How to Cite AI-Generated Content in APA, MLA, and Harvard (2026 Rules)

Knowing how to cite AI-generated content has become one of the most pressing practical questions in academic writing. When a student uses ChatGPT to brainstorm a literature review outline, asks Claude to summarise a methods chapter, or queries Gemini about statistical procedures, the output is a source — and in 2026, every major citation style has published clear official guidance on how to document it. Yet citation errors persist, partly because the rules were published in stages across 2023–2025 and partly because the three dominant styles — APA 7th edition, MLA 9th edition, and Harvard (Cite Them Right) — each treat the AI tool differently.

This article consolidates the current official guidance from the APA Style blog, the MLA Style Center, and the Cite Them Right Harvard framework into a single reference document. Each section provides the exact template, a worked example for ChatGPT, Claude, and Gemini, and an explanation of the underlying logic so you can adapt the format to tools not covered here. The article also addresses when citing AI is — and is not — appropriate, and provides transparency statement templates that satisfy the disclosure requirements of most UK and US universities.

Why You Must Cite AI-Generated Content

The ethical foundation of citation has always been traceability: a reader should be able to locate and evaluate the sources that shaped an argument. AI-generated text satisfies this standard imperfectly — it cannot be re-generated identically — but the obligation to acknowledge it remains. All three major style authorities agree on a core principle: if AI-generated content influenced the ideas, language, data, or structure of your work, the reader must know.

Beyond ethical obligation, there is a legal and institutional dimension. In 2026, the majority of UK universities (including members of the Russell Group) and US research universities have revised their academic integrity policies to require explicit disclosure of AI use. Presenting AI-generated text without attribution is treated as a form of undisclosed assistance — functionally equivalent to the kind of contract cheating that triggers formal misconduct proceedings.

The question of what universities formally permit is distinct from how you cite what you did use. For a full breakdown of which AI tools are allowed for thesis writing, how Grammarly differs from generative AI in institutional policy, and what the disclosure obligations are, the companion guide on whether Grammarly and AI editors are allowed for a thesis in 2026 covers the full policy landscape across UK and US universities.

There is also a methodological argument for citation. When AI tools are used to process, synthesise, or generate content in a research context, they constitute a step in the research process. A Methods section or methodology chapter that omits this step is inaccurate regardless of whether the institution requires disclosure. Citing correctly is, in this sense, not merely a compliance act but a precision act.

APA 7th Edition: Full Rules and Examples

The American Psychological Association first published guidance on citing ChatGPT in April 2023 and has since updated that guidance to cover the range of generative AI tools now in common academic use. The APA Style blog post Citing generative AI in APA Style: Part 1 — Reference formats (published at apastyle.apa.org) is the authoritative source. The framework below reflects the current guidance as of May 2026.

Core Principle: The Developer Company as Author

APA does not treat the AI itself as an author. Authorship in APA requires human accountability for the content. Instead, the company that developed and operates the model is listed as the author. For ChatGPT, that is OpenAI; for Claude, that is Anthropic; for Gemini, that is Google.

APA distinguishes between two scenarios: citing a specific chat (where a stable, shareable URL exists) and citing the tool generally (where no specific output is referenced).

Template: Specific Chat Session (with Shareable URL)

Author/Company. (Year, Month Day). Title of chat in sentence case [Generative AI chat]. Model name. URL

Template: Tool Generally (No Specific Chat)

Author/Company. (Year). Tool Name [Bracketed descriptor]. URL

Worked Examples

ChatGPT — specific chat:

OpenAI. (2026, May 15). Strategies for reducing measurement error in survey research [Generative AI chat]. ChatGPT GPT-4o. https://chatgpt.com/share/abc12345

ChatGPT — tool generally:

OpenAI. (2025). ChatGPT [Large language model]. https://chatgpt.com/

Claude (Anthropic) — specific chat:

Anthropic. (2026, May 20). Essential grammar topics for advanced academic writing [Generative AI chat]. Claude Sonnet 4. https://claude.ai/share/329173b2-ec93-4663-ac68-4f65ea4f166d

Gemini (Google) — tool generally:

Google. (2025). Gemini [Large language model]. https://gemini.google.com/

In-Text Citations (APA)

In-text citations follow the standard APA author–year format:

• Parenthetical: (OpenAI, 2026)
• Narrative: OpenAI (2026) generated the following summary…

Additional APA Requirements

The APA Style blog recommends that when you quote AI output directly, you include the prompt that produced it in the body of the text or in a footnote, and that you append the full text of AI-generated content used substantively as an appendix. This is because AI responses cannot be independently reproduced, making standard reproducibility expectations unmet. For full details on applying the APA 7th edition framework to all source types, see the APA citation format guide covering every source type.

