
How to Cite ChatGPT and AI Sources (APA, MLA, and Harvard)
By Priya Anand 10 min read
The moment you use an AI tool for your research, one question follows: how do you cite it? Here are the official APA, MLA, and Harvard formats, plus an honest look at what an AI citation can and cannot do for your paper.
If you used ChatGPT, Claude, Gemini, or any other AI tool while working on your assignment, you may need to cite it. Not because someone is trying to catch you out, but because academic writing runs on one simple rule: you show where your words and ideas came from. AI is no different.
Citing AI properly is not an admission of guilt. It is the opposite. It tells your marker that you used the tool openly and that you understand the difference between a tool that helped you and a source you are leaning on for evidence. Hiding it is what gets students into trouble.
This guide covers the official citation formats for APA, MLA, and Harvard, with in-text examples for each. It also covers the part most guides skip: an AI citation does not turn AI-generated text into a credible source. You still need real, verifiable references. We will come back to that.
One thing to check first
Guidance for citing AI is still evolving, and every institution is slightly different. Before you rely on any format below, check your course handbook or ask your instructor. Some departments require a specific disclosure statement, some allow AI only for limited tasks, and some do not allow it at all. The policy always wins over a general guide.
Do You Actually Need to Cite AI?
Not every use of an AI tool needs a formal citation, but almost every use needs to be acknowledged somewhere. The line usually falls between using AI to produce content that ends up in your paper and using it as a background assistant.
Cite or disclose it when you used AI to:
- Generate text, ideas, or an outline that shaped what you wrote.
- Produce code, data, or figures that appear in your work.
- Summarise or paraphrase a passage that you then used.
- Translate content that appears in your paper.
If you only used AI to check your grammar, tidy your formatting, or brainstorm topics you did not end up using, many institutions ask for a short disclosure statement rather than a formal citation. When in doubt, a one-line note in your methods or acknowledgements section (“I used ChatGPT to help refine the phrasing of this section”) is safer than saying nothing.
A formal citation has two parts, exactly like any other source: a reference list entry and an in-text citation. The formats below give you both for the three most common styles.
How to Cite ChatGPT in APA (7th Edition)
APA treats a large language model like software. The author is the company that made the model, not the model itself. So for ChatGPT the author is OpenAI, for Claude it is Anthropic, and for Gemini it is Google. The year is the year of the version you used, and you add a bracketed description that names it as a large language model.
Reference list format:
Company. (Year). Tool name (Version) [Large language model]. URL
Example:
OpenAI. (2026). ChatGPT (June 2026 version) [Large language model]. https://chatgpt.com
In-text:
(OpenAI, 2026)
APA also recommends that you describe how you used the tool in the body of your paper, usually in the method section, and that you keep a record of the prompt you entered and the response you received. Because another reader cannot retrieve your exact conversation, APA suggests including the full prompt and output in an appendix if the response is central to your work. That way your use is transparent and reproducible in spirit, even if the output cannot be re-generated exactly.
How to Cite AI in MLA (9th Edition)
MLA takes a different approach. It does not list the AI tool as an author. Instead, you describe the prompt you used as the title of the source, name the tool as the container, and add the version, the company, the date, and the URL. MLA is clear that you should acknowledge AI in your text and never present its output as your own.
Works Cited format:
“Prompt text” prompt. Tool name, Version, Company, Date, URL.
Example:
“Explain the main causes of the 2008 financial crisis” prompt. ChatGPT, June 2026 version, OpenAI, 10 June 2026, https://chatgpt.com.
In-text:
(“Explain the main causes”)
Because the prompt stands in for the title, your in-text citation uses a shortened version of that prompt in quotation marks. Keep the prompt in your reference short and descriptive so the reader can see exactly what you asked.
How to Cite AI in Harvard Style
Harvard is an author-date system, and the most widely used variant (Cite Them Right) handles AI much like APA. The company is the author, the year is the version year, you add the large language model description, and you include an access date because the output is not permanently retrievable.
Reference list format:
Company (Year) Tool name (Version) [Large language model]. Available at: URL (Accessed: Date).
