How to cite ChatGPT in 2026
APA, MLA, Chicago, and Harvard examples for citing ChatGPT and other AI tools without overcomplicating the paper.
The cleanest ChatGPT citation starts with a boring question: did the AI output become part of the work your reader is judging? If you quote it, paraphrase it, summarize it, adapt its wording, use its analysis, or build a section around its answer, cite or disclose it. If it only helped you brainstorm search terms and you then read the real sources yourself, cite the real sources.
That distinction is still the practical rule in 2026. The APA Style Blog treats ChatGPT as software when you cite the tool. The MLA Style Center says not to list the AI tool as the author, because the author is the person or group responsible for the work. The Chicago Manual of Style is more comfortable with notes and disclosure, especially when the conversation is not public.
Before you format anything, check the course policy. Some instructors allow AI for outlining but not drafting. Some allow grammar support but require a note. Some ban it for the assignment. A perfect citation will not fix a use that the assignment did not permit. Citation answers the reader's "where did this come from?" question. Permission answers the instructor's "was this allowed?" question.
APA example
APA 7 has two pieces to remember. The reference entry points to the tool, and the paper explains the specific use. If the model name appears in your interface, include it. If your instructor asks for prompts or transcripts, put them in an appendix rather than stuffing the reference list.
Reference list example: OpenAI. (2026). ChatGPT (GPT-4o) [Large language model]. https://chatgpt.com
In-text example: ChatGPT helped generate a first-pass comparison table, which the author revised and fact-checked (OpenAI, 2026).
APA also expects ordinary source discipline. If ChatGPT suggests that a 2022 article proves your point, the AI is not your source for that claim. Find the article, read it, and cite the article. If the article does not exist or does not say what the tool claimed, leave it out. A ChatGPT citation should not become a shield for unverified evidence.
MLA example
MLA 9 is built around containers, so the generated response is the thing you are citing and ChatGPT is the container. MLA also wants enough detail for a reader to understand the prompt without pretending the private chat is a normal published source.
Works cited example: "Prompt asking for three counterarguments to a proposed thesis about remote learning." ChatGPT, GPT-4o, OpenAI, 29 May 2026, chatgpt.com.
In-text example: The generated response grouped the objections into access, motivation, and assessment concerns ("Prompt asking for three counterarguments").
If your prompt is long, do not paste the whole thing into the title position. Use a short description that lets the reader match the in-text citation to the entry. The full prompt can go in a note or appendix if the assignment needs that level of audit trail.
Chicago example
Chicago often handles AI output cleanly in a note. If you reproduce words from ChatGPT, use a footnote or endnote. If the use is general drafting help, a short disclosure in the text may be enough, depending on the assignment.
Note example: 1. Text generated by OpenAI, ChatGPT, GPT-4o, response to "Suggest three objections to this thesis about remote learning," May 29, 2026, https://chatgpt.com.
Bibliography rule: if the conversation has no public share link, many Chicago uses leave it out of the bibliography and keep the detail in the note.
Harvard example
Harvard is a family of university styles, not one global rulebook. Your school may have its own AI policy. A safe Harvard-style entry usually names the company, year, tool, model or descriptor, date used, URL, and access date.
Reference list example: OpenAI (2026) ChatGPT, GPT-4o. Response to prompt on 29 May. Available at: https://chatgpt.com (Accessed: 29 May 2026).
In-text example: OpenAI (2026) grouped the draft objections into three categories.
What to keep
Keep a small AI-use note while you work: tool, model if shown, date, prompt summary, what you used, and what you changed. Do not cite ChatGPT for facts it cannot verify. Use it as a cited tool when its output matters, and cite primary or scholarly sources when those sources carry the evidence.
The most defensible habit is boring documentation. Write down enough that you could explain the use two weeks later without reconstructing it from memory. If the AI helped with wording, say that. If it helped with structure, say that. If it produced text you quoted, cite it directly. The citation should make your process clearer, not make the AI step look more scholarly than it was.
Related reading
- ChatGPT citation rules, explained
- The real cost of bad citations
- MLA 9 changed how you handle 3+ authors