How to Cite ChatGPT in Harvard, APA and MLA (UK University Rules 2026)

If you are a UK student staring at your reference list at 2am wondering how to cite ChatGPT in Harvard style without losing referencing marks — or worse, being accused of academic misconduct — you are not alone. Students at Bolton, Kaplan International College, BPP, Arden, Coventry and dozens of other UK institutions ask us the same three anxious questions every week: what is the exact format, do I italicise ChatGPT, and can my lecturer detect that I used it?

This guide answers all three. You will get copy-paste Harvard, APA and MLA references that match 2026 UK expectations, a clear ruling on the italics question, the difference between citing a prompt and paraphrasing, what Kaplan and Bolton specifically expect, whether lecturers can see your ChatGPT history, when you do not need to cite, and a declaration statement you can adapt. If you need the broader foundations first, see our general ChatGPT citation guide, our full Harvard referencing guide and our APA 7th edition guide.

Quick Answer: Copy-Paste ChatGPT References (Harvard, APA, MLA)

In a rush before a deadline? Copy the box that matches your required style, then replace the version, model and access date with your own. Citing correctly never makes undisclosed banned use acceptable — it only fixes the referencing part. If your brief says AI use is prohibited for that task, do not use it at all.

Harvard Deep-Dive: How to Cite ChatGPT in Harvard Style (UK)

Harvard is not one fixed style — it is a family of author-date styles. In the UK, however, almost all universities base their Harvard on Cite Them Right (12th edition), the standard textbook by Pears and Shields. That is the version Kaplan, Bolton, BPP, Arden, Coventry, Manchester Met and most post-92 universities point you to. So when your handbook says “Harvard” without further detail, assume Cite Them Right.

The Cite Them Right formula for generative AI

Cite Them Right treats ChatGPT as software / generative AI, not as a human author. The pattern is:

Organisation (Year) Name of tool (version details) [Large language model]. Available at: URL [Accessed date].

A real 2026 example for your reference list:

Break it down element by element so you never lose a mark:

• Author — OpenAI (2025): the company is the author, because an AI cannot be an author. Never write “ChatGPT (2025)” as the author. In-text this becomes (OpenAI, 2025). For a narrative citation: “OpenAI (2025) states that…”
• Tool name + version — ChatGPT (GPT-4o, 12 May version): include the model (GPT-3.5, GPT-4, GPT-4o, GPT-5) and the version date shown in the interface or share link. This matters because ChatGPT gives a different answer every time — the version tells your marker which ChatGPT you used. If you only know the year, write ChatGPT (2025 version).
• [Large language model]: square-bracketed descriptor so the reader knows what kind of source this is. Copy it exactly. If you used ChatGPT-4 with image input, some guides allow [Large multimodal model] — follow your handbook.
• Available at: URL: use the direct access URL (https://chat.openai.com/chat or your share link if you created one via the Share button). Do not link to the OpenAI homepage.
• [Accessed date]: the date you generated the text, e.g. [Accessed 10 October 2026]. Harvard requires this because AI output is non-retrievable — another reader typing the same prompt will get different text.

Is ChatGPT italicised in Harvard referencing?

No. This is the single most searched question — harvard citation chatgpt italicized — and the cause of many lost marks. In Cite Them Right Harvard, ChatGPT is written in roman (plain) type, not italics and not in quotation marks. Contrast this with APA 7th (italicised) and MLA 9th (italicised as container). So:

Why the difference? Harvard CTR treats the tool name as a software label in roman, while APA treats it as a titled work (like a book title) and MLA treats it as a container (like a website name). Do not mix them. If your Kaplan or Bolton handbook explicitly shows ChatGPT in italics, follow that local variant — house style always beats a blog, including this one.

Harvard in-text examples (direct quote vs paraphrase)

ChatGPT output is non-retrievable, so APA and Harvard guidance both say: give the prompt in your text, then cite. For example:

When prompted with “Summarise three causes of the 2008 financial crisis in 200 words”, the ChatGPT-generated text identified subprime mortgage securitisation, credit rating failures and excessive leverage as central factors (OpenAI, 2025).

