismatched citation styles across sections is one of the most common reasons a group submission gets flagged.
Conclusion
Citing AI tools is not really about memorizing one format. It is about being transparent when you use a source that readers cannot retrieve. Name the tool and version, credit the developer, record the date and prompt, and disclose how you used it. APA, MLA, Chicago, and IEEE all orbit that same principle, even when they place the information in different spots. Check the official style site for current wording, follow your journal or instructor’s policy when it is stricter, and verify every fact the tool produced. Do those things consistently and AI becomes just another source you handled responsibly, which is exactly what academic integrity asks for.
Frequently asked questions
Can I list ChatGPT as a co-author on a paper?
No. COPE and ICMJE both state that AI tools cannot be authors because they cannot take responsibility for the work, consent to publication, or be held accountable. You may describe AI assistance in the methods or acknowledgments, but authorship credit belongs only to people.
Do I need a reference list entry for AI-generated text in APA style?
Usually not. APA treats AI output as a non-recoverable source, so an in-text citation with the model version and developer is typically enough. If you include the full output in an appendix so readers can retrieve it, a reference entry may be appropriate. Check the APA Style blog for the current rule.
How do I cite an AI tool in MLA style?
MLA asks you to use a description of your prompt as the title, with the AI tool as the author and container, followed by the version, publisher, and date. The exact format is documented on the MLA Style Center, which published formal guidance in 2023 and has updated it since.
What should I do if my journal’s policy conflicts with my citation style?
Follow the journal. Publisher and journal policies, such as the Nature portfolio’s requirement to document LLM use in the methods section, override general style guides whenever the two disagree. Many journals also require a separate disclosure statement beyond the citation itself.
Should I include the exact prompt in my citation?
Yes, whenever the style asks for it. The prompt is often the only reproducible element of an AI citation, since the same query can produce different output over time. Recording the prompt and date at the moment you generate text is far easier than reconstructing them later.
Does citing an AI tool mean I don’t have to check its facts?
No. Citation is transparency, not endorsement. Language models can produce inaccurate claims and fabricated references, so every fact and source they generate needs independent verification before it appears in academic work.
