How to cite AI?
Cite AI as the company that made it, not the AI. In APA: OpenAI. (2026, February 10). Title of chat [Generative AI chat]. ChatGPT. URL. MLA uses a Title-of-source style with the prompt described. Chicago says AI is not an author and often prefers a note. All three require you to disclose AI use and verify every fact yourself.
Why — the first-principles explanation
The reason citation styles struggled with AI is that a citation has one core job: let a reader retrieve the same source you saw. That works for a book because every copy of the book says the same thing forever. It breaks for AI because the same prompt can produce a different answer tomorrow, for a different user, on a different model version. In library terms, an AI chat is a non-recoverable source — closer to a personal conversation than to a published article.
Styles resolved this in two ways. First, they assign authorship to the company, not the model. ChatGPT did not choose to write anything; OpenAI built and released the tool. So OpenAI is the author, and ChatGPT is the source. Second, they lean on the fact that most AI tools now offer a share link that freezes one specific conversation at a public URL. That link restores retrievability, which is why APA now builds its reference around a shareable chat with a date and URL. If your chat has no share link, the fallback is to treat it like personal communication — cite it in text only, and put the full prompt and output in an appendix so a reader can at least see what you saw.
The styles differ mostly in taste. APA formats the tool as a retrievable work with an author-date-title-source shape. MLA leans on its universal template, describing the AI's role and the prompt in the entry. Chicago is the most conservative: it holds that AI is not an author in any meaningful sense, so it often keeps AI out of the bibliography entirely and handles it in a footnote or in the text, because a bibliography is meant to list sources others can consult.
The part everyone skips matters most. APA states plainly that you must review AI output, verify its accuracy, and take responsibility for it. Citing an AI does not transfer blame to the AI. If the model invents a study and you cite the chat, you published a fabrication — the citation just documents where you got it. And citation is not permission: your instructor's or journal's policy decides whether AI use is allowed at all. A perfectly formatted citation on an assignment that banned AI is still academic misconduct.
An example that makes it click
Citing an AI is like quoting a very well-read stranger you met on a train. You don't write "according to the train," and you don't write "according to a man." You write who they were and when you spoke — because that conversation happened once, and nobody else can go hear it.
Now suppose that stranger told you a fact and you put it in your report. If the fact was wrong, nobody blames the stranger. It's your report. Saying "a man on a train told me" doesn't rescue you — it just shows you knew your source was weak and used it anyway. That's exactly why every style guide pairs "here's the format" with "go check what it told you."
How to do it
- Check the rules first. Confirm whether your instructor, journal, or employer allows AI use at all — citation format is irrelevant if the use is prohibited.
- Get a share link. In ChatGPT, Claude, or Gemini, use the share function to generate a permanent URL and title for that specific chat.
- Identify the company as author: OpenAI for ChatGPT, Anthropic for Claude, Google for Gemini.
- APA format: OpenAI. (2026, February 10). Top 10 writing topics for high school graduates [Generative AI chat]. ChatGPT. https://chat.openai.com/share/...
- MLA format: use the universal template — describe the prompt, name the tool as Title of Source, add the version/company as Publisher, and the date and URL.
- Chicago format: prefer a footnote describing the tool, prompt, and date; Chicago generally keeps AI out of the bibliography since readers cannot consult it.
- Describe your method in the text — say what you asked the AI to do and how you used the output (APA suggests the Method section for research papers).
- If there is no share link, cite it in text as you would personal communication and reproduce the full prompt and response in an appendix.
- Verify every factual claim, quote, and reference the AI produced against a primary source before it goes in your work.
Key facts
- APA assigns authorship of AI output to the company (e.g., OpenAI), not to the model, and formats a shared chat as: Author. (Year, Month Day). Title [Generative AI chat]. Tool. URL.
- APA explicitly requires authors to review AI-generated text, verify its accuracy, and take responsibility for it before submission or publication.
- MLA advises citing generative AI with its universal template, documenting the prompt and describing the AI's role in the work.
- The Chicago Manual of Style (18th ed.) treats AI as not an author and generally recommends handling it in a note or in text rather than in the bibliography.
- APA recommends describing AI use in the Method section (or an equivalent methods description) of a research paper.
- Citation does not equal authorization: APA notes that submitting AI-generated work as your own remains unethical and can be academic misconduct.
▶ The 60-second explainer (script)
How do you cite AI? Rule one: cite the company, not the AI. Not "ChatGPT says." ChatGPT didn't decide to write anything — OpenAI built it. So OpenAI is the author, ChatGPT is the source. Here's the APA shape: OpenAI, paren, year comma month day, title of the chat, bracket Generative AI chat, ChatGPT, and the URL. That URL matters, and here's why. A citation exists so a reader can go see what you saw. That works for a book — every copy says the same thing forever. It breaks for AI, because the same prompt gives different answers tomorrow, to a different user, on a different model. So use the share button. It freezes one conversation at a public link and makes it retrievable again. No share link? Then treat it like a private conversation: cite it in text, and dump the full prompt and response into an appendix. MLA does the same job with its universal template, describing your prompt. Chicago is the strict one — it says AI isn't an author, so keep it in a footnote, out of the bibliography. And the part everybody skips. APA says flat out: you must verify AI output and take responsibility for it. Citing the AI does not move the blame. If the model invents a study and you cite the chat, you published a fake — you just documented where you got it. One more: citing isn't permission. If the assignment banned AI, a perfect citation is still misconduct.
What authoritative sources say
People also ask
Can I list ChatGPT as a co-author?
No. Major style guides and journal policies reject AI authorship because an author must be able to take responsibility for the work and approve it. AI cannot do either.
Do I cite AI if I only used it for grammar or brainstorming?
Usually not for basic spellcheck or grammar tools. Disclose it when AI shaped the substance — generating text, ideas, code, or analysis. When unsure, ask your instructor or editor, and default to disclosing.
What if the AI gave me a citation to a real-looking study?
Find and read the original before citing it. Language models can produce plausible references to papers that do not exist. Cite the study itself, never the chat, when the study is real.
My chat has no share link. Now what?
Cite it in text like personal communication and put the full prompt and output in an appendix so readers can see the exact material you worked from.
Does citing AI make it allowed?
No. Citation is disclosure, not permission. Your instructor's or publisher's policy determines whether AI use is acceptable in the first place.