A paper that quotes ChatGPT, Claude, or Gemini needs a citation, the same way a paper quoting a book or an article does. What trips students up is not whether to cite it. It is the mechanics: which format to use, what counts as the author, and what to do with the fact that an AI chat cannot be looked up by anyone else the way a published article can. This is purely a how-to on formatting. APA, MLA, and Chicago each answer that mechanical question differently, and each one has been updated recently enough that older advice floating around online is often already out of date.
None of the three style guides treat this as a footnote issue confined to computer science papers. A student citing a chatbot's phrasing in a literature essay, a psychology paper's discussion section, or a history assignment's research notes runs into the identical formatting question, and the answer does not change based on subject. What changes is only the format itself, and getting that format wrong is one of the more avoidable ways to lose easy points on an otherwise solid paper.
Why the Format Keeps Changing
Part of what makes this confusing is that the guidance itself has moved more than once since ChatGPT reached a mainstream audience. Early advice treated a chat as something close to unpublishable software, cited by version number and a generic homepage link, because there was no way to share a specific conversation with a reader. As AI tools added shareable conversation links, the style guides updated their recommendations to point at that specific link instead of a generic one. APA's most recent revision, published by its Style team in September 2025, is a direct result of that shift: once a chat can be shared the way a webpage can, the reference entry can look far more like an ordinary retrievable source than the earlier, more improvised formats did. MLA's own guidance page carries the words "updated" and "revised" in its title for the same reason. A format memorized from a 2023 class handout is worth double-checking against the current guidance before it goes into a paper due this semester.
The Common Mistake: Treating AI Output Like a Person You Talked To
The most frequent error is reaching for a "personal communication" citation. In APA Style, a personal communication is the standard category for anything a reader cannot retrieve for themselves: a phone call, a private email, a text message. Since an AI chat session used to feel similarly unrecoverable, a lot of students default to writing something like "(personal communication, March 2024)" for a quoted line of ChatGPT text, the same way they would cite a conversation with a professor. APA Style has said directly that this is the wrong move. Quoting AI-generated text is closer to citing the output of an algorithm than to quoting a person, so the company behind the tool, OpenAI for ChatGPT, Anthropic for Claude, Google for Gemini, gets credited with an actual reference list entry and a matching in-text citation, not treated as an unretrievable personal exchange.
The second common mistake is simpler and just as damaging: leaving out the prompt entirely. A different prompt produces different text, even from the same model on the same day, which means the exact wording a student typed in is part of what makes the citation meaningful. All three major style guides build a place for the prompt into their format, just in different spots. APA asks for it in the surrounding text or an appendix. MLA turns the prompt description into the actual title of the source. Chicago has a student quote the prompt directly inside the footnote itself. Skipping that step is the difference between a citation a reader can make sense of and one that just names a company with no way to tell what was actually asked.
APA Style: Reference List Entry and In-Text Citation
APA's current guidance, updated by the APA Style team in September 2025, treats an AI chat the way it treats any other authored, dated, titled source: the company is the author, the date is when the chat happened, the chat gets its own title, and the chat's own shareable link replaces the tool's generic homepage URL that older advice used to point to. A reference entry built that way looks like this: OpenAI. (2025, October 2). Themes in Frankenstein [Generative AI chat]. ChatGPT. https://chatgpt.com/s/t_68dec13c9d408191b545f5ddbb3f7eec. The elements in order are the company as author, the full date in parentheses, an italicized title describing the chat followed by a bracketed descriptor identifying it as a generative AI chat, the name of the tool, and the link to that specific conversation. The in-text citation follows APA's ordinary author-date pattern: (OpenAI, 2025).
This replaced an earlier, simpler format from APA's original 2023 guidance, which cited the tool by version number, for example "ChatGPT (Mar 14 version) [Large language model]," paired with the tool's general web address rather than a link to the specific conversation. A student working from an older style guide, an older class handout, or an AI citation tool that has not been updated may still encounter that version. Worth checking which one an instructor or publication actually expects before submitting, since both have circulated widely enough to still show up in searches.
There is one more detail APA's guidance addresses directly: because a chat produces a unique response even from an identical prompt, a reader cannot independently verify a quoted line the way they could look up a page number in a printed book. APA's own guidance handles that by recommending the full text of a long AI response go into an appendix or supplemental material, with the in-text citation pointing there, in a form such as "(OpenAI, 2024; see Appendix A for the full transcript)." That step matters most when a paper quotes more than a line or two, since a reader genuinely has no other way to see what the model actually said.
MLA Style: Works-Cited Entry Format
MLA takes a different structural approach, built around its standard template of core elements rather than a fixed reference-list formula. The guidance is explicit that the AI tool itself should not be treated as an author. Instead, the prompt becomes the Title of Source, the tool's name becomes the Title of Container, the specific model becomes the Version, the company becomes the Publisher, and the date the content was generated fills the Date slot, with the Location slot taking either a stable, shareable link to the conversation or the tool's general URL if no shareable link exists.
A works-cited entry built from those elements looks like this, matching the example format MLA's own style center provides: "Describe the theme of nature in Jane Austen's Mansfield Park" prompt. ChatGPT, model GPT-4o, OpenAI, 23 Sept. 2024, chatgpt.com/share/66f1b0a0-d704-8000-be9a-85f53c850607. The in-text citation is parenthetical and keyed to that same prompt description rather than to an author's last name, since there is no author in the traditional sense, something like ("Describe the theme").
MLA's guidance also covers a related situation worth knowing about separately: if an AI tool's response cites its own sources, the recommendation is to track down and cite those original sources directly rather than citing the AI's summary of them, since the AI-generated summary is a step removed from whatever it was actually drawing on.
