How to cite ChatGPT or Claude in a thesis: APA, MLA, ISO 690

You used ChatGPT or Claude while working on your thesis, and your university’s rules say you have to disclose it. Fair enough, but how? Does the AI go in the reference list, in a footnote, in an appendix? And what do you do when your citation style, ISO 690 for instance, says nothing about AI? Here’s what APA, MLA and the Chicago Manual of Style officially recommend, how to adapt ISO 690, and the crucial difference between citing an AI and disclosing your use of AI.

In short: an AI is never cited as an author, only as a tool. In APA, the author is the company behind the model: OpenAI. (2023). ChatGPT (version) [Large language model]. URL, with (OpenAI, 2023) in the text. In MLA, the title of the source describes your prompt, followed by the tool, the model, the publisher, the date and a share link. In Chicago, a footnote is enough. ISO 690 has no official rule for AI: you adapt the format for online resources and get it approved. In every case, citing doesn’t replace an AI-use statement, and an AI is never the source of a fact: go back to the primary source.

First: citing is not permission

Before “how” comes “whether.” Every institution sets its own rules, and they vary a lot (we compared several in our article on academic integrity). The Université d’Angers charter in France, for example, requires the agreement of the instructor or supervisor before using generative AI, and forbids presenting AI output as your own.

Citing an AI response correctly makes your use transparent. It doesn’t make it allowed. If your assignment bans AI, a flawless reference won’t change that.

APA 7: the model’s developer as author

In April 2023, APA published a reference post, “How to cite ChatGPT,” by Timothy McAdoo. The principle: a conversation with an AI isn’t a “personal communication,” because no person is communicating. So you cite the algorithm and credit its creator as the author.

Reference (example from APA’s post):

OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat

In the text: (OpenAI, 2023), or OpenAI (2023) as a narrative citation.

Each element has a job: the author (the company that built the model), the year of the version, the tool’s name in italics, the version in parentheses, the description in square brackets, then the URL. APA also advises explaining in your text how you used the tool (in your method section, for example), reproducing the prompt, then the part of the response you’re quoting. If the response is long, the full version can go in an appendix.

For Claude, the same structure gives, for example (an adaptation of the APA pattern, using the version shown in your interface):

Anthropic. (2026). Claude (Opus 5.5 version) [Large language model]. https://claude.ai

MLA: your prompt becomes the title

MLA revised its guidance in August 2025. The first rule is unchanged since 2023: don’t treat the tool as an author. Fill in the usual MLA template like this:

• Title of Source: a description of what the AI generated, often your prompt;
• Title of Container: the tool’s name (ChatGPT, Claude);
• Version: the model name or number, as specifically as possible (new in 2025);
• Publisher: the company that made the tool;
• Date: when the content was generated;
• Location: ideally the conversation’s stable share link, otherwise the tool’s general URL (also new in 2025).

MLA’s official example:

“Describe the theme of nature in Jane Austen’s Mansfield Park” prompt. ChatGPT, model GPT-4o, OpenAI, 23 Sept. 2024, chatgpt.com/share/66f1b0a0-d704-8000-be9a-85f53c850607.

In the text, use the start of the title: (“Describe the theme”).

Applied to a thesis with Claude, it could look like this (an adaptation): “Explain the difference between correlation and causation” prompt. Claude, model Claude Opus 5.5, Anthropic, 27 Sept. 2026, followed by the share link if you created one.

MLA adds two rules many students miss. All “functional” uses of the tool, such as editing your prose or translating words, should be acknowledged in a note, in your text or in another suitable place. And any sources the AI hands you must be checked (more on that below).

Chicago: just a note

The Chicago Manual of Style is the lightest. According to its Q&A, which follows the 18th edition (2024), you need to credit the tool whenever you use text it generated. For most writing, naming it in your text is enough. For academic work, a footnote looks like this:

1. Text generated by ChatGPT, OpenAI, January 10, 2026, URL.

If the prompt isn’t in your text, it can go in the note:

1. ChatGPT, response to “Explain how to make pizza dough from common household ingredients,” OpenAI, January 10, 2026, URL.

The URL is the one from the Share button. And if you edited the generated text, say so, for example with “edited for style and content.”

ISO 690: no official rule, an adaptation to get approved

ISO 690 is the international standard (NF Z 44-005 in France), common in the humanities, languages, history and the arts, and used by many universities outside the English-speaking world (we covered it in our APA, ISO 690 and Vancouver guide). The catch: it has no specific format for generative AI, and we found no shared national guidance on it. The sensible fix is to apply ISO 690’s general rules for online resources, as university library guides such as the one from the Université de Lorraine library set them out: [online] after the title, the access date in square brackets, then “Available at:” and the URL.

