AI citation workflow · Updated 24 September 2026
How to Cite ChatGPT and AI in APA, MLA, and Chicago
The three styles do not agree with each other, and APA alone gives you two different reference formats to choose between. Here is how to work out which artifact your paper actually needs — an in-text acknowledgment, a reference entry, an appendix transcript, or some combination.
Before you start
You need three things open before the style question means anything: your course AI policy, your assignment instructions, and the chat itself (or your notes about it — the date, the model name, and the share link if the tool offers one). If the chat is gone and you cannot reconstruct what happened, say so in your disclosure rather than inventing details.
Quick answer
- Policy first. If the use was not allowed, a citation does not fix it. Ask your instructor before submitting.
- Work out which artifact you owe, not just which format. Depending on your course, a use may need only an in-text acknowledgment, or an entry at the end, or an appendix transcript — and these are separate decisions. The decision aid below sorts this.
- Copy the structure from the current official page, not from a chatbot's memory. In APA there is a second question first: are you citing the tool or one specific chat? Those are two different reference formats, both current.
- Cite sources you opened yourself. If the AI pointed you to an article, open it, read the relevant part, and cite it directly — MLA says the same thing, because AI tools hallucinate and mis-summarise sources.
- Keep a log as you go. Reconstructing AI use the night before a deadline is where fabricated details creep in.
APA now has two formats, and picking the wrong one is the common error
APA's early advice, in How to cite ChatGPT, treated the tool like software: the company as author, the tool and model, and a [Large language model] description. In September 2025 APA published a three-part series, and Part 1 — Reference formats added a second format pitched at a different target: one particular conversation, with the date of that chat, the chat's own title, a [Generative AI chat] description, and the chat's URL.
Both formats are current. This is the detail that trips students up, because it looks like a version change and is not one. University guides reproducing APA's current templates — UBC Library, Oregon State University Libraries, Red Deer Polytechnic — show the tool-level entry and the chat-level entry side by side, UBC with a tool-level example carrying a 2025 date and a current model name. So a correctly built APA reference list can contain both kinds of entry, because they are citing two different things.
| Citing the tool | Citing one chat | |
|---|---|---|
| Elements | Company, year, tool and model, [Large language model], the tool's own URL. | Company, the full date of the chat, the chat's title, [Generative AI chat], tool and model, the chat's URL. |
| Use it when | The tool did work that is not the source of any idea in your paper — editing or organising your own writing, translating, generating code, tables or figures, or as part of your methodology — or when a specific chat cannot appropriately be cited. | A specific conversation generated ideas, content, or information that appears in your work. |
| Needs a share link? | No. The URL is the tool's public address. | Yes — and it has to actually resolve for a reader. |
Red Deer Polytechnic's summary of the rule is the clearest one-liner: cite the tool generally when it would be “unhelpful, unethical, or otherwise inappropriate to cite a specific chat”, or when you want to point at the tool's existence rather than at information from it. Our reading of this: that clause is doing real work for students. A chat containing your own draft, your classmates' text, or anything you should not publish is a chat you should not be linking to, and the tool-level reference is the route APA leaves open for exactly that case.
Either way, keep your own records. Purdue University Libraries' AI guide, updated on 23 September 2026, shows the chat-level template with its shareable URL and advises keeping copies of important prompts and outputs in case you are asked to put a transcript in an appendix. That is the same material the log at the end of this page collects.
One genuine lag is worth knowing about: some library guides still show only the older tool-level entry as the way to cite a chat — UWM's guide still presents the 2023 software-style entry for ChatGPT — so if a handout gives you one format with no mention of the other, check it against APA's current page rather than assuming it is complete.
