Skip to content
← Guides & helpTrust & verification8 min readBy The CiteDash team

How to cite AI tools like ChatGPT in APA and MLA

APA 7 and MLA formats for citing ChatGPT and other AI tools, with copyable templates, in-text examples, and when you need a disclosure instead.

Whether you may use ChatGPT in your thesis is a policy question for your university. How to cite it when you do is a style question, and the major style bodies have answered it. This guide gives you the APA 7 and MLA formats with templates you can copy, what Chicago and Harvard users should do, and the difference between citing an AI tool and disclosing that you used one.

One framing note before the formats. You cite an AI tool when its output appears in your text or is the thing you are analysing. You disclose an AI tool when it shaped your process. Most graduate students need the second far more often than the first, and confusing the two is the most common mistake in this whole area.

The formats below name ChatGPT because it is the tool students ask about most, but the same patterns apply to Claude, Gemini, Copilot, and any other generative tool. Substitute the company as the author or publisher, the tool as the titled work or container, and keep the version and the date you used. Nothing about the logic is ChatGPT-specific.

When you need to cite an AI tool, and when you don't

Cite the tool when you quote its output directly, when you paraphrase a specific response, or when the output itself is your object of study, for example a project analysing how language models describe historical events. In those cases the AI text is functioning as source material, and source material gets cited, whatever produced it.

Do not cite the tool for help that shaped your process rather than your text: brainstorming, suggesting structure, tightening prose, or explaining a method you then wrote up yourself. That assistance belongs in an acknowledgment or a disclosure statement, whose form your university probably specifies. And never transplant references an AI mentions into your bibliography: models fabricate plausible-looking citations, and a fabricated reference in your thesis is your responsibility, not the tool's.

A useful test when you are unsure: would a careful reader want to know that these exact words or this exact analysis came from a model? If yes, cite. If the reader instead needs to know how the thesis was produced, disclose. If neither, you are probably describing ordinary tool use, like a spellchecker, that your policy may not require you to mention at all, though when in doubt the safer default is the disclosure.

How to cite ChatGPT in APA 7

APA's published guidance treats the company as the author and the tool as the titled work, with the version you used and a bracketed description of what kind of thing it is. The date is the year of the version.

  • Reference list template: OpenAI. (Year). ChatGPT (Version date) [Large language model]. General URL of the tool
  • Worked pattern: OpenAI. (2026). ChatGPT (June 2026 version) [Large language model], followed by the tool's general URL
  • Parenthetical in-text citation: (OpenAI, 2026)
  • Narrative in-text citation: OpenAI (2026)

APA also wants your method described

The reference entry is only half of APA's guidance. It also recommends describing how you used the tool in your method section or introduction, quoting the relevant prompt in your text where it matters, and placing long transcripts in an appendix.

The appendix is not bureaucratic decoration. The same prompt can produce different output on different days, so your quoted material cannot be re-derived the way a page reference can be checked. The transcript in the appendix is what makes the quote verifiable, and examiners in AI-adjacent topics increasingly expect to find it there.

To adapt the pattern for another tool, keep the structure and swap the parts: the company that makes the tool is the author, the tool is the title, and the version and year are whatever you actually used. If you accessed the model inside another product, name that product in the version information so a reader can tell which interface produced the text.

How to cite ChatGPT in MLA

The MLA Style Center takes a different angle: the AI is not an author. Instead you build the works-cited entry from MLA's core elements, using a description of what was generated, usually the quoted prompt, as the title of the source, with the tool as the container.

  • Works-cited template: "Your prompt text" prompt. ChatGPT, version used, OpenAI, date you generated the text, general URL of the tool
  • Worked pattern: "Summarise the main critiques of social capital theory" prompt. ChatGPT, June 2026 version, OpenAI, 14 June 2026, general URL of the tool
  • In-text citation: a shortened form of the title, for example ("Summarise the main critiques")

What about Chicago, Harvard, and the other styles?

Chicago's published guidance handles AI-generated text in a numbered footnote rather than the bibliography, treating it much like a personal communication: name the tool, the company, and the date the text was generated, and include the prompt in the note if it is not already in your text.

