How to Cite ChatGPT, Claude, and Gemini in Academic Papers (APA, MLA, Chicago)

Academic citation practice has caught up with AI tools. As of 2026, APA, MLA, and Chicago all have published guidance on how to cite AI-generated content, and the majority of universities have formal policies requiring disclosure of AI use even when no AI-generated content appears verbatim. This guide covers what to cite, what to disclose, and the specific formatting conventions across the three major style manuals.

If you are looking for the broader picture of when AI use is appropriate at all, see our AI academic integrity guide. If you want the pattern for citing AI-generated content outside academic papers (blog posts, essays, creative writing), see how to cite AI-generated content. This piece is the academic-specific reference for APA 7, MLA 9, and Chicago.

Should You Cite AI-Generated Content At All?

The short answer in 2026 is: cite or disclose, always. The specific mechanism depends on how you used the tool.

Three distinct use cases

Verbatim output. If AI-generated text appears in your paper — a sentence, a paragraph, a summary you did not rewrite — that output needs to be cited. The reader needs to know which words originated with the AI and which words are yours.

Substantive shaping. If AI shaped the paper without appearing verbatim — the AI helped you outline, the AI suggested the argument structure, the AI summarised papers that you then drew conclusions from — this needs to be disclosed in a methods section or acknowledgement. A reference list entry is usually not required, because no AI text appears in the paper.

Mechanical use. If AI was used only for mechanical tasks — grammar checking, rephrasing for clarity, formatting — disclosure is usually not required. Most institutions treat AI grammar tools the same as human copy-editing: expected, standard, not separately cited.

The boundary between "substantive shaping" and "mechanical use" is where most real-world disputes happen. A safe rule: if you would be embarrassed to tell a reviewer about the AI use, you should probably be disclosing it.

The institutional landscape

Most universities now have formal AI-use policies. These policies generally fall into three categories:

• Full disclosure required for all use. Some institutions, especially in graduate programmes, require a one-paragraph disclosure statement on every submission describing which AI tools were used, how, and to what extent.
• Disclosure required for substantive use. The most common stance. Grammar and formatting tools do not need disclosure; anything that shaped content does.
• Discipline-specific policies. Law schools, medical schools, and some humanities departments have stricter policies than the university-wide default.

Check your institution's and specific course's policies. When in doubt, over-disclose rather than under-disclose.

APA 7 Format for AI Tools

APA 7 updated its AI guidance substantially in 2023 and has refined it through subsequent revisions. The current standard treats AI output as the work of the organisation that built the tool, not as a personal communication (which was an earlier interim guidance).

APA 7 reference list entry

The standard format is:

Organization. (Year). Tool name (Version) [Large language model]. URL

Concrete examples:

OpenAI. (2026). ChatGPT (GPT-4o) [Large language model]. https://chat.openai.com

Anthropic. (2026). Claude (3.5 Sonnet) [Large language model]. https://claude.ai

Google. (2026). Gemini (1.5 Pro) [Large language model]. https://gemini.google.com

A few specific conventions:

• The tool name is italicised.
• The version goes in parentheses after the tool name. If you cannot find the exact version, note the date of use in the methods section instead.
• The descriptor [Large language model] is in square brackets and not italicised.
• The URL is to the tool's main interface, not to a specific conversation (which is not retrievable anyway).

APA 7 in-text citation

(OpenAI, 2026) (Anthropic, 2026) (Google, 2026)

For direct quotations of AI output, include the date of the specific query if you have it. This handles the non-retrievability of AI conversations by anchoring the citation to a specific interaction.

APA 7 methods-section disclosure

For any substantive AI use, APA 7 recommends a methods-section or author-note disclosure. The content should cover:

• Which tools were used (name and version).
• What they were used for.
• How the output was verified.