The disagreement that decides your reference list
This is the question students actually get stuck on, and the three styles answer it three different ways for the same AI use.
| APA | MLA | Chicago | |
|---|---|---|---|
| End-of-paper entry? | Yes — a reference-list entry, at chat level for a conversation you drew content from, at tool level otherwise. | Yes — a works-cited entry. | Not shown in the public guidance. A numbered note carries the credit; a university library guide describes the exchange as handled like personal communication, noted but not listed in a bibliography. |
| Who is the author? | The company that made the tool. | No author. MLA explicitly does not recommend treating the AI tool as the author. | Credit is phrased in the note or your own sentence rather than an author slot. |
| What sits in the title slot? | The chat's title in a chat-level entry, or the tool and model in a tool-level entry — each with its own bracketed description. | Your prompt in quotation marks, or a description of what was generated. | A description such as text generated by the named tool, with the prompt optionally quoted in the note. |
| Where does the model name go? | With the tool in the source element. | The version slot — name the model as specifically as you can. | No model slot. The published note example names the tool and the company only. |
| Lightest acceptable option | Disclose the use, then a tool-level reference where no idea or wording came from a particular chat. | Works-cited entry with matching in-text citation. | For most writing, simply naming the tool in your own sentence. |
| Official page | APA Style, Part 1 | MLA Style Center | CMOS Q&A |
The practical consequence: you cannot build one citation artifact and reuse it across styles. A Chicago paper and an APA paper describing the identical chat look different in kind, not just in punctuation.
Step 1: Find your use in the decision aid
Citation is not permission. If your instructor says not to use AI for a task, adding a citation does not make the use acceptable — start with the course rule, then the assignment instructions, then the style guide. Once the use is allowed, the question is which artifact you owe. Find your use in the left column.
| What the AI actually did | In-text acknowledgment | Entry at the end | Keep a transcript |
|---|---|---|---|
| Suggested search terms or a topic angle; none of its wording or ideas survived | Only if your class requires an AI-use statement | No | Optional |
| Named sources you then found, opened, and read yourself | Only if your class requires it — cite the sources, not the chat | No entry for the chat; full entries for the sources | Useful if you are ever asked how you found them |
| Generated wording, an outline, or an idea that appears in what you submit | Yes | APA: chat-level reference. MLA: works-cited entry. Chicago: a note. | Yes — save the share link now |
| Produced a table, figure, image, or code, or ran part of your method | Yes, in the caption or the method description | APA: tool-level reference — this is the case APA's general-tool format is for. MLA: works-cited entry. Chicago: a note. | Yes, plus the prompt that produced it |
| Edited, organised, or translated your own writing; gave feedback you acted on | Usually yes, phrased as editing or feedback rather than generation | APA: tool-level reference, with the use disclosed. MLA and Chicago: judgment call; ask your instructor. | Yes, in case you are asked what changed |
Two things that trip people up on this table. Grammar and spell-check sit awkwardly in the last row and courses genuinely differ on whether they count at all, so this is an ask-your-instructor case rather than a guess. And a share link is only worth including if it actually resolves for someone else — open it in a private window before you paste it into a reference. If it does not resolve, or if the chat contains material you should not be publishing, that is an argument for the tool-level reference rather than for a broken link.
Step 2: Build the log first, the citation second
Almost every fabricated AI citation starts the same way: the student reconstructs a chat from memory and fills the gaps with plausible-looking model names and dates. Run this prompt in the chat you actually used, while it is still open.
You are helping me document this conversation for an academic AI-use log. Do not write a citation and do not guess anything.
From this conversation only, report:
1. The title shown for this chat, if it has one.
2. The model name or version you are running as, exactly as you can state it. If you cannot state it with certainty, write "unknown - check the interface".
3. The date and time of my first message in this chat, if you can see it; otherwise write "unknown".
4. A one-sentence description of what you generated for me.
5. Every place in this chat where you named an outside source: list the sources and mark each "quoted from a document I gave you" or "produced from your own training".
Then list which of these fields I must verify myself in the interface rather than take from you.The end of that prompt is the part that matters. Item 5 gives you the shortlist of sources you still have to open yourself, and the closing instruction keeps you honest about fields a model cannot reliably self-report — its own version string and your chat's timestamps are exactly the details chatbots will state confidently and get wrong. Read those off the interface, not off the reply.
Step 3: Stress-test the draft entry against the official page
Do not ask a chatbot to produce the citation. Ask it to attack one you drafted from the official template — a much easier task to check, because every criticism it raises is something you can confirm or dismiss in thirty seconds against the style page.
I drafted the reference below by copying the structure from my style guide's current official page. Do not rewrite it and do not produce a corrected version.
Instead:
1. Break my entry into its elements and label each one (author / date / title / description / container / version / publisher / location).
2. For each element, tell me what I would need to check on the official page to confirm it is right.
3. Flag any element that looks like something I could not have known without checking the tool interface - those are the ones I may have invented.