Harvard is not one standard but a family of institutional variants, so follow your own university's Harvard guide; most guides that have addressed AI adapt the APA-style pattern of company, year, tool, and version. Whatever style you write in, the ordinary sources in your bibliography still need correct formatting: CiteDash's free APA citation generator and MLA citation generator cover those, and the Reference Manager inside the platform formats bibliographies in 12 styles, including both Chicago variants.

Notice that all of these styles are converging on the same small set of facts: which tool, which company, which version, which date, and what prompt. If you record those five things whenever AI output might end up in your text, you can produce a correct citation in any style later, including a style you have not chosen yet.

Quoting versus paraphrasing AI output

Quoted AI text takes quotation marks and a citation, exactly like quoted human text. Paraphrased AI output still needs the citation, because the ideas in that passage came from the tool's response, not from a source your reader can consult. If the passage matters to your argument, quote it: paraphrasing machine output invites the suspicion that you are laundering generated text into your own voice.

Keep the full transcript of any exchange you cite. AI output is not reproducible the way a published page is; your citation points to an interaction only you witnessed, so the transcript is the closest thing to an archival record, and an examiner is entitled to ask for it.

Never cite the sources an AI tool cites

A model asked for literature will produce confident, well-formatted references, and some of them will not exist. Others exist but do not say what the summary claims. Every reference that reaches your bibliography must be one you retrieved and read yourself: check that the DOI resolves, get the paper, and confirm it supports the sentence it is attached to.

This is the failure mode CiteDash was built to remove. In the platform, a citation is a database object that resolves to a real paper, never a free-typed string, and the Fact Checker verifies each AI-assisted claim against the held full-text source before it reaches you. If a claim cannot be verified against a real source, you see that, rather than a plausible fabrication.

Citing an AI tool does not make it evidence

A correctly formatted AI citation documents provenance; it does not confer authority. The style guidance exists for transparency, not because model output is citable evidence for factual claims. If a sentence in your literature review needs support, that support must come from the literature, and the AI can at most have helped you find or summarise it.

Examiners are increasingly familiar with these formats and read them accordingly. A ChatGPT citation attached to a claim about your field signals that the claim has no real source. The same citation attached to a methods note explaining that you analysed model output signals rigour. The format is identical; the credibility difference is entirely about what job the citation is doing.

Citation or disclosure? You may need both

A citation covers AI text that appears in your document. A disclosure statement covers AI involvement in how the document came to be, and many universities now require one in the thesis front matter regardless of whether any AI text is quoted. The two do different jobs, and one does not substitute for the other.

If you quoted a ChatGPT response in your methods discussion and also used AI assistance while drafting, you need the citation for the quote and the disclosure for the assistance. Our guide to the AI use disclosure statement covers the wording, the placement, and the level of detail examiners expect.

Where does the disclosure live if your university has no template? The emerging convention is the front matter, near the declarations page, in a short titled statement. Some students add a one-line pointer in the acknowledgments as well. What matters is that an examiner can find it without hunting for it.

Common mistakes when citing AI tools

Most citation problems with AI tools come from treating the model like a normal published source. These are the errors that come up most often:

  • Giving the model a personal-style author name instead of following your style's format
  • Citing a shareable chat link as the location: those links can break or be deleted, which is why style guidance points to the tool's general URL
  • Citing the AI as the source of a factual claim instead of citing the literature the claim actually rests on
  • Omitting the version or the date, which makes an already unreproducible source completely untraceable
  • Copying references the model suggested straight into the bibliography without retrieving them
  • Using an AI citation to pad a minimum-sources requirement, which examiners read exactly as you would expect
  • Discarding the transcript, leaving nothing to show if the quoted passage is questioned

The bottom line on citing ChatGPT and other AI tools

Check your policy first, because permission precedes format. Cite AI output only when it appears in your text, using the APA company-as-author pattern or the MLA prompt-as-title pattern above. Disclose process use separately, keep transcripts, and verify every real reference yourself. The formats are the easy part: the discipline of separating cited output, disclosed assistance, and verified literature is what keeps your bibliography defensible in an examination.

And keep the five facts recorded as you work: tool, company, version, date, prompt. They take seconds to note in the moment and cannot be reconstructed later.

Ready to try it on your own thesis?

Get Started Free

Do this in CiteDash

More guides