A template:

This paper was drafted with the assistance of ChatGPT-4o (OpenAI, 2026) for outlining and initial paragraph drafting. All factual claims and citations were independently verified by the author against primary sources. No AI-generated text appears verbatim in the submitted manuscript.

For papers where AI-generated text does appear verbatim, the template shifts:

Paragraphs X, Y, and Z include AI-generated content from ChatGPT-4o (OpenAI, 2026), lightly edited for clarity. All citations within those paragraphs were independently verified against primary sources.

See our APA 7 citation guide for the full style manual coverage beyond AI use.

APA 7 Appendix: Archiving Prompts and Responses

APA 7 recommends archiving the prompts you used and the responses you received, because AI conversations are not retrievable by readers. The standard mechanism is an appendix.

A typical appendix entry:

Appendix A: ChatGPT Prompts and Responses

Date: 4 April 2026

Prompt: Outline the main arguments for and against a universal basic income from an economic efficiency perspective.

Response: [verbatim response, in a block quote]

If you used AI across multiple sessions, archive each session separately. The point is that a reviewer who wants to see what the AI actually produced can see it, even though the conversation itself is not retrievable through the tool.

MLA 9 Format for AI Tools

MLA 9 treats AI tools as containers and applies the same general container framework it uses for other digital sources. The key difference from APA 7 is that MLA 9 uses the prompt as the equivalent of a "title" for the specific AI output.

MLA 9 works-cited entry

The standard format:

"Prompt text, truncated if long" prompt. Tool name, version, Organization, date of interaction, URL.

Concrete example:

"Summarize the main arguments for universal basic income" prompt. ChatGPT, GPT-4o, OpenAI, 4 Apr. 2026, https://chat.openai.com.

Conventions:

• The prompt goes in quotation marks followed by the word "prompt" (or "response" if what you are citing is a specific response to a multi-turn conversation).
• The tool name is italicised.
• The version follows the tool name, separated by a comma.
• The date is the date of the interaction.

MLA 9 in-text citation

("Summarize the main arguments")

MLA 9 uses shortened titles of the works-cited entry for in-text citation. Since the "title" is the prompt, the in-text citation is a shortened version of the prompt in quotation marks.

MLA 9 disclosure conventions

MLA 9 is more permissive than APA 7 about where disclosure happens. A methods section is appropriate for research papers; an acknowledgement or author's note is acceptable for shorter pieces; a footnote is often used for passing references to AI assistance.

Our MLA 9 citation guide covers the full style manual if you need the broader reference.

Chicago Format for AI Tools

Chicago has two systems — notes and bibliography (NB), favoured in humanities, and author-date, favoured in sciences. AI citation works in both, with slightly different conventions.

Chicago Notes and Bibliography

A full first note:

1. ChatGPT, response to "Summarize the main arguments for universal basic income," OpenAI, April 4, 2026, https://chat.openai.com.

Subsequent short note:

2. ChatGPT, response to "Summarize the main arguments."

Bibliography entry:

OpenAI. ChatGPT. Accessed April 4, 2026. https://chat.openai.com.

Chicago NB treats the specific prompted interaction as an unpublished source, broadly similar to how it handles personal communications. Some Chicago-NB style guides prefer that AI use be disclosed in a preface rather than cited, especially when the AI was used for substantive shaping rather than verbatim text.

Chicago Author-Date

Reference list entry:

OpenAI. 2026. ChatGPT (GPT-4o). Large language model. https://chat.openai.com.

In-text citation:

(OpenAI 2026)

Chicago author-date is close to APA 7 in structure, with minor formatting differences (no comma after author, different handling of version information).

Our Chicago style citation guide has the full manual if you want to look up non-AI conventions.

What to Disclose vs Cite

A useful way to decide between a reference-list entry and a methods-section disclosure:

The last row is important. If you used CiteDash's deep research or Elicit to find papers, you cite the papers themselves — not the tool that found them. The tool is a research aid, like a library catalogue; the sources it surfaces are what you cite.

