How to cite ChatGPT in 2026

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.

This guide gives you the exact format for APA 7, MLA 9, Chicago, and Harvard, with both a reference-list entry and an in-text citation for each, plus a side-by-side table, the mistakes that cost real marks, and answers to the questions students actually ask. By the end you will be able to drop a correct ChatGPT citation into any paper in under a minute and know when you should be citing the underlying source instead.

First, decide what you are actually citing

Most citation mistakes with AI happen before formatting, at the decision step. Run your use through three quick questions and the right move becomes obvious.

1. Did the AI's words or analysis end up in the work? If yes, cite or disclose the tool. If the AI just pointed you somewhere and you wrote from the real source, cite the real source.
2. Is the output a factual claim? A model is not evidence. Verify the claim against a primary or scholarly source and cite that source, not the chatbot.
3. Was this use allowed? Check the assignment. Citation does not turn a banned use into a permitted one.

Once you know you are genuinely citing the tool, pick your style below. Every example uses the same scenario so you can compare them directly: a student asked ChatGPT (GPT-4o) on 29 May 2026 for three counterarguments to a thesis about remote learning, then revised and fact-checked the result.

APA 7 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.

The structure is author (the company), year, italicised title with the model in parentheses, a bracketed description of the type, and the URL. There is no retrieval date, because APA treats the cited version as fixed. If your instructor wants the prompt and full response, APA puts those in an appendix and you point to it in the text, for example "(see Appendix A for the full ChatGPT transcript)." Keep the reference list entry lean.

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 9 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.

MLA orders the entry as a description of the prompt in the title position, the AI tool as the container in italics, the version, the publisher (OpenAI), the date of the conversation, and the site. The in-text citation uses the first words of that description so the reader can match it to the entry. If your prompt is long, do not paste the whole thing into the title position. Use a short description and move the full prompt to a note or appendix if the assignment needs that level of audit trail.

Chicago example (notes-bibliography)

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.

Chicago is comfortable handling AI in a note because a note can carry the full context without pretending a private chat is a published work. If you used the author-date system instead of notes-bibliography, you would cite it more like the APA pattern, with an in-text (OpenAI 2026) and a matching reference-list entry. Pick the Chicago system your assignment specifies and stay consistent across the whole paper.

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.

Because Harvard varies by institution, treat the entry above as a strong default rather than a universal rule. The components your marker will expect are stable across most variants: who made the tool, the year, the tool and model, what you asked it for, where it lives, and when you accessed it. If your university publishes a referencing guide, match its exact punctuation and capitalisation, since Harvard markers tend to be strict about format.

APA vs MLA vs Chicago vs Harvard at a glance

The four styles agree on the substance and differ mostly on ordering and punctuation. This table shows how the same ChatGPT use maps across them so you can spot the differences without rereading each section.

Common mistakes (and the fix)

• Citing ChatGPT for a fact instead of the real source. The model is not evidence. Fix: track down the article, book, or dataset it referred to, confirm it says what you think, and cite that. If you cannot verify it, do not use the claim.
• Listing ChatGPT as the author. The chatbot is not a person responsible for the work. Fix: use OpenAI as the maker of the tool (APA, Harvard), a prompt description (MLA), or name OpenAI inside a note (Chicago).
• Pasting the entire prompt into the title or reference slot. It makes the entry unreadable. Fix: write a short description in the citation and move the full prompt to an appendix if the assignment requires one.
• Trusting an AI-generated reference list. Models invent plausible-looking citations to sources that do not exist. Fix: verify every entry against the actual source before it goes in your bibliography, or run your list through a checker.
• Citing the tool but ignoring the policy. A citation does not authorise a banned use. Fix: read the assignment rules first. If AI drafting is not allowed, citing it only documents the breach.
• Using a publication date instead of the date you ran the prompt. A private chat has no publication date. Fix: record the exact day you generated the response and format that date for your style.
• Mixing styles in one paper. Switching between APA parentheses and Chicago notes signals carelessness. Fix: pick the style your assignment requires and apply it to every entry, AI or not.

