How to Cite ChatGPT and AI Tools

Both lines are APA's own published examples, copied as printed on APA's AI references page. Cite the chat when a share link makes the exact exchange retrievable and seeing it would help your reader. Cite the tool when it would be "unhelpful, unethical, or otherwise inappropriate to cite a specific chat" — APA's own list includes editing or analysing your own writing, translating, working on code, and research where a reference per chat would compromise participant confidentiality or produce an unusably long reference list.

In-text, both forms key off the company and that reference's year: (OpenAI, 2026) or OpenAI (2026). The whole difference lives in the reference list.

Author, date, title, source — APA's ordinary four elements. The author is the company, because APA defines that element as whoever is "responsible for a work", and an AI cannot be an author "because it is not a living, conscious human who can give consent" to the responsibilities authorship carries. The date is the specific year, month and day the chat occurred or concluded, not just a year. The title is the chat's own title, in italic sentence case, followed by a bracketed description — [Generative AI chat] is APA's example, and APA says the wording is flexible. The source element opens with the tool (ChatGPT, or a model name like ChatGPT-5, not italic here) and ends with the chat's URL.

Two details that catch people out. The chat form takes no version parenthetical — APA names the tool in the source element instead. And rename the chat in ChatGPT before you cite it: that is APA's own advice, because the title is what your reader sees. Don't lift the browser tab title either. On APA's own example share page the tab reads "ChatGPT - Top 10 Writing Topics" while APA's reference reads Top 10 writing topics for high school graduates — the tab title is both truncated and title-cased.

Paste the share link into the generator and it resolves to this form, or build it under Manual Entry → AI conversation.

Here the tool name is the title and it carries the italics — ChatGPT, Gemini 3 Flash — followed by [Large language model], which is never italic. The date is the year the tool was most recently updated, falling back to the site's copyright date if it isn't stated. The source element is only the tool's URL, because the author and the publisher are the same company and APA names it once.

So which year do you actually type? In practice, the current one. You can't audit a tool's release history, and you don't need to: these tools ship constantly — ChatGPT alone shipped updates throughout 2026 — so the year it was last updated and the year you're writing in are the same year. That's not a shortcut around the rule, it's the rule's own fallback: a live tool's copyright date is the current year. Don't write (n.d.) — a tool that is being updated has a date, and "no date" says the opposite. The only time to reach for an older year is a tool that has actually stopped: a retired model, or a version frozen at a known point.

This is also why the two references carry different dates. The chat is dated the day your conversation happened; the tool is dated the year the tool was last updated. Those are facts about two different things, so they can disagree — and in-text they usually collapse to the same thing anyway, because both come out as the maker and a year.

Version numbers are gone. Under its own heading "Version numbers in AI references—no more", APA's current post says it no longer advises including a version number by default: give the model name if there is one (ChatGPT-5), or just the tool name (ChatGPT). ChatGPT no longer publishes version numbers at all. If a tool does show one you may still add it, in parentheses and unitalicised, after the name — but the dated (Mar 14 version) form from APA's 2023 post is retired, and that is the single thing most AI-citation guides still get wrong.

The third line is MLA's own worked example. The prompt still leads: MLA puts a description of what was generated in the Title of Source element, and it still does not recommend treating the AI tool as an author. Note that the word "prompt" sits outside the closing quotation mark, the tool is the italic container, and the publisher comes before the date. MLA's in-text form is the shortened description — ("Describe the theme") — never the company.

Quote the prompt, or describe it — the punctuation differs. Quotation marks belong to the prompt's actual text, as in MLA's example above. If you'd rather give a short description of the exchange, MLA's own alternative reads: Theme of nature in Austen's Mansfield Park prompt. No quotation marks there, and no italics — under MLA's rule for untitled works you "capitalize the description sentence-style... and do not italicize it or enclose it in quotation marks." So quoted prompt text keeps its quotes and a description takes none; either way it is sentence case, never title case.

Two elements changed in MLA's August 2025 revision. Version now carries the model, written model GPT-4o, not a dated "13 Feb. version". And Location now prefers a stable, shareable URL for the conversation, falling back to the tool's general URL only if the tool has no share feature. Everything else in MLA's advice stands.

MLA's current guidance is citing generative AI, updated and revised, published 13 August 2025. Its March 2023 post is still online and now carries its own supersession note, so don't work from that one.

APA's 2023 advice was to cite only the tool, because at the time a specific chat could not be retrieved and shared — and retrievability is a core requirement for most APA references. Most tools, ChatGPT included, now give each conversation its own URL and title, which is precisely why the chat reference now exists. So retrievability is the question you answer, not a blanket impossibility: generate the link and you can cite the conversation; don't, and you are citing the tool.

Your prompt is still not part of the APA reference, and APA gives four reasons: it isn't one of the four reference elements, it doesn't help a reader retrieve the work, it can't be reported fully and accurately in one line (most chats are several refinements), and it would usually run too long. So describe it in your text, or put it in full in your Method section or an appendix as agreed with your instructor or editor. MLA is the opposite here — it wants a description of the prompt in the title position.

A share link does more than freeze a transcript: a reader who opens it can carry the conversation on with the AI. That sounds alarming for a citation and isn't. In the tools APA tested, a continuation did not appear to be saved for anyone but the person who typed it — every other reader opening your link still lands on the original exchange, which is the thing your reference points at. And never cite AI as the source of a fact: verify the claim against a real source and cite that.

• Check the policy first — many courses require disclosure or ban AI for graded writing. APA has a whole post on whether AI is "allowed" in APA Style.
• Don't cite AI for facts — models hallucinate; verify and cite the primary source instead.
• Generate the share link, and rename the chat first — the link is the retrievable thing, and the chat's title is what lands in your reference.
• Keep the exchange — save the prompt and response (appendix or footnote) so your use is transparent.
• It's a tool, not an author of record — you remain responsible for everything in your paper.