For a direct quotation under 3 lines, use single quotation marks (UK Harvard convention) plus citation: ‘excessive leverage amplified relatively small mortgage defaults into a systemic crisis’ (OpenAI, 2025). For longer AI output, put the full transcript in an appendix and cite it: (OpenAI, 2025; see Appendix A for full transcript). Never present AI text as your own analysis — that is the fast track to a misconduct panel even if the reference is technically perfect. Learn to rewrite properly with our paraphrasing guide.

How to Cite ChatGPT in APA 7th Edition (2026 Update)

APA’s official guidance (McAdoo, APA Style Blog, updated September 2025) is stable into 2026: cite ChatGPT using the software template (Manual, section 10.10). OpenAI is the author, the year is the version year, ChatGPT is the italicised title, the version follows in parentheses, then the bracketed descriptor and URL.

The classic APA example still quoted in university slides uses the March 2023 version — OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com/chat — but you must use the year and version you actually used, not copy “Mar 14” blindly. In 2026 that will typically be a GPT-4o or GPT-5 dated version; check the footer of the interface or your share link.

• In-text parenthetical: (OpenAI, 2025).
• In-text narrative: OpenAI (2025).
• Method transparency: APA explicitly asks you to describe how you used the tool in your Method section (or Introduction for essays): the prompt, the portion of output you relied on, and — for long outputs — the full transcript in an appendix or supplementary file. Example: “When prompted with ‘Is the left-brain/right-brain divide real or a metaphor?’ the ChatGPT-generated text indicated that ‘the notion that people can be characterised as left-brained or right-brained is considered an oversimplification’ (OpenAI, 2023).”
• Vet everything: APA warns that ChatGPT confidently invents references — in their own test, one of five supplied sources did not exist. If ChatGPT gives you sources, track down and read the originals, then cite those — not the AI’s summary of them.

For general APA list rules (hanging indent, double spacing, alphabetical order), see our APA 7th edition citation guide.

How to Cite ChatGPT in MLA 9th Edition

MLA 9th takes a different philosophy: do not treat the AI as author. Instead, the Works Cited entry starts with a description of your prompt (the Title of Source element), followed by the word “prompt”, then the italicised tool name as container, version, publisher, generation date and URL. This is why MLA entries look “backwards” compared to Harvard and APA.

MLA’s own examples (Style Center, March 2023, revised August 2025) illustrate the pattern across use cases:

• Paraphrase: describe the prompt, then paraphrase in your prose with the entry above. Keep the description sentence-case (not title case) — it is a description, not a published title.
• Direct quote: name the tool in your sentence (“When asked to describe the symbolism… ChatGPT provided a summary…”), quote briefly, then list the follow-up prompt if relevant, e.g. “In 200 words, describe the symbolism of the green light” follow-up prompt to list sources. ChatGPT, 13 Feb. version, OpenAI, 9 Mar. 2025, chat.openai.com/chat.
• Editing or translating your own prose: MLA says to acknowledge all functional uses — editing, translating, brainstorming — in a note or in your text, even if you do not quote output. You do not need a Works Cited entry for every grammar fix, but you do need a transparent acknowledgement (see the declaration section below). Your tutor decides whether that level of use was permitted.
• Images or poems: caption AI images with prompt + tool + version + date; for a generated poem, use its first line or assigned title as the description.

If ChatGPT now gives you a Share link, MLA advises using that unique URL instead of the general chat.openai.com/chat address. And if the AI cites secondary sources, click through and verify them — hallucinated citations are endemic. Cite the source you actually read.

Kaplan, Bolton and Other UK University Variants

Students searching “Kaplan Harvard ChatGPT” usually study at Kaplan International College / Kaplan Living Learning (­pathway programmes articulating to UK universities) or Kaplan Business School partners. Kaplan’s study-skills guides specify Cite Them Right Harvard — i.e. the roman-type format in the Harvard box above. Kaplan markers are strict on two points: the access date and the version. A reference without “[Accessed …]” or without a model/version routinely loses the referencing mark, and repeated sloppy AI references can trigger an academic-integrity tutorial.