Chicago Style: A Footnote, Not a Bibliography Entry
Chicago handles the retrievability problem differently than APA or MLA do. Because reading a specific ChatGPT conversation requires the original account's login credentials, Chicago's guidance treats an AI chat the way it treats a personal communication for retrievability purposes: it never appears in the bibliography or reference list at all. It only shows up in a numbered footnote or endnote, attached to the sentence where the AI-generated text is used.
Chicago's own guidance gives two template shapes, depending on whether the prompt already appears in the surrounding text or needs to be spelled out inside the note itself: Text generated by ChatGPT, OpenAI, [date], [URL]. Or: ChatGPT, response to "[the exact prompt used]," OpenAI, [date]. The first form fits when a student has already described what they asked in their own sentence. The second folds the prompt directly into the note, which Chicago's guidance recommends specifically for cases where the prompt has not been mentioned anywhere else in the text. Either way, the citation stays confined to a note. It never migrates into the bibliography the way a book or journal article would.
That is worth pausing on, because it puts Chicago on the opposite side of the exact question APA had to answer. APA explicitly rejected calling an AI chat a personal communication, on the grounds that no actual person is communicating. Chicago keeps the comparison to personal communication anyway, but for a narrower and more literal reason: not because a person is involved, but because the specific conversation sits behind a login only the original user can access, which is the same practical barrier that keeps a private email or phone call out of a Chicago bibliography. Both style guides are reasoning from the same starting fact, that an AI chat is not independently retrievable the way a published source is. They just land in different places, because APA has since built a workaround using shareable links while Chicago's current guidance has not adopted that same workaround.
Citing Claude and Gemini, Not Just ChatGPT
Every template above swaps cleanly to a different AI tool, because none of the three style guides wrote their guidance around ChatGPT specifically. The company that built the tool takes the author or publisher slot: Anthropic for Claude, Google for Gemini, OpenAI for ChatGPT. The tool's own name, and its specific model where one is named, for instance a particular Claude or Gemini version, fills the title or version slot the same way "ChatGPT" and "model GPT-4o" do in the examples above. The date, the prompt, and the link-or-lack-of-one all work exactly the same regardless of which company built the model that generated the text. A student who has the APA, MLA, or Chicago template down for one AI tool already has it for all three, since the mechanical structure is about the type of source, a generative AI chat, rather than about any single company's product.
Where Everyone Agrees, and Where They Don't
Lined up side by side, the three style guides agree on more than it first appears, even though the formatting looks nothing alike on the page. All three insist the AI tool cannot stand in as a human author. All three require the prompt to be documented somewhere, even though each one places it in a different spot. Where they diverge is retrievability: APA and MLA now both lean on a shareable conversation link and treat the result as a normal, citable source, while Chicago still treats the same conversation as fundamentally unrecoverable by another reader and keeps it out of the bibliography on that basis alone.
| Citation Element | APA | MLA | Chicago |
|---|---|---|---|
| Where it appears | Reference list entry plus an in-text (Author, Year) citation | Works-Cited entry plus an in-text citation keyed to the prompt | A numbered footnote or endnote only, never in the bibliography |
| Who counts as the author | The company behind the tool: OpenAI, Anthropic, or Google | No author. The AI tool is named as the Title of Container instead | The company, cited the way a personal communication would be |
| Does the prompt have to appear | Yes, in the surrounding text or an appendix for long responses | Yes. The prompt description becomes the Title of Source itself | Yes, quoted directly inside the footnote or already in the text |
| What link to use | The chat's own shareable link and title, per the Sept 2025 update | A stable, shareable link, or the tool's general URL if none exists | Not required. The citation never appears in a retrievable bibliography |
Keeping the Details Straight Across a Semester of Chats
The formatting rules above only work if a student still has the underlying details by the time a paper is due: which tool generated a given line, what the exact prompt was, and the date the chat happened. Those details are easy to lose across a semester spent bouncing between ChatGPT for one assignment, Claude for another, and Gemini for a quick check on a third, especially once a browser history gets cleared or a chat gets buried under dozens of newer ones. MemX keeps a running, private record of which AI tool produced what and when, across ChatGPT, Claude, and Gemini, so the prompt, the date, and the source tool are still there when it comes time to build the citation, private by architecture rather than dependent on a single chat history that may or may not still be open.
01What is the biggest mistake students make when citing AI?
Citing an AI chat as a "personal communication," the way a phone call or private email would be cited. APA Style specifically advises against this, since quoting AI text is closer to citing an algorithm's output than a person's words, and recommends a full reference list entry credited to the company instead.
02What is the current APA format for citing ChatGPT?
As of the APA Style team's September 2025 update, a reference entry names the company as author, the date, an italicized chat title with a "[Generative AI chat]" descriptor, the tool name, and a link to that specific shareable chat, cited in text as (Company, Year).
03How does MLA cite an AI-generated response?
MLA treats the prompt as the Title of Source, the AI tool as the Title of Container, the specific model as the Version, the company as the Publisher, and either a shareable or general URL as the Location, with the AI tool itself never listed as an author.
04Does Chicago Style require a bibliography entry for ChatGPT?
No. Chicago treats an AI chat as unretrievable by other readers, similar to a personal communication, so it appears only in a numbered footnote or endnote and is left out of the bibliography entirely.
05Do I need to include my exact prompt when citing AI?
Yes, in some form, in every major style. APA wants it in the text or an appendix, MLA turns it into the source's title, and Chicago has it quoted directly inside the footnote, since a different prompt would have produced different text.
