Quick answer. All four major citation styles now have an official format for ChatGPT and other large language models. APA 7: OpenAI. (2024). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat. MLA 9: “Prompt text in quotes.” prompt. ChatGPT, version, OpenAI, date. Chicago 17: text-and-notes or author-date with model name, version, date, and OpenAI as author. Vancouver / AMA: text-only acknowledgement plus reference list entry with model name and date. Cite when you quote, paraphrase, or use a substantive AI-generated idea. Disclose all AI assistance even if not cited.
Since 2023 every major citation manual has issued guidance on referencing generative-AI tools. The conventions are still settling; what they share is a requirement to be specific about which model, which version, and what date — because LLM outputs are non-reproducible. This guide gives you the official format from each style, plus the disclosure conventions Canadian universities now expect in dissertations and submitted papers.
Why ChatGPT Needs a Citation
ChatGPT outputs are neither published works (no fixed text, no publisher) nor personal communication (no individual author). Style manuals treat them as a third category: software with a specific corporate creator, a specific version, and a specific output you preserved. Three reasons to cite:
- Attribution. If the wording, structure, or substantive idea originated with ChatGPT, you are required to attribute it just as you would attribute any source.
- Reproducibility. A reader trying to verify your work needs the model name, version, and date so they can attempt to reconstruct the prompt-output relationship.
- Academic integrity. All Canadian universities now require explicit disclosure of AI assistance. Citing satisfies disclosure even if your institution’s rules don’t specifically demand it.
APA 7th Edition
APA classifies ChatGPT outputs as software with the author as OpenAI (the company, not the model). Reference list entry:
OpenAI. (2024). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat
In-text citation:
The model summarised the methodology as a four-step process (OpenAI, 2024).
APA also recommends quoting the actual prompt and response in an appendix when the output substantially influenced your analysis. For Claude, use Anthropic as author; for Gemini, use Google.
MLA 9th Edition
MLA treats AI outputs as containers with the prompt as the “title”. Reference list entry:
“Summarise the key methodological choices in qualitative research” prompt. ChatGPT, 4 May version, OpenAI, 14 March 2024, chat.openai.com/chat.
In-text citation:
(“Summarise the key methodological choices”)
MLA emphasises that you should not treat ChatGPT as an author because the model does not have intentionality. The prompt becomes the title; OpenAI becomes the publisher.
Chicago 17th Edition
Chicago offers two styles — Notes-and-Bibliography (humanities) and Author-Date (sciences). Both treat ChatGPT as software with corporate author OpenAI.
Notes-and-Bibliography footnote:
1. ChatGPT, response to “Summarise qualitative methods,” OpenAI, March 14, 2024, https://chat.openai.com/chat.
Author-Date in-text citation:
(OpenAI 2024)
Reference list entry (Author-Date):
OpenAI. 2024. ChatGPT. Mar 14 version. Large language model. https://chat.openai.com/chat.
Vancouver / AMA Style
Vancouver (used by ICMJE-aligned medical journals) and AMA 11 do not allow AI tools to be listed as authors but require their use to be disclosed in the methods section. Standard form:
ChatGPT (OpenAI, GPT-4, version dated March 14, 2024) was used to draft initial summaries of three included studies. All summaries were verified against the original publications and revised by the first author.
AMA 11 added a specific reference-list format in 2023:
OpenAI. ChatGPT. Version dated March 14, 2024. Accessed March 15, 2024. https://chat.openai.com/chat
Hire An AI-Free PhD Writer
- 100% human-written by a PhD specialist
- Turnitin AND AI-detector screened
- Plagiarism-free, originality report
- Money-back if either check fails
- Confidential under PIPEDA
When to Cite vs When to Disclose Only
| Use case | Required action |
|---|---|
| Direct quote from ChatGPT | Cite (in-text + reference list) |
| Substantive idea or argument generated by ChatGPT | Cite |
| Summary or paraphrase of an AI-generated explanation | Cite |
| ChatGPT used to brainstorm but ideas heavily reworked | Disclose in acknowledgements or methods, no citation needed |
| Grammar / spelling check by an AI tool | Disclose if institution requires, no citation |
| AI used to translate quotes | Cite the AI translation; note original-language source |
| AI used to generate computer code | Disclose; cite if code is a substantive contribution |
Disclosure in the Manuscript
Beyond citation, almost every journal and Canadian university now requires explicit AI-use disclosure. Standard placement:
- Methods section — if the AI was used in data collection, analysis, or generating any reported result.
- Acknowledgements section — if the AI was used in writing, drafting, or polishing prose.
- Cover letter to editor — many journals (Nature, Science, JAMA) require a separate AI-use statement in the cover letter on submission.
Standard disclosure language:
The author(s) used ChatGPT (OpenAI, GPT-4, dated [date]) to [specific task: e.g., draft initial summaries of literature, refine wording in the discussion section, generate Python code for data preprocessing]. All output was reviewed, verified, and edited by the author(s) prior to inclusion. The author(s) take full responsibility for the final content.
University-Specific Rules
- University of Toronto: AI use must be disclosed in any work submitted for assessment. Faculty members set per-course policies; default is “disclose all uses”.
- McGill: AI tools may be used if the syllabus permits, with explicit disclosure. Undisclosed AI use is a violation of the Code of Student Conduct.
- UBC: AI use governed by instructor; institution-wide policy under development.
- Waterloo, Queen’s, Western: Discipline-specific rules; check faculty graduate handbook.
For our take on the ethical landscape see Is using ChatGPT ethical for PhD research?
Frequently Asked Questions
Can I list ChatGPT as a co-author?
No. All major journals (Nature, Science, JAMA, BMJ, Cell, ICMJE) explicitly prohibit listing AI tools as authors because they cannot bear responsibility for the content.
What if I used ChatGPT just to fix grammar?
Most journals require disclosure for any AI use including grammar checking. Use a brief acknowledgement: “ChatGPT was used for language polishing of the manuscript.” No reference-list entry needed.
Do I cite Grammarly the same way?
Grammarly is treated similarly. Most journals require disclosure of any AI-assisted editing including Grammarly. Citation format follows the same software-citation pattern.
How do I cite Claude or Gemini?
Same formats, different authors and URLs. Author is Anthropic for Claude (claude.ai), Google for Gemini (gemini.google.com). Use the specific model name (Claude 3 Opus, Gemini Ultra) and version date.
Do I need to save my ChatGPT conversations?
Yes, until publication or thesis acceptance. Export the conversation to PDF and keep it for verification. Some journals now request the conversation log as supplementary material.
What about citing AI in a dissertation versus a journal article?
Identical citation format. Dissertations also typically require a stand-alone AI-use statement in the front matter (after acknowledgements). See our guide to AI in dissertations.