University of Bolton likewise uses a Harvard variant based on Cite Them Right (see the Bolton Library referencing guide). The expected form is identical in substance — OpenAI as author, ChatGPT in roman, [Large language model], Available at + [Accessed date] — though Bolton’s examples historically render the access date as “(Accessed: 3 April 2025)” with a colon and parentheses rather than square brackets. Both are accepted provided you are consistent. The same applies at BPP, Arden, Coventry, Hertfordshire and most modern UK universities: same content, minor punctuation house-style.

Three practical tips for Kaplan/Bolton students:

1. Copy your handbook’s punctuation exactly. If your PDF shows “Available at: URL (Accessed: date)”, use parentheses. If it shows “[Accessed date]”, use square brackets. Consistency across the whole list matters more than which variant you pick.
2. File the evidence. Save your ChatGPT share link or export the transcript as a PDF and keep it until after results. Kaplan and Bolton tutors increasingly ask for it when AI use is declared.
3. Declare even “small” uses. Both institutions treat undisclosed substantive AI content as misconduct. A grammar polish with a declaration is a non-issue; the same polish hidden and then flagged by a detector becomes a panel hearing. When in doubt, declare.

How to Cite Your Prompt vs Paraphrasing AI Text

A frequent confusion: do you cite the question you asked or the answer you used? The answer is both — the prompt identifies which output you mean, and the reference authenticates it. Think of the prompt as the title of an unrepeatable conversation.

• If you quote AI wording directly (even a single sentence), use quotation marks, give the prompt in your text, and cite. Example (Harvard): When prompted with “Define academic integrity in 50 words”, ChatGPT returned ‘the commitment to honesty, trust and responsibility in scholarship’ (OpenAI, 2025).
• If you paraphrase AI ideas (same ideas, your words), you still cite — changing words does not remove the debt. Example: ChatGPT-generated summaries of the 2008 crisis foreground securitisation and leverage (OpenAI, 2025). Then demonstrate your own analysis on top; a paragraph that only restates AI output earns no criticality marks. Practise proper technique with our paraphrasing guide.
• If you used AI for ideas but cite real sources: the gold standard. Use ChatGPT to orient yourself, then read the journal articles or books it pointed to, verify they exist, and cite them. Your reference list then contains peer-reviewed sources, with the AI acknowledged separately in a declaration — far stronger than a list with a lone ChatGPT entry.
• Keep prompts concise in MLA: MLA allows a shortened but accurate description rather than pasting a 400-word prompt verbatim. “Summarise three causes of the 2008 crisis” prompt is fine if that faithfully describes what you asked.

One hard rule across all three styles: ChatGPT is never a substitute for a scholarly source. No lecturer awards evidence marks for “(OpenAI, 2025)” where a peer-reviewed citation was expected. Cite AI for transparency; cite scholarship for authority.

Can Your Lecturer Detect ChatGPT History? (The Honest 2026 Answer)

Short answer: no, your lecturer cannot see your ChatGPT history — but that does not mean undisclosed use is safe.

OpenAI does not give universities access to your account, chats, timestamps or prompts. There is no “lecturer portal” into ChatGPT. Deleting chats or using temporary chats removes them from your sidebar (subject to OpenAI’s retention policies), and no UK university can subpoena that data for a routine essay. So the specific fear — “can lecturer detect ChatGPT history?” — can be retired.

