ChatGPT launched in November 2022, and within six months every major style body had to decide what to do about it. The result, by 2026, is a settled but still-evolving set of rules for citing AI-generated content. This guide walks through what APA 7, MLA 9, and Chicago currently require, then covers the edge cases most researchers actually run into.
Citing a book or a peer-reviewed paper works because the source is retrievable. Anyone who reads your citation can find the same text you read. AI-generated content isn't retrievable that way — the same prompt to the same model, seconds later, may produce a different response. Worse, the model itself is updated regularly. The "ChatGPT" you queried in 2024 is not the same model that will respond in 2026.
Every style's answer to this problem has been, roughly: treat the AI as the author, cite the model version, document the date, and preserve the prompt.
APA's guidance, formalized in 2023 and refined since, treats AI output as a kind of software.
In-text: (OpenAI, 2024) — treat OpenAI (or Anthropic, Google, etc.) as the author.
Reference list:
OpenAI. (2024). ChatGPT (September 25 version) [Large language model]. https://chat.openai.com/
Note the three required elements beyond a normal reference:
[Large language model] in square brackets as a source descriptionIf you quote a specific output in your paper, APA 7 recommends including the prompt in your text or in an appendix — the reasoning being that your reader can at least evaluate what you asked, even if they can't regenerate the exact response.
MLA's Works Cited format:
"Summarize the key findings of recent AI safety research" prompt. ChatGPT, 25 Sept. version, OpenAI, 25 Sept. 2024, chat.openai.com.
Notable differences from APA:
MLA's guidance explicitly says that if the output would normally be a footnote-able reference (a direct quote, a paraphrase), it needs a Works Cited entry. Casual mentions of "I asked ChatGPT" in your writing do not need a Works Cited entry, though you should still disclose that you used AI.
Chicago 17th edition's author-date format:
In-text: (ChatGPT 2024)
Reference list:
ChatGPT. 2024. Response to "What are the leading theories of consciousness?" OpenAI, September 25. https://chat.openai.com/.
For the notes-bibliography style, the full footnote format:
ChatGPT, response to "What are the leading theories of consciousness?", OpenAI, September 25, 2024, https://chat.openai.com/.
The same rules apply to any large language model. Substitute the developer and model name:
APA:
Anthropic. (2024). Claude (Sonnet 4.5) [Large language model]. https://claude.ai/
APA:
Google. (2024). Gemini (1.5 Pro) [Large language model]. https://gemini.google.com/
Same pattern — the developer is the author, the product name is the work, and the model version goes in parentheses.
AI Overviews are ephemeral in a particularly bad way: they don't even have a stable URL to the AI response, just to the search query. Most styles now recommend citing the underlying search query with the date, and noting that an AI Overview was consulted:
Google. (2024, October 15). [AI Overview for the query "climate change projections 2030"]. Google Search. https://www.google.com/search?q=climate+change+projections+2030
Check your target journal's guidance — some specifically prohibit citing AI Overviews.
If you used ChatGPT to help outline a paper or rewrite a paragraph, but the ideas and sources are your own, you don't cite ChatGPT as an author — the ideas aren't its. Most style bodies do, however, now require a disclosure statement in the paper's methods or acknowledgments. Something like: "The authors used ChatGPT-4 to help polish prose in the introduction and methods sections; all scientific content and conclusions are the authors'."
Many journals have made this disclosure mandatory. Check your target journal's policy — they're usually in the author guidelines under "AI and automated tools."
If you ask ChatGPT to summarize a specific paper and cite its conclusions, do not cite ChatGPT. Read the paper, and cite it directly. Citing ChatGPT as the source of a claim that came from a peer-reviewed paper is a citation error — and a minor research-integrity issue, because ChatGPT can misrepresent what a paper actually says.
Citing an AI-translated quote is a gray area. APA 7 recommends treating the translation as a normal paraphrase, citing the original source, and noting in brackets that it was translated (e.g., "translated by ChatGPT"). The reader is trusted to evaluate whether that's a problem for your claim.
Before you submit any paper that includes AI-generated content:
Get those six right and you're already ahead of most submissions.
Expect more nuance in the next two years. There are already style-body proposals for citing fine-tuned models (do you cite the base model or the fine-tuned variant?), for citing agentic workflows (how do you cite a response that was produced by an AI agent using five tools?), and for citing multi-modal generation (how do you cite an AI-generated image embedded in an AI-generated argument?).
The safest rule through all of it: be more explicit than you think necessary. Over-disclosure ages much better than under-disclosure.
If you want the in-browser tools to format these citations correctly, DEEPNOTIS's free citation generator supports every current AI-content template across the major styles, and our how-to guides for specific AI tools are kept current with the latest style-body revisions.