Example:
OpenAI (2026) ChatGPT (June 2026 version) [Large language model]. Available at: https://chatgpt.com (Accessed: 10 June 2026).
In-text:
(OpenAI, 2026)
Harvard is not one fixed standard. Your university may use its own variant with small differences in punctuation or the order of elements. Check your institution's Harvard guide and match it exactly. Our Harvard referencing guide walks through the wider system if you need a refresher.
Chicago, Vancouver, and IEEE (In Brief)
The three styles above cover most students, but if your field uses a different one, here is the short version.
Chicago
Chicago usually handles AI in a footnote rather than the bibliography, because the content is not publicly retrievable. A typical note reads: “Text generated by ChatGPT, OpenAI, 10 June 2026, https://chatgpt.com.” If your instructor wants a bibliography entry as well, follow their guidance.
Vancouver and IEEE
Both are numbered systems common in science and engineering. They treat AI output much like software or a personal communication, with a numbered reference that names the company, the tool, the version, and the URL. Because conventions here are still settling, confirm the exact format with your department before you submit.
For a full side-by-side of how the six major styles handle sources in general, see our guide on how to cite sources in a research paper.
Quick Reference: The Three Main Styles
APA
Author is the company. In-text: (OpenAI, 2026). Describe your use in the method and keep the prompt.
MLA
Prompt is the title, tool is the container, no author. In-text: a shortened prompt in quotation marks.
Harvard
Author is the company, include an access date. In-text: (OpenAI, 2026). Match your institution's variant.
What an AI Citation Cannot Fix
Here is the part that matters most, and the part that a citation format alone will never solve. Citing ChatGPT correctly does not make the AI a credible source of evidence. AI is a tool, not an authority. If you write “AI use is rising in higher education (OpenAI, 2026),” you have cited the tool properly and still supported a factual claim with something that cannot back it up.
For any claim of fact, you need a real, verifiable source: a peer-reviewed article, a book, a dataset, an official report. AI can point you toward those sources, but it can also invent them. Language models are known to produce references that look perfect and do not exist, with plausible authors, titles, and journal names attached to nothing. Never paste an AI-generated reference list into your paper without checking that every single entry is real and says what the AI claims it says.
This is exactly why universities are moving toward process-based evaluation: they care less about whether you touched AI and more about whether your work is grounded in real sources and real understanding. Verify your references against a database likeGoogle Scholar, and if you are not sure where to look, our guide on how to find academic sources covers the ground.
5 Common Mistakes When Citing AI
1. Listing the AI as the author
Neither APA nor MLA treats the model as the author. APA uses the company; MLA uses the prompt as the title with no author. Writing “ChatGPT (2026)” as an author is the most common slip.
2. Citing AI as evidence for a fact
An AI citation shows where a piece of text came from. It does not give the claim authority. Back every factual statement with a real source, and use the AI citation only to acknowledge the tool.
3. Not saving the prompt and output
Because the conversation cannot be retrieved later, you may be asked to show what you actually did. Save the prompt and the response as soon as you use them, and keep them with your working files.
4. Trusting AI-generated reference lists
Fabricated references are the fastest way to lose a marker's trust. Open every source the AI suggests and confirm it exists, that the authors are right, and that it genuinely says what you are citing it for.
5. Skipping your institution's policy
A perfect citation does not help if your department banned the tool or wanted a specific disclosure format. Read the policy first, then cite. If you need help wording it, our guide on what AI can and cannot do for a literature review covers responsible use in more depth.
Sources You Can Actually Cite
The safest way to use AI in research is to keep it pointed at sources you can verify. That is how Litrevu is built. You upload the real papers you want to work from, and Litrevu synthesises a cited draft where every claim links back to a source you can open and read yourself.
Litrevu is an AI literature review assistant that turns the papers a researcher has already gathered into a cited first draft, with every citation traceable to the uploaded source.
No invented references, no citing a chatbot as evidence. You review every line, you check every source, and you own every word. Try it free, 800 words, no credit card needed.
Start writing for free