Citing ChatGPT in APA is treated as citing software, not as citing a person or a web page. APA published guidance on this in 2023, and the pattern it gives puts the company in the author position and the model version in the title. This page shows the reference, the in-text form, and the appendix requirement that most instructors add on top.
APA treats a chatbot response as the output of an algorithm, so the reference describes the tool rather than the conversation. The company that made the model is the author, because the model itself is not a legal or intellectual author.
The company is the author, so the in-text citation carries the company name and the year. This looks odd the first time you write it, because the sentence is about ChatGPT but the citation says OpenAI.
Parenthetical: (OpenAI, 2024)
Narrative: OpenAI’s (2024) ChatGPT produced a three-part structure.
Naming the model in the sentence: When prompted, ChatGPT summarised the argument in three parts (OpenAI, 2024).
Pointing at the appendix: The full exchange appears in Appendix A.
Reference list entry
OpenAI. (2024). ChatGPT (Aug 8 version) [Large language model]. https://chat.openai.com/chat
In-text, first mention
(OpenAI, 2024)
Describing the prompt in your text
When prompted to summarise the argument, ChatGPT produced a three-part structure (OpenAI, 2024). The full exchange appears in Appendix A.
AI citation policy is moving quickly and varies by institution. These are the situations that come up before the reference format is even the question.
| You used it to brainstorm but copied nothing | Most policies still require disclosure. Many ask for a statement of use rather than a formal citation. |
|---|---|
| You used it to edit your own writing | Usually disclosed in a methods or acknowledgements note rather than cited as a source. |
| It gave you a reference that turned out to be real | Read the actual paper and cite the paper. The model is not the source. |
| It gave you a reference that does not exist | Remove it. A fabricated citation in submitted work is treated as misconduct at most institutions. |
| You used a different model | Same pattern, different company: Anthropic. (2024). Claude (version) [Large language model]. URL |
| Your course prohibits AI entirely | A correct citation of a prohibited tool is still a violation. Check the handbook before using it at all. |
APA classifies a chatbot response as algorithm output, which puts it in the same category as statistical software rather than in the category of things you cite for what they say. That classification carries a consequence most students miss: it means the citation documents that you used a tool, not that the tool is evidence for a claim.
The distinction matters when a model tells you something factual. If ChatGPT says a study found a particular result, citing ChatGPT does not support that claim, because a reader cannot check it and the model cannot be held to it. The correct move is to find the study, read it, and cite the study. If the study turns out not to exist, which happens often, you have learned something important about the claim.
There is also a reproducibility problem that the reference format cannot solve. Two people giving the same model the same prompt on the same day can get different answers, and the model itself changes between versions. The version number in the reference is an attempt to bound this, and the appendix requirement exists because it is the only way a reader can see what you actually got. Neither fully fixes it, which is why AI output is weak evidence even when cited correctly.
Usually yes. Most institutional policies require you to disclose AI use even when no text was copied, and many ask for a statement of how it was used rather than a formal citation. Check your course handbook, because policies vary widely and change often.
You cannot cite it as a retrievable source, because your reader cannot open it. APA asks you to cite the tool and put the exchange in an appendix. A shared conversation link is not a substitute, since the content can be edited or deleted.
No. The pattern is the same with a different company and model name: Anthropic. (2024). Claude (version) [Large language model]. URL. What changes is only the author and the title.
That is the common failure mode, and a fabricated reference in submitted work is treated as academic misconduct at most institutions. Check every AI-supplied reference against a real database before you submit.
After the reference list, labelled Appendix A. Include the prompt exactly as you typed it and the response in full, so a reader can judge what the model contributed.
For research writing, often yes. A methods note describing which model, which version, and what it was used for is more informative than a citation, and many journals now require exactly that.