How to cite ChatGPT and other AI tools in APA, MLA and Chicago

The rules every style agrees on

• An AI tool is not an author. COPE, the ICMJE and Elsevier say so explicitly, and Nature Portfolio keeps accountability with the human authors: a tool cannot take responsibility for the work, approve the final version or answer for its accuracy.
• Cite output you quote or paraphrase. If text, an idea or a figure produced by the tool appears in your work, it gets a citation like any other source.
• Cite the original source, not the AI's summary of it. When the tool tells you what a paper found, find the paper, read it and cite it. Assistants can invent references that look real.
• Disclose functional use. Using a tool to translate, edit or restructure your own writing is not quoting it, but schools and journals usually want it acknowledged — in a note, the methods section or a declaration.
• Keep a record. Save the prompts, the date and a shareable link to the conversation; a reviewer or examiner may ask for them.

APA Style

In September 2025 the APA Style team replaced its 2023 guidance. The current approach cites the specific chat you had, not the tool in general. The author is the company that makes the tool, the date is the date of the chat, and the title is the chat's own title in italics, followed by a description in square brackets:

OpenAI. (2025, August 21). Urban heat island mitigation in small parks [Generative AI chat]. ChatGPT.
    https://chatgpt.com/share/...

The in-text citation is author and year: (OpenAI, 2025) or OpenAI (2025). The same pattern works for other tools — Google for Gemini, Anthropic for Claude, Microsoft for Copilot — with the tool's name after the bracketed description.

You will still see the 2023 format in course handouts: OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat. APA now keeps that post only as an archive. If your instructor or journal asks for a particular version of the rules, follow theirs.

MLA Style

MLA updated its guidance in August 2025. It uses MLA's usual core elements, with the AI-specific choices below:

• Author — none. The tool is never the author.
• Title of source — a description of what was generated, normally the prompt in quotation marks followed by the word prompt.
• Title of container — the tool, in italics: *ChatGPT*.
• Version — the model name or number, such as GPT-4o.
• Publisher — the company: OpenAI.
• Date — the day the content was generated.
• Location — a stable, shareable link to the conversation; the tool's general address only when no shareable link exists.

Put together, a works-cited entry looks like this:

“Explain the urban heat island effect in two sentences” prompt. ChatGPT, model GPT-4o,
    OpenAI, 12 Oct. 2026, chatgpt.com/share/...

In the text, cite the shortened title: (“Explain the urban heat island”). MLA also asks you to follow the tool's own links to its sources and cite those, because AI tools can make sources up.

Chicago style

Chicago's advice is to say in your text that you used the tool — for example, “The following summary was generated by ChatGPT” — and, in more formal writing such as a research article, to add a numbered note. Its two example notes are:

1. Text generated by ChatGPT, OpenAI, January 10, 2026, URL.
1. ChatGPT, response to “Explain how to make pizza dough from common household
   ingredients,” OpenAI, January 10, 2026, URL.

Use the first when your text already gives the prompt, the second when it does not. Chicago's guidance shows notes only; ask your instructor or publisher whether they also want a bibliography entry.

Disclosing AI use in a journal article

Citing a quotation is not the same as disclosing how you worked. Publishers now ask for both, in different places:

• Nature Portfolio grades AI use by risk. Assistive use — polishing language, suggesting structure, translation — is permitted, and disclosing it is encouraged. Use that shapes interpretation is permitted only with human oversight, verification and disclosure. AI that replaces scholarly judgement, or is used without transparency, is not permitted.
• Elsevier asks for a separate declaration at the end of the manuscript, just before the references, naming the tool, what it was used for and how you oversaw it. Basic grammar, spelling and punctuation checks need no declaration.
• The ICMJE asks authors to describe AI use in the cover letter and in the submitted work — in the acknowledgments for writing assistance, in the methods for data collection or analysis.

Policies change and differ between journals, so read the author guidelines of the journal you are submitting to. A short, factual statement works everywhere: which tool and version, what you used it for, and that you checked and take responsibility for the result.

Mistakes that get papers questioned

1. Listing ChatGPT as a co-author or in the author byline.
2. Citing ChatGPT for a fact or a statistic instead of the paper the fact comes from.
3. Copying references an assistant suggested without opening them — check each one with the citation generator, which only formats works it can find in OpenAlex or Crossref.
4. Leaving out the date and the model; the same prompt gives different answers on different days.
5. Pasting long generated passages into the text. If the conversation matters, put it in an appendix and cite it.

If you used AI to help write, read can you use AI to write a research paper? for what journals allow, and are AI detectors accurate? before you worry about a detector score.