The exception is if the tool's summary or synthesis appears verbatim in your paper. In that case, you cite both the underlying sources and disclose the tool in your methods. This is analogous to citing a review article you quoted alongside the primary studies.

Institutional Policies: What to Check

Before finalising a citation approach, check three policy documents:

Your university's overarching AI policy

Most universities now have a top-level AI-use policy. Find the current version — these have been revised frequently since 2023 — and note what it requires.

Your department or programme's supplementary policy

Individual departments often add requirements on top of the university policy. Law schools, medical schools, graduate programmes in humanities, and computer science programmes are common examples where supplementary policies are stricter.

The specific course syllabus

Course-level policies can be stricter than department policies. A professor who forbids AI use entirely in a specific assignment has the authority to do that, and the forbiddance takes precedence over a permissive department policy. The syllabus is usually where the operational rule appears.

The submission guidelines for journals or conferences

For research papers intended for publication, the relevant policy is the journal's or conference's. Most major academic publishers added AI-use policies in 2023–2024 and have been refining them. Check the specific submission requirements.

Sample Methods-Section Wording

Here are several ready-to-use methods-section statements for common patterns. Adapt as appropriate.

Minimal AI use (brainstorming and grammar only)

AI assistance in preparing this manuscript was limited to brainstorming support (ChatGPT-4o, OpenAI, 2026) and grammar/style checking (Grammarly Pro). All substantive content was drafted, researched, and verified by the author. No AI-generated text appears in the manuscript.

Moderate AI use (literature scoping, outlining)

Initial literature scoping for this review was conducted using CiteDash's deep research tool (CiteDash, 2026), which retrieves papers from Semantic Scholar, OpenAlex, and CrossRef. All cited papers were independently reviewed by the author. ChatGPT-4o (OpenAI, 2026) was used for outlining support. All final text was written and verified by the author.

Substantial AI use (AI-drafted paragraphs)

This manuscript was drafted collaboratively with ChatGPT-4o (OpenAI, 2026). Draft paragraphs were generated by the model, then edited, fact-checked, and integrated by the author. All citations in the final manuscript were independently verified against primary sources via CrossRef and Semantic Scholar. The author takes full responsibility for the accuracy of all claims and citations.

Research synthesis tool use

Research synthesis for this paper was supported by [tool name] (developer, 2026), a retrieval-first academic AI system that searches peer-reviewed literature before generating summaries. All claims attributed to specific sources were verified by the author against the primary papers. The author is responsible for all interpretations and conclusions.

Final Notes on AI Citation Practice

Three pieces of advice that do not fit cleanly into a style-manual section:

Do not cite AI output as a factual source. If you ask ChatGPT a factual question and get an answer, the citation you need is to the underlying source, not to ChatGPT. Verify the source exists. Cite it directly. This is the specific pattern that has led to most of the citation fabrication problems in 2023–2025 — students treating AI output as authoritative and citing ChatGPT as if it were a reliable reference.

Archive, archive, archive. AI conversations are not retrievable. If you plan to cite AI output, save the prompt-response pair immediately, ideally in a persistent appendix or supplementary material. Relying on being able to reproduce the conversation later is a bad bet — the model may have been updated, the specific response is stochastic, and even re-running the same prompt rarely produces identical output.

When in doubt, over-disclose. There is almost no penalty for disclosing AI use that was arguably not required to disclose. There is a very real penalty for not disclosing AI use that was required. In 2026, err toward transparency. The norms are still settling and the conservative reading is the safer one.

Citation practice around AI is still evolving, and the style manuals have been updating their guidance quickly. The specific format conventions above are current as of April 2026. Check the style-manual publisher's website for any subsequent revisions before final submission. For most undergraduate and graduate work, the combination of a reference-list entry for verbatim AI use and a methods-section disclosure for substantive AI use is the practice that maps cleanly onto every current style guide — and the practice that will be defensible under any future revision.