MLA 9th Edition: Full Rules and Examples

The Modern Language Association updated its guidance on citing generative AI through a post titled How do I cite generative AI in MLA style? (Updated and Revised) published at style.mla.org. MLA applies its container model — the same framework used for citing essays within books or articles within journals — to AI-generated content. The approach is meaningfully different from APA and reflects MLA’s emphasis on the research process.

Core Principle: The Prompt as Source Title, the Tool as Container

MLA does not list the AI company or the AI tool as the author of the work. The AI has no authorial voice in the MLA framework. Instead, the user’s prompt is treated as the title of the source (placed in quotation marks), and the AI tool (e.g., ChatGPT, Claude) is the container (italicised). This makes practical sense: the specific content generated is a direct response to a specific prompt, making the prompt the identifying descriptor of that particular piece of content.

Works Cited Template

"Prompt text as submitted to the AI tool" prompt. Tool Name, model version, Publisher/Company, Day Month Year, URL.

Worked Examples

ChatGPT:

“Explain the difference between deductive and inductive research approaches” prompt. ChatGPT, model GPT-4o, OpenAI, 26 May 2026, chatgpt.com/share/abc12345.

Claude:

“Summarise the key arguments for and against snowball sampling” prompt. Claude, model Claude Sonnet 4, Anthropic, 20 May 2026, claude.ai/share/329173b2.

Gemini:

“List the main limitations of a cross-sectional research design” prompt. Gemini, model Gemini 1.5 Pro, Google, 18 May 2026, gemini.google.com/app/share/def67890.

In-Text Citations (MLA)

In MLA, parenthetical citations reference the first element of the Works Cited entry — which, for AI citations, is the prompt text. Use a shortened version if the prompt is long:

• Parenthetical (full short prompt): (“Explain the difference”)
• Parenthetical (longer prompt): (“Summarise the key arguments”)

MLA’s Guidance on Not Citing AI Summaries of Other Sources

MLA explicitly advises researchers to verify and cite the original sources that an AI tool references, rather than citing the AI’s summary of those sources. If ChatGPT describes a finding from a 2019 paper, locate that paper, read it, and cite it directly. This is both an integrity requirement and a practical safeguard against the hallucination risk inherent in current generative models. AI tools regularly fabricate citations that look real but resolve to non-existent papers; the guide to the best AI citation checkers for detecting fake and hallucinated references covers the tools that verify your bibliography against live academic databases. For full MLA formatting rules across all source types, the MLA format guide for 2026 provides comprehensive coverage of MLA 9th edition.

Harvard (Cite Them Right): Full Rules and Examples

The Harvard referencing system used in most UK universities does not have a single governing body equivalent to the APA or MLA. The de facto standard is Cite Them Right (published by Palgrave Macmillan), which has issued specific guidance on citing generative AI. Some universities — including UCL and Edinburgh — have published their own Harvard AI citation guidance that broadly aligns with Cite Them Right.

Core Principle: Corporate Author with Prompt as Title

Harvard treats the AI tool’s developer as a corporate author. The prompt submitted is italicised as the title of the source (since AI output is unique to the prompt). The citation records both the URL and the access date, because AI-generated content is not persistently archived in the way that published documents are.

Template: With Shareable URL

Name of AI company/tool (Year) Prompt submitted to AI in sentence case. AI tool name response to [Your Name]. Available at: URL (Accessed: Day Month Year).

Template: Without Shareable URL (Personal Communication)

Name of AI company/tool (Year) AI tool name response to [Your Name], Day Month.

When no shareable URL is available, the session cannot be independently accessed by readers. Harvard treats this equivalently to a personal communication: it appears in the in-text citation but is typically not included in the reference list, as it is not retrievable. Check your institution’s specific Harvard guide, as practices vary.

Worked Examples

ChatGPT — with shareable URL:

OpenAI ChatGPT (2026) What are the main ethical considerations in qualitative research with vulnerable populations? ChatGPT GPT-4o response to J. Smith. Available at: https://chatgpt.com/share/abc12345 (Accessed: 26 May 2026).

