Cite AI output as a non-retrievable personal communication with the tool as author, the year of the version, and the date you generated it. APA 7th treats it as software with the developer as author, Harvard requires the prompt and date, and OSCOLA handles it as an unpublished source in a footnote.
Before You Cite: Check You Are Allowed to Use It
Citation format is the second question. The first is whether your institution permits AI use for this assessment at all, and in what role. Policies now fall into three broad groups: prohibited entirely, permitted with declaration, and permitted for defined tasks such as brainstorming or proofreading but not for drafting submitted text.
Citing correctly does not make prohibited use permitted. A citation is evidence of transparency, not authorisation.
APA 7th Edition
APA treats generative AI as software. The developer is the author, the version year is the date, and the model name is the title.
Reference list:
OpenAI. (2025). ChatGPT (Nov 2025 version) [Large language model]. https://chat.openai.com/chat
In text: (OpenAI, 2025)
APA also advises that because the output is not retrievable by your reader, you should include the full prompt and response in an appendix, and describe in your method how you used the tool.
Harvard (Cite Them Right)
Cite Them Right, the variant most UK universities follow, treats AI output as a personal communication and asks for the prompt.
Reference list:
OpenAI ChatGPT (2025) ChatGPT response to Subrata Saha, 14 March. Available at: https://chat.openai.com/chat (Accessed: 14 March 2025).
In text: (OpenAI ChatGPT, 2025)
Because the exchange is not publicly retrievable, keep a copy or screenshot. Several UK universities now ask for the transcript as an appendix.
OSCOLA
OSCOLA has no dedicated rule for AI, so it is handled under the guidance for unpublished and personal sources: a footnote, no entry in the bibliography.
Footnote:
1 ChatGPT (OpenAI) response to author's prompt regarding directors' duties under s 172 Companies Act 2006 (14 March 2025).
Law schools are among the strictest on AI use, largely because hallucinated case citations are both easy to produce and catastrophic. Never cite a case an AI tool gave you without verifying it in a law report.
MLA 9th
"Explain the four actions framework" prompt. ChatGPT, Nov 2025 version, OpenAI, 14 Mar. 2025, chat.openai.com/chat.
The Declaration Most Students Forget
A growing number of institutions require a statement of AI use even when nothing is quoted — including where a tool was used only to check grammar or generate an outline. It usually sits at the end of the methodology or on the cover sheet, and reads plainly:
"I used ChatGPT (OpenAI, Nov 2025 version) to generate an initial outline for Section 3 and to check grammar in Sections 2 and 4. All analysis, argument and final wording are my own. Prompts and outputs are in Appendix C."
An undeclared use that surfaces later is treated far more seriously than a declared one, because the concealment becomes the offence rather than the use.
The Hallucination Problem
Generative tools produce plausible references that do not exist — correct-looking author names, real journals, invented articles and page numbers. This is the most common way AI use is detected: a marker checks a reference and finds nothing. Verify every citation in the actual database before it goes in your list. If you cannot find it in your library catalogue, it is not a source.
Frequently asked questions
How do you cite ChatGPT in Harvard referencing?
Cite Them Right Harvard formats it as: OpenAI ChatGPT (2025) ChatGPT response to [your name], [date]. Available at: https://chat.openai.com/chat (Accessed: [date]). In text this becomes (OpenAI ChatGPT, 2025). Because the exchange is not retrievable by your reader, keep the transcript and include it as an appendix if your department asks.
Do I have to declare that I used AI?
Most universities now require a declaration of any generative AI use, including for outlining or grammar checking. Check your programme handbook, because policies differ by department and sometimes by module. An undeclared use discovered later is treated much more seriously than a declared one.
Can I cite an article that ChatGPT gave me?
Only after you have found and read the actual article. Generative tools routinely produce references that look correct but do not exist, and markers checking a reference list is the most common way AI use is detected. If it is not in your library catalogue or the publisher’s database, it is not a source.
Does citing ChatGPT protect me from a misconduct allegation?
It protects you from the concealment element, not from a prohibited use. If your assessment brief bans AI, citing it correctly still breaches the rules. Citation demonstrates transparency; it does not grant permission.
Want This Checked Before You Submit?
Our tutors review your draft against your own rubric and explain what to change and why — used by 45,000+ students across the UK, USA, Canada and Australia.