How to Cite ChatGPT and AI Tools in APA, MLA, Harvard and Chicago

The mechanical part of this is easy, and the templates below will get you through it in a minute. The part that actually matters is the decision that comes before: a chatbot response is not a retrievable source, carries no peer review, and cannot be checked by your reader. Citing one correctly does not make it evidence.

This post covers the formats for the four styles most students need, where the prompt belongs, the distinction between citing AI as a source and disclosing AI assistance, and the one mistake that turns a formatting question into an academic integrity case.

First Decide Whether to Cite It at All

There are three distinct situations, and they call for three different responses.

The third row is where most students go wrong. A model asserting that something is true is not evidence that it is true, and a citation to a chatbot in support of a substantive claim signals to an examiner that the claim was never verified. If the model told you something useful, go and find the paper it was paraphrasing - then cite the paper.

Before any of this, check your own institution's regulations. Some prohibit citing generative AI output outright, some require a specific disclosure format, and a local rule overrides any style guide.

APA 7

APA treats AI output as software: the company that made the model is the author, and the model is the title.

Three details people get wrong:

• The author is the company, not the model. "ChatGPT (2026)" is incorrect. OpenAI made it; OpenAI is the author.
• The year is the year of the version you used, not the year the product launched.
• Include the version in parentheses. Models change behaviour between versions, so the version is what makes your description reproducible - as far as it can be.

APA also expects you to make the interaction visible: either describe the prompt in your text, or put the full prompt and response in an appendix if it is long. A reader cannot retrieve your session, so the appendix is doing the work the URL normally does.

MLA 9

MLA builds the entry from its standard template, with the prompt in the Title of Source position and the model as the container.

MLA's guidance goes further than format. It asks you to do three things: cite the tool whenever you quote, paraphrase or incorporate its output; acknowledge purely functional uses in a note or your text rather than in Works Cited; and vet any secondary sources the tool mentions, because it may have invented them.

Chicago (Notes and Bibliography)

Chicago's position is the clearest of the four: a chat session is not retrievable by your reader, so it belongs in a note or in your running text, not in the bibliography.

No bibliography entry is needed, on the same logic Chicago applies to personal communications. If your department or publisher requires one anyway, follow their instruction - it overrides the style guide.

Harvard

Harvard is not one style but a family of author-date conventions that differ between institutions, so this is the one to check locally before anything else. The Cite Them Right form, which many UK universities follow, looks like this:

Many Harvard-using departments also ask for the prompt and the full response in an appendix, referenced from the in-text citation. If your library publishes a referencing guide, use its wording rather than this one.

Quick Comparison

Style guides have been revising this guidance repeatedly since 2023, and the versions above reflect the current published formats. Check the style's own site before submission - this is the single fastest-moving corner of citation practice right now.

Citing AI as a Source vs Disclosing AI Assistance

These are different obligations and they are frequently confused. A citation credits a source you drew content from. A disclosure states how a tool was used in producing the work. You can owe one, both, or neither.

Where the disclosure goes varies: declaration page, methods section, acknowledgements, or a dedicated AI use statement. Follow the specified location, and write it as you work rather than reconstructing it at submission. Our guide to using AI ethically in academic writing covers disclosure in more depth.

The Mistake That Becomes an Integrity Case

Language models generate references that look real and are not: plausible author names, a real journal, a DOI that resolves to nothing or to an unrelated paper. Citing one is recorded as fabrication, and intent is not a defence - the reference either exists or it does not.

The rule has no exceptions: open every source before it enters your reference list. Not the model's summary of it, not the abstract shown in a search result - the paper itself. Confirm the authors, the year, the journal, and that it says what your sentence claims. If a reference cannot be located through a library catalogue or its DOI, delete it.

This is also the practical argument for tools that retrieve from indexed databases rather than generating references from model memory: the citation exists because it was found, not because it was predicted.

FAQs About Citing AI Tools

Do I have to cite ChatGPT if I only used it for ideas?

Not as a source, because you are not quoting or paraphrasing its output. Many institutions still require you to disclose it. Check your regulations, and when they are ambiguous, disclose.

Can I list ChatGPT as a co-author?

No. Every major publisher and style body prohibits it, on the grounds that authorship entails accountability for the work and an AI system cannot be accountable. Journals will reject submissions that do this.

How do I cite AI-generated images or code?

The same way, with the description in brackets changed: [Text-to-image model] for an image generator, [Large language model] for code produced by a chatbot. Note that image generators raise separate copyright and licensing questions that a citation does not resolve.

What if the AI is built into another tool, like Grammarly or Word?

Routine spelling and grammar checkers are normally treated as standard software and not cited or disclosed. Generative features that draft or substantially rewrite prose are a different matter and usually do require disclosure. If you cannot tell which one you used, err toward disclosing.

My university says not to cite AI at all. Then what?

Follow the university. Either the prohibition means the use itself is not permitted, in which case do not use it, or it means AI output does not belong in a reference list, in which case use a disclosure statement instead. Ask your supervisor in writing which they mean and keep the reply.

Do I need to keep my chat logs?

Keep them. Some institutions require a prompt log as part of disclosure, and even where they do not, being able to show what you asked and what came back is the strongest response available if your work is ever questioned. Reconstructing a log afterwards is not possible.

How do I cite a specific version when the model updates silently?

Record the version name shown in the interface and the date of the session, and state both. Where no version is displayed, give the date and note in your methods that the provider did not expose version information - accuracy about what you can and cannot know is better than a guess.

Getting the format right takes a minute. The judgment underneath takes longer and matters more: chatbot output is not evidence, and a correctly formatted citation to it does not become evidence. Cite AI when the output itself is what you are examining. Everywhere else, find the source it was standing in for. Our complete citation guide covers the styles for sources that can actually be retrieved.