Tools that make this faster

Formatting one ChatGPT citation by hand is quick. Doing it consistently across a full reference list, in the right style, with no stray punctuation, is where most marks leak away. A few free Phrasit tools handle the mechanical parts so you can spend your time verifying sources instead of fighting commas.

• Build correctly formatted entries in APA, MLA, Chicago, or Harvard with the free citation generator, then drop in the AI-tool details from the examples above.
• Run your finished reference list through the citation checker to catch missing fields and format slips before you submit.
• Confirm that an AI-assisted draft is genuinely your own writing with the plagiarism cross-check, which compares passages so you can spot text that stayed too close to the source.
• Keep an assignment inside its limit while you revise with the word counter, useful when an instructor caps the length of an AI-use note or appendix.

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.

Frequently asked questions

Do I have to cite ChatGPT if I only used it for ideas?

If the AI only helped you brainstorm and you then found, read, and wrote from real sources, you cite those sources, not ChatGPT. You cite or disclose the tool when its output actually shows up in the work: quoted wording, a paraphrase, a summary, an analysis, or a structure you adopted. The test is whether a reader is judging the AI's contribution. Disclosure is separate from citation, and some instructors want a short acknowledgment of any AI use even when nothing is quoted, so check the assignment.

Can I list ChatGPT as the author of a source?

No. MLA is explicit that you do not treat the AI tool as an author, because the author is the person or group responsible for a work. APA cites the company that makes the model (OpenAI) as the author of the software, not ChatGPT itself. So the author position is either a human, OpenAI as the maker of the tool, or a short description of the prompt, depending on the style. It is never the chatbot persona.

What date do I use when I cite ChatGPT?

Use the date you ran the prompt, not a publication date, because there is no fixed publication date for a private chat. APA uses the year of the version you used. MLA and Chicago want the full day, month, and year of the conversation. Harvard styles usually want both the date used and an access date. When in doubt, record the exact day you generated the response and you can format it for any style.

How do I cite a model other than GPT-4o?

Name whatever model your interface shows at the time, in the same slot where the examples here put GPT-4o. If you used a reasoning model, a vision model, or a newer release, write that label exactly as it appears. If the interface does not name a specific model, you can omit the version and just cite ChatGPT as the tool. The same pattern works for other assistants: swap OpenAI and ChatGPT for the company and product you actually used.

Should I include the full prompt and the AI's full answer?

Not in the reference entry. Put a short description of the prompt in the citation so a reader can match it to the in-text reference, then move any long prompts or full transcripts to an appendix if the assignment asks for that level of audit trail. Stuffing a paragraph-long prompt into the title or reference position makes the entry unreadable and does not help the reader.

Can I cite ChatGPT as the source of a fact or statistic?

No. A language model is not evidence for a factual claim, and it can produce confident citations to articles that do not exist or do not say what it claims. If ChatGPT points you toward a study, find the study, read it, and cite the study. If you cannot verify it, leave it out. A ChatGPT citation should describe how you used the tool, not stand in for a source you never checked.

Is using ChatGPT plagiarism if I cite it?

Citation answers where text came from. It does not answer whether the use was allowed. If an assignment bans AI-generated drafting, citing the tool does not make the use acceptable, it just documents a rule you broke. Plagiarism is presenting work as your own without attribution. The bigger risk with AI is usually a permission problem, not an attribution one, so read the course policy before you rely on the output.

Which citation style should I use for AI tools?

Use whatever style your course or journal requires. APA is common in psychology, education, and the sciences. MLA is common in the humanities. Chicago and Turabian appear in history and some social sciences. Harvard is widely used across UK and Australian universities, but it is a family of variants rather than one rulebook, so follow your department's guide. The underlying decision (cite the tool, describe the use, verify real sources separately) is the same across all of them.

Related reading

• ChatGPT citation rules, explained
• The real cost of bad citations
• MLA 9 changed how you handle 3+ authors
• Chicago vs Turabian: which one your paper needs