Claude — with shareable URL:

Anthropic Claude (2026) Describe the key differences between grounded theory and thematic analysis. Claude Sonnet 4 response to A. Patel. Available at: https://claude.ai/share/329173b2 (Accessed: 20 May 2026).

Google Gemini — without shareable URL (personal communication):

Google Gemini (2026) Gemini response to T. O’Brien, 18 May.

In-Text Citations (Harvard)

• Paraphrasing: (OpenAI ChatGPT, 2026)
• Direct quote: “…” (OpenAI ChatGPT, 2026, p. n/a — AI output has no page numbers; omit the page element)
• Narrative: OpenAI ChatGPT (2026) noted that…

For a thorough grounding in Harvard referencing conventions beyond AI sources, the Harvard referencing guide with 2026 rules, examples, and templates covers the full range of source types.

Side-by-Side Comparison Table

The table below summarises the structural differences between the three styles for citing a ChatGPT session with a shareable URL, accessed on 26 May 2026.

When to Cite vs. When to Acknowledge Without Citing

Not every interaction with an AI tool requires a formal citation entry. Style authorities and universities generally distinguish between three tiers of AI involvement, each with different disclosure requirements.

Tier 1: Content Use — Formal Citation Required

A formal citation in the reference list (or Works Cited, or reference section) is required when AI-generated text appears directly in the submitted work — whether quoted verbatim or paraphrased — and when AI-generated analysis, interpretations, or data summaries shape the argument. This includes: reproducing a paragraph edited from an AI draft, using an AI-generated outline that became the structure of the paper, or incorporating an AI-generated table or list.

Tier 2: Process Assistance — Acknowledgement Without Full Citation

When AI was used as a writing process tool — grammar checking, brainstorming, finding synonyms, improving sentence flow — without contributing any content that appears in the final document, a formal reference entry is typically not required. However, a brief acknowledgement in the Methods section, a footnote, or an author note is still expected by many institutions. MLA notes explicitly that process uses “do not require citation but should be acknowledged.”

Tier 3: No Disclosure Needed

Using an AI-powered spell-checker or predictive text — the kind embedded in word processors for decades — generally falls outside disclosure requirements. The boundary is less clear for tools like Grammarly with its advanced AI suggestions. When in doubt, disclose. The institutional risk of under-disclosure is far greater than the cost of including a one-sentence acknowledgement. For a detailed analysis of what AI tools universities formally classify as permissible, the guide on whether you can use AI to write your dissertation in 2026 maps out university policy across the UK and US.

The Non-Shareable Session Problem

A structural difficulty common to all three citation styles is the non-shareable session: AI conversations that cannot be linked to a stable public URL. ChatGPT’s share feature, Claude’s shared conversation links, and similar tools in Gemini address this for users who enable sharing. Where sharing is disabled or unavailable, APA recommends including the full text of the AI-generated content as an appendix. MLA suggests providing the general tool URL. Harvard treats it as a personal communication. In all cases, including the exact prompt in the text itself provides at least partial transparency.

Transparency Statement Templates

A transparency (or AI use) statement is a brief, structured disclosure placed either in the Methods section (for research papers and dissertations) or in an Acknowledgements note (for coursework essays). Many journals, including those indexed in Web of Science and Scopus, now require one as a condition of submission. The following templates are adaptable to all three citation styles.

Template A: Limited Use (Editing Only)

Template B: Substantive Use (Content Generation)

Template C: Research Methods (Dissertation / Thesis)

These templates align with the transparency requirements published by the Committee on Publication Ethics (COPE) and the disclosure norms at major UK and US research universities. The same transparency logic that applies to AI citations applies to recycled prior work — see the detailed guide on self-plagiarism and reusing your own work in a thesis for the parallel disclosure framework.

Before submission, verify your entire reference list — including any AI-generated citations — against a live academic database. Tools designed for this task are ranked in the guide to the best AI citation checkers that catch fake and hallucinated references in 2026.

Frequently Asked Questions

External sources used in preparing this article: APA Style blog — How to cite ChatGPT; APA Style blog — Citing generative AI: reference formats; MLA Style Center — How do I cite generative AI? (Updated); RGU Library — Harvard Cite Them Right: Generative AI templates.