What lecturers can do in 2026 is broader and, frankly, more effective than reading your history:

• Turnitin AI detection + originality reports: Turnitin’s AI writing indicator scores passages for likelihood of AI generation alongside classic text-matching. It is probabilistic, not proof — false positives happen, especially with formulaic or EAL writing — and several universities (and Turnitin guidance itself) warn against treating the score as a sole verdict. But a high AI score plus a missing declaration reliably triggers a conversation, a viva, or a request for drafts and notes.
• Style discontinuity: markers who have read your seminar posts, drafts or previous essays notice overnight jumps in vocabulary, syntax or referencing sophistication. Sudden perfection is a signal.
• Hallucinated or non-existent references: fabricated DOIs, wrong years, journals that never published the article — classic AI fingerprints. One invented citation damages credibility more than any detector.
• Viva voce and process checks: many UK modules now reserve the right to viva any submission. “Explain this paragraph in your own words” defeats ghostwritten text quickly. Draft histories, notes and tutorial engagement all count as evidence of authorship.
• Declaration cross-checks: if your cover sheet says “no AI used” and the detector, style and references suggest otherwise, the dishonesty element — not the AI use itself — becomes the misconduct charge, which carries heavier penalties.

When NOT to Cite ChatGPT (Grammar Checks vs Content)

Not every contact with AI needs a reference-list entry. UK guidance converges on a content-vs-polish distinction — but the boundary is set by your handbook, so treat this as a starting point, not permission.

• Usually NO reference entry needed (but often still a declaration): spell-checking, basic grammar correction, rewording your own sentence for clarity where ideas and evidence are unchanged, formatting references you already wrote, or using AI as a thesaurus. MLA calls these “functional uses” to be acknowledged in a note; many UK handbooks want a one-line declaration even for proofreading-level help.
• ALWAYS cite + declare: any AI-generated idea, argument, structure, summary, data interpretation, literature synthesis, case application, or wording you quote or paraphrase. If removing the AI would remove content from your essay, it must be cited.
• NEVER acceptable even with citation: submitting largely AI-written work as your own where the brief requires independent authorship; using AI in closed-book or invigilated assessments; fabricating AI-invented sources as if you read them. A reference does not launder misconduct — it only documents it.

Rule of thumb: if you are unsure whether it counts as content, treat it as content — cite it and declare it. Over-transparency has never triggered a misconduct panel; under-transparency routinely does.

AI Declaration Statement Example (UK Universities)

Since 2024, most UK universities require — or strongly recommend — a short AI use statement separate from the reference list, often on the cover sheet, in a footnote, or in a methodology/acknowledgement section. It states what tool you used, for what, and affirms the work is yours. Adapt this template (Harvard student, 2026):

AI use declaration: I used OpenAI ChatGPT (GPT-4o, May 2025 version) to brainstorm initial essay structure and to check grammar on my own draft. All arguments, analysis and conclusions are my own; all AI-generated wording was rewritten and verified against scholarly sources. Full prompts and outputs are available on request. Reference: OpenAI (2025) ChatGPT (GPT-4o, 12 May version) [Large language model]. Available at: https://chat.openai.com/chat [Accessed 10 October 2026].

Variants for different levels of use:

• No AI used: “I confirm no generative AI tools were used in producing this work beyond standard spell-checking.”
• Polish only: “I used ChatGPT solely to proofread my own draft for grammar and clarity. No content, ideas or references were AI-generated.”
• Substantive permitted use: describe prompts, dates and which sections were informed, plus the appendix reference. Keep the share link.

Place the declaration where your brief says — usually the assignment cover sheet or immediately before the reference list. If the brief provides its own wording, use theirs verbatim.

Frequently Asked Questions

Final Thoughts

Citing ChatGPT correctly in 2026 is straightforward once you separate the two jobs: format and permission. Format is mechanical — Harvard (OpenAI, year, roman ChatGPT, version, [Large language model], Available at + Accessed date), APA (OpenAI. year. ChatGPT version [Large language model]. URL) or MLA (prompt first, italicised ChatGPT container, version, OpenAI, generation date). Copy the boxes above, swap in your version and dates, and your referencing marks are safe.

Permission is where students get caught. No citation rescues work your brief forbade you to generate with AI, and no deleted chat history protects undisclosed substantive use from detectors, vivas and hallucinated references. Use AI for what your handbook allows, verify every claim against real scholarship, declare everything in one honest sentence — and keep your share links until results day. Do that, and ChatGPT becomes what it should be: a study aid that speeds up orientation, not a shortcut that costs you your degree.