Start free

How to cite AI use in academic work

Every major style guide now treats generative AI as a tool, not an author. Here is how APA, MLA, Chicago, and leading journals expect you to disclose and cite it.

The general principle

Every current style guide treats generative AI as a tool, not an author. You cite it the way you would cite a software package: with the model, developer, version, and (where applicable) the date and access information. Most guides additionally ask you to describe what the AI did: drafting, brainstorming, editing, summarizing. Transparency is the common denominator.

APA style

APA has specific guidance for ChatGPT and similar models. The basic reference form cites the developer and year.In-text: (OpenAI, 2024). APA recommends you describe the prompt and how the AI was used in the methods section. The full current guidance is on the APA style blog, which is the source of truth as it updates.

MLA style

MLA treats generative AI as a source cited by paraphrase or quotation. Their format describes what was generated and by which model.The MLA guidance on citing generative AI covers paraphrase, quotation, and derivative use separately. MLA explicitly cautions against treating AI as an author.

Chicago style

Chicago uses a note-based citation. You cite the model, developer, and date in a footnote or endnote rather than in the bibliography.Chicago recommends describing the prompt and your editorial role in the text, not just in the footnote. See the Chicago Manual of Style Q&A on AI for the current format.

Journal policies

Major journals have explicit AI policies that go beyond what style guides cover. Key examples:
  • Nature: LLMs cannot be authors. Use must be documented in methods or acknowledgments, with specific restrictions on AI-generated imagery.
  • JAMA: Authors must disclose AI use, identify which sections involved AI, and take responsibility for the content.
  • COPE (Committee on Publication Ethics): Sets the international standard that AI tools are not authors and cannot bear authorship accountability.
  • Elsevier: Specifies AI tools do not meet authorship criteria and must be disclosed in the manuscript.

What to write in a methods section

If your work used AI for anything substantive, include a short paragraph naming the model and developer, the date or version used, what it was used for (such as brainstorming, outlining, or drafting), and how you reviewed and revised the output.This level of specificity satisfies most current journal requirements and protects you if questions come up later.

What NOT to do

  • Do not list ChatGPT as an author. No major journal accepts this.
  • Do not fabricate a retrievable URL. ChatGPT conversations are not public. Cite the model, not a fake link.
  • Do not claim AI-generated content as original. If the text is AI-drafted and you reviewed it, say that. Do not pretend the writing is entirely yours.

What this has to do with detection

Citation and detection are two sides of the same coin. If you disclose AI use honestly, detection does not threaten your submission: the methods section answers the question a detector raises. It is when AI use is undisclosed and flagged that consequences follow. Our piece on humanization ethics covers the boundary between acceptable rewriting and concealment.If you want to see how your draft scores before submitting, to make sure your own voice has not been diluted by AI assistance, run it through Leap's free detector. It runs in your browser, needs no account, and highlights the sentences that weigh most on the score. Remember the score is a signal, not proof, and it should never be the only basis for any accusation. For the academic-integrity backdrop to these citation standards, read our overview of ChatGPT policies at schools and our piece on AI detectors vs plagiarism checkers: different tools, different burdens of proof.

Frequently asked questions