How to cite ChatGPT in your thesis: when to cite, when to acknowledge (with APA, Harvard and MLA examples)
You're writing your literature review, and you ask ChatGPT to explain a concept you are still struggling with after multiple tries. Its answer is clear, better than anything you've read so far, and you'd like to use part of it. So you search for how to cite ChatGPT.
Before you format anything, ask yourself one question: what is ChatGPT doing in this part of your thesis? The answer decides whether you cite it, acknowledge it, or cite something else entirely.
The short answer
Most of the time, ChatGPT belongs in your AI declaration or acknowledgements, and the studies it helped you find belong in your reference list. You cite ChatGPT itself when its own words or output appear in your thesis, or it was used as a tool in your methods.
That's the position Michael Rowe and I take in our book on AI for doctoral students, Still Yours. Many university library guides take a similar approach: you reference AI when you use content it produced, and you acknowledge it when it helped with your process. I would add a third case: if it carried out part of your method, you describe that in your methods section and cite the tool. The rest of this post explains why, and then shows you how to format a citation when you need one. Although I refer to ChatGPT, the principles in this post are relevant to any genAI tool, be it Claude, Gemini, Copilot or any other.
Why ChatGPT is a poor source
A citation sends your reader to a source they can find, read and judge for themselves. ChatGPT's answers fall short of that in three ways:
Nobody stands behind the answer. You can share a ChatGPT conversation and put the link in your citation, so your reader can open the same conversation you had. The link takes them to the whole conversation, and they'll need to scroll to find the part you used. What they find there is text produced by a system that predicts likely words. Nobody reviewed it, and nobody stands behind it. Ask the same question tomorrow, and you may get a different answer, so the shared link is the only record of what you saw.
It can be confidently wrong. AI tools sometimes produce references that look real and don't exist, or describe a real paper in a way its authors wouldn't recognise. If you cite the AI's summary, you inherit its mistakes.
It isn't where the knowledge comes from. When ChatGPT explains a concept well, it's drawing on work that other people published. Those people deserve the credit, and your reviewers and examiners will expect you to know their work first-hand.
Four situations and each of their approaches
1. ChatGPT helped you find or understand sources
Say ChatGPT suggested five papers on your topic, or explained a theory so you could make sense of the original. Find each paper, read it, check that it says what ChatGPT claimed, and cite the paper. Some of the five may turn out not to exist, and some may say something rather different from the summary.
The help ChatGPT gave you goes in your AI declaration. I explain how to write one, with examples, in How to declare AI use in your thesis.
2. ChatGPT helped with your writing
If it suggested clearer phrasing or commented on your structure, it helped with your process. Nobody needs to look up its output to follow your argument. This also goes in your declaration, and no citation is needed.
3. ChatGPT's output is part of what you're investigating
Here you cite it. Some examples:
- You're researching how AI tools answer patients' health questions, and you quote the responses.
- You include an AI conversation as part of your data.
- You quote ChatGPT's words directly for any other reason.
In each of these examples, ChatGPT's output is what you're writing about, so your reader needs to be able to find it.
4. ChatGPT was a tool in your methods
Sometimes ChatGPT does part of the research work itself. Say you have 412 open-ended survey responses. You give ChatGPT a framework and ask it to sort each response into the framework's categories, which you then verify. ChatGPT is now part of your method, in the same way as the software you use for statistics or qualitative coding.
Describe what it did in your methods section or chapter, in enough detail that someone else could repeat it:
- the tool, the model and the dates you used it
- the task you gave it and the framework you provided
- your prompts in full, usually in an appendix
- how you checked its work, for example by reviewing every categorisation yourself, or by having a second researcher code a sample independently and comparing the results
- what you did when you disagreed with it
The checking step is the part your examiners will look at most closely. ChatGPT can put a response in the wrong category with complete confidence, so your methods need to show that a person checked its work and made the final decisions. Here's an example:
I used ChatGPT ([model name], OpenAI) in [month and year] to assign 412 open-ended survey responses to the five domains of [framework]. The prompt, which included the definition of each domain and three worked examples, is in Appendix C. I checked every assignment against the framework definitions and changed [number] of them. A second researcher independently coded a random sample of 80 responses, and we resolved disagreements through discussion. All final coding decisions were mine.
