How to Cite AI Tools (ChatGPT, Claude, Gemini) in APA, MLA, and Chicago

To cite ChatGPT or another AI tool, name the company, the model, the date you used it, and a shareable link to the conversation, formatted to whichever style your course uses. APA, MLA, and Chicago all have official rules for this now, and they don't agree on the details, so the format below is taken straight from each style's own guidance. The harder question comes first, though: whether you need to cite the AI at all, and whether the thing you should be citing is the chatbot or the real study it pointed you to.

First: do you even need to cite the AI?

You cite an AI tool when you use what it produced: text you quoted or paraphrased, an image it generated, an outline it drafted, data it summarized. That's the baseline all three style guides share. Using a chatbot only to brainstorm, to check your grammar, or to explain a concept you then wrote up yourself is usually closer to using a dictionary or a calculator, and often doesn't need a citation at all.

But the rule that overrides every format below is this: policies vary, so check yours first. Some instructors ban AI in graded work entirely. Some allow it but want a short usage statement instead of a formal citation. Some only care when AI shaped the actual content, not when it fixed a comma. Your syllabus, assignment brief, or department's academic-integrity page is the real authority here; the style guide only tells you how to format the citation once you've decided you need one. Getting the APA format perfect won't help if your course didn't allow the tool in the first place.

The format, by style

Here is the short version of each official format, with a correct in-text citation and a reference-list entry. Read the per-style sections below for the details that trip people up.

APA

APA's current guidance treats the AI tool's developer as the author, because in APA an author is whoever is "responsible for" a work, not necessarily who typed it. A reference to a specific chat follows the usual author, date, title, source shape. The official APA template is: Company. (year, month day). Title of the chat in italics [Generative AI chat]. Tool name and model. URL of the chat. So a Claude conversation would read:

In text, you cite the author and year like any other source: parenthetically as (Anthropic, 2026), or narratively as Anthropic (2026). APA's older, still-cited "How to cite ChatGPT" guidance showed a slightly different shape, with the version in parentheses after the title and a "[Large language model]" descriptor; the newer template above is what APA now recommends, and it handles shareable chat links cleanly. Either way, APA says to describe your prompt in the running text rather than cram it into the reference, and to put the full text of long responses in an appendix so a reader can see exactly what the tool produced.

MLA

MLA builds every citation from its template of core elements, which is why it adapted to AI without inventing a new rule. Per the MLA Style Center, you describe what the AI generated (often by quoting your prompt) as the title of the source, name the tool as the container, give the model in the Version element, the company as publisher, the date you used it, and a stable shareable URL as the location. A worked entry:

The in-text citation points back to the first words of that entry, so it reads ("Describe the theme"). MLA's August 2025 update made two things explicit: prefer a stable, shareable link to the exact conversation, and falling back to a general link to the tool only when no shareable URL exists; and always put the specific model name or number in the Version element. MLA also flags the trap this whole article is about: if the AI's answer leans on sources, click through, read them, and cite those sources directly rather than the AI's summary of them.

Chicago

Chicago handles AI more like a personal communication than a published work. Per the Chicago Manual of Style Q&A, you credit the tool whenever you use its words, but in most writing you can do that in the text itself ("the following was generated by ChatGPT") or in a footnote, rather than a bibliography entry. A note looks like:

If you want the prompt in the note, Chicago's variant is: ChatGPT, response to "Explain how to make pizza dough from common household ingredients," OpenAI, March 7, 2023. For author-date papers, the parenthetical is simply (ChatGPT, March 7, 2023). The key Chicago rule: don't put the AI in a bibliography or reference list unless the conversation has a publicly available link, because a reader can't retrieve a private chat any more than they can retrieve your phone call. And if you edited the generated text, say so, either in the text or at the end of the note.

Describe the prompt, the tool, and the date

Across all three styles the same three pieces of information do the real work, so capture them as you go rather than rebuilding them later. Record the exact prompt you used, because every style wants it described and MLA puts it in the entry itself; paraphrasing your prompt afterward isn't the same thing. Record the specific tool and model, not just "ChatGPT" but "ChatGPT GPT-4o" or "Claude Sonnet 4," since the same product behaves differently across versions and a reader needs to know which one answered. And record the date you ran it, because these tools change week to week and the output isn't reproducible; the date is what tells a reader the answer reflects the model as it stood that day. If your tool offers a shareable link to the conversation, grab it, every style now prefers one.

The point everyone misses: the AI is not the source

Here is the conceptual move that matters more than any format. An AI tool is not a source for facts. It's a way of generating text, including text that looks like a fact with a citation attached. So when a chatbot tells you "a 2019 study found that sleep improves memory," the chatbot is not your evidence for that claim. The study is, if it exists.

That changes what you cite. If the AI pointed you to a real study, your job is to open that study, read the part that backs your sentence, and cite the study, not the conversation that mentioned it. You verify it the same way you'd verify any reference: find it by title, resolve its DOI, confirm the authors are real, and check it actually says what you're about to claim. We walk through that in how to check if a citation is real. The reason this matters so much with AI is that chatbots routinely invent studies that don't exist, complete with plausible authors and a correctly formatted DOI that leads nowhere, which is its own deep problem covered in why AI makes up citations.

So in practice you cite the AI tool itself only in the narrow case where the AI's own output is the thing you're discussing: you quoted what it wrote, you're analyzing its behavior, you used an image it generated. The far more common case is that the AI helped you find or phrase something, and the real citation is a paper you opened and read. If you can't find that paper, you don't have a citation; you have a sentence that needs a different source, or no support at all. Getting a chatbot to surface real, openable sources in the first place is a skill of its own, and we cover the prompts and settings that help in how to get ChatGPT to cite real sources.

Where CiteOwl fits

This is the gap CiteOwl is built to close. It doesn't write a claim and then guess at a citation. It searches for real papers, reads them, and only writes a sentence it can attach to a source it actually retrieved, showing you the quote behind each one. So most of the time, the thing you end up citing isn't the AI at all; it's a real study with a working link, the kind you'd cite normally in APA, MLA, or Chicago. You'd still credit CiteOwl as a tool if your course requires disclosing AI use, the same way you'd note any writing assistant, but the references in your bibliography point at genuine scholarship you can open. That's the honest version of "AI helped me write this": the help was real, and so are the sources.