Following the logic of the English-language styles (developer as author, model as version), that gives:

OPENAI. ChatGPT [online]. Version GPT-4o. [Accessed 23 September 2026]. Available at: https://chatgpt.com/share/…

ANTHROPIC. Claude, response to the prompt “Explain the difference between correlation and causation” [online]. Version Opus 5.5. [Accessed 27 September 2026]. Available at: https://claude.ai

These are adaptations, not an official rule. Show them to your library or your supervisor before using them throughout: some programs have their own instructions, and those take priority.

Quick comparison

Style	Where the reference goes	Author	Key element	In the text
APA 7	Reference list	The developer (OpenAI, Anthropic)	Version + [Large language model]	(OpenAI, 2023)
MLA 9	Works Cited	None	Prompt as title, model as version	(“Start of prompt”)
Chicago 18	Footnote	None (the tool opens the note)	Date + share URL	Note number
ISO 690	Bibliography	Adaptation: the developer	[online], access date	Per your system (name-date, numeric, notes)

Disclosing your use: broader than citation

MLA puts it clearly in an August 2025 post: acknowledging AI use is “slightly different from, and more comprehensive than, citation.” A citation covers a specific passage you reuse. A disclosure covers everything AI touched in your work: brainstorming, reshaping your research question, proofreading, translation, analysis code. MLA suggests notes, a section in your methods, appendices, even transcripts of the exchanges.

Universities are moving the same way. The Angers charter requires AI use to be mentioned in the final document (a dedicated section, appendix, footnote or introduction) with an explanation of exactly how it was used, and provides students with a disclosure questionnaire. The University of Cape Town in South Africa has opted for assignments in which students disclose and reflect on their AI use.

Here’s a sample statement to adapt to your situation:

Statement on the use of generative AI. I used Claude (Anthropic, Opus 5.5 model) between March and May 2026 to: (1) turn my starting question into a research problem through a series of follow-up questions; (2) check the coherence of my outline; (3) proofread the spelling in chapters 2 and 3. No passage of this thesis was written by the tool. All cited sources were consulted directly. The exchanges are reproduced in Appendix 4.

If your institution asks for prompts in an appendix, save them as you go: exporting three months of conversations the night before the deadline is a nightmare. This statement also protects you if an AI detector raises doubts, as we explain in our article on detector reliability.

Mistakes to avoid

• Citing the AI as the source of a fact. “According to ChatGPT, the youth unemployment rate is…”: no. An AI can be wrong, and MLA points out that AI tools can make up sources or misrepresent the ones they cite. It recommends consulting and citing the sources the AI links to directly, rather than the AI’s response. For a figure, a date or a definition, go back to the primary source (article, report, database) and cite that.
• Listing the AI as an author. None of APA, MLA or Chicago treat the tool as an author in the sense of someone who stands behind ideas. Responsibility for the content stays with you.
• Leaving out the version and date. Answers change from one model to the next. Without a version or date, your reference says very little.
• Giving a URL nobody can open. The private address of your chat is useless to your reader. Use a share link or the tool’s general URL, and put the exchanges in an appendix if asked.
• Staying quiet about “invisible” uses. Proofreading, translation, help with code: not a citation, but still something to disclose.
• Mixing styles. One style throughout the thesis, AI included.

What this means for you

In practice: first check what your institution’s rules and your supervisor allow. Start a small usage log from day one (date, tool, model, what you asked, what you did with it). Cite an AI response only when you’re analyzing or quoting what the AI said, never to support a fact. And write your AI-use statement as you go rather than at the last minute. It takes ten minutes a week, and it’s what will make the difference at your defense if someone on the panel asks.

FAQ

Does ChatGPT go in the reference list?

In APA and MLA, yes, if you quote or paraphrase a response: the tool gets its own entry. In Chicago, a footnote is enough. Either way, follow your institution’s instructions first.

How do I cite a Claude response in APA?

Same structure as for ChatGPT: the developer as author, the year, the tool’s name in italics, the version, [Large language model] and the URL. For example: Anthropic. (2026). Claude (Opus 5.5 version) [Large language model]. https://claude.ai. In the text: (Anthropic, 2026).

Is there an official ISO 690 rule for citing AI?

Not that we know of. You adapt the rules for online resources ([online], access date, “Available at:”) and get the format approved by your library or your supervisor.

Should I put my prompts in an appendix?

Only if your institution or supervisor asks, but it’s good practice when AI played a significant role. MLA and the Université d’Angers charter both list an appendix among the places where you can document your use.

Can I cite ChatGPT for a definition or a statistic?

No. An AI can be wrong or invent a source. Find the definition or figure in a reliable source (textbook, article, official report, database) and cite that source instead.

Further reading

• The bibliographie-normes skill: formats your references in APA, ISO 690 or Vancouver, without ever inventing missing information.
• The citer-sans-plagier skill: quote, paraphrase and summarize honestly, with or without AI.
• The memoire-problematique skill: go from your starting question to a research problem and outline, with a coach that asks questions.
• The articles-scientifiques connector: find the primary sources to cite instead of an AI response, with their APA, ISO or Vancouver reference.