Cite the tool as well, using the second APA template below, or your library's guidance on citing software if you use Harvard or MLA.
Before you paste any participant data into ChatGPT, check that your ethics approval and your participants' consent cover it. Many universities advise against putting research data into public AI tools, and some provide an approved version for this kind of work.
How to cite ChatGPT in APA 7
In September 2025, the APA Style team updated its guidance. You can now cite a specific conversation if you've created a shared link to it, and the reference points to that chat (APA Style, 2025).
Template for a specific chat:
AI Company. (Year, Month Day). Title of the chat [Generative AI chat]. Tool or model name. URL of the shared chat
Example:
OpenAI. (2026, March 14). Common misconceptions about statistical power [Generative AI chat]. ChatGPT. https://chatgpt.com/share/[your chat ID]
If you know which model you used, add it after the tool name, for example ChatGPT GPT-5.
In-text: (OpenAI, 2026) or OpenAI (2026).
The title is the title of the chat, so rename the chat to something descriptive before you share it. APA also suggests describing your prompts in the text, in your methods section or in an appendix. They don't go in the reference itself.
If ChatGPT was a tool in your methods (situation 4), cite the tool itself. APA gives a second template for this:
OpenAI. (2026). ChatGPT [Large language model]. https://chatgpt.com/
Here, the year is when the tool was last updated, not when you used it.
How to cite ChatGPT in Harvard
Harvard comes in many local versions, so check your university library's guide first. Many UK universities use Cite Them Right. Its template for an AI response you can share a link to looks like this (South East Technological University, Cite Them Right: content created by your prompts):
Name of AI company (Year) AI-generated text by [name of AI tool] with prompt '[your prompt]', Day Month. Available at: URL (Accessed: date).
Example:
OpenAI (2026) AI-generated text by ChatGPT with prompt 'What are common misconceptions about statistical power?', 14 March. Available at: https://chatgpt.com/share/[your chat ID] (Accessed: 14 March 2026).
In-text: (OpenAI, 2026).
If you can't share a link, the reference records who received the response and when, for example: OpenAI ChatGPT (2026) ChatGPT response to Sam Taylor, 14 March. Check your library's guide, as this format is being updated.
How to cite ChatGPT in MLA
MLA updated its guidance in 2025 (MLA Style Center). It doesn't treat the AI tool as an author. The title describes what the tool generated, and can include your prompt.
Example:
"What are common misconceptions about statistical power?" prompt. ChatGPT, model [model name], OpenAI, 14 Mar. 2026, chatgpt.com/share/[your chat ID].
In-text: ("What are common misconceptions").
Save the link on the day
Whichever style you use, create the shared link and note the date of the conversation. If you delete the chat, the shared link stops working, and you can't get the conversation back. Save a copy of the conversation as well as the link. Keep a copy of your prompts too. APA suggests you keep them for your own records even when they don't appear in the thesis, and they'll help you answer questions in your viva about how you used the tool.
The formatting is the easy part
A reference generator will format a ChatGPT citation in seconds. The harder questions are whether you used AI at the right time, for the right reasons, and in the right way. Is the paper ChatGPT mentioned real? Have you read it? Does ChatGPT's summary match what the authors found? Could you explain in your viva why that paper belongs in your argument?
Our book, Still Yours: A Doctoral Researcher's Guide to AI, deals with those questions. Michael Rowe and I outline a literature workflow that uses AI for scoping and databases for searches you can document and repeat, and we explain the difference between reading a summary and understanding a paper well enough to defend it. We also cover when to cite AI, when to acknowledge it, and how to keep a record of your AI use so your declaration is easy to write.
The book is coming soon from Springer Nature.
***
The views in this post are my own and don't necessarily reflect those of my employers.
0 comments
Leave a comment
Please log in or register to post a comment