4. Tell me which elements are missing entirely.
5. Tell me whether my in-text citation and my end-of-paper entry start with the same words. If they do not, say so.
6. I have told you below what I am citing. Tell me whether the bracketed description and the date in my entry are consistent with that, or whether they describe a different kind of thing.
Style: [APA 7 / MLA 9 / Chicago notes-bibliography]
What I am citing: [this specific chat / the tool in general]
My draft entry: [paste]
My in-text citation: [paste]How to check what you get back
Take the labelled element list the prompt produces and lay it next to the official page. Four checks settle it: every element the style requires is present; in APA, the bracketed description matches what you decided you were citing; the first words of your in-text citation match the first words of your end entry (this is what lets a reader find it); and nothing in the entry is a detail you could not have seen with your own eyes. If the model produced a corrected citation despite being told not to, discard the correction — that is the output you cannot verify.
The failure that actually gets students in trouble
It is not a mis-placed comma. It is the confident, well-formatted entry for a source the student never opened. MLA's guidance says this plainly: click through to any sources an AI response points to and cite those directly, because AI tools can invent sources or summarise them incorrectly. A reference list is a claim about what you read. A tidy citation for an unread paper is a false claim even when every element is formatted correctly.
Worked example: one AI use, documented three ways
Illustrative example, not a real chat. Say you asked a chatbot to generate a comparison paragraph about two theories, edited roughly half of it, and kept it in your essay. The chat is titled “Two theories compared”, it ran on a model whose name is visible in the interface, and the tool gave you a share link. Here is what each style asks you to produce — note that it is not one artifact in three costumes.
| Style | What goes in the body | What goes at the end |
|---|---|---|
| APA | A sentence naming what you used the tool for, with a parenthetical pointing to the company and the chat's date. | A chat-level reference-list entry — because this chat generated content that survived into the essay: company as author, the chat's date, the chat title with a bracketed generative-AI-chat description, then the tool name and the chat URL. Had the tool only tidied your own sentences, the tool-level entry would be the right one instead. |
| MLA | A parenthetical using the opening words of the works-cited entry — that is, the start of your quoted prompt or description. | A works-cited entry with no author: your prompt or a description in the title slot, the tool as container, the model as version, the company as publisher, the date generated, and the share link. |
| Chicago | Often enough on its own: a sentence saying the passage was generated by the named tool and edited by you. | A numbered note naming the generated text, the tool, the company, the date and the URL, and stating that you edited it. No bibliography entry shown in the public guidance. |
Because you edited the generated paragraph substantially, Chicago asks you to say so — a phrase such as “edited for style and content” in the note or in your own sentence. That disclosure of editing has no direct equivalent in the other two styles, which is another reason a Chicago paper and an APA paper describing this identical chat do not look alike.
Step 4: Cite original sources, not AI summaries
If AI points you to a journal article, book, dataset, website, or report, open the original source yourself. Read the relevant section. Then cite the original source if it actually supports your claim.
Do not cite an AI answer as a shortcut for sources you never opened. AI can summarize, misread, omit context, or invent details. Your bibliography should show the sources you personally used.
Copy-ready AI-use log
Keep one of these per chat, in the same document as your notes. Every field is something you read off the screen yourself — if a field is blank at submission time, leave it blank and say so rather than filling it in from memory.
Assignment:
Instructor AI policy says (allowed / allowed with disclosure / not allowed):
Tool used:
Model or version, copied from the interface (not from the chatbot):
Chat title shown in the sidebar:
Date of chat:
Share link: Opens in a private browser window? yes / no / no link offered
Prompt purpose:
Full prompt saved? yes / no
Output used in submitted work? none / ideas only / wording / outline / image / code / data / other
How much of it survived my editing:
Sources the AI named: Which ones I opened and read myself:
What I am citing (this specific chat / the tool in general):
Artifact I owe (in-text only / entry at the end / note / appendix transcript):
Where I disclosed or cited it:
Still to ask my instructor:
How to Become a Straight-A Student
A strategic study blueprint designed specifically for college students. Demystifies top performance and shows how to prepare for exams efficiently without cramming.
AI Study Pilot receives a small commission from qualifying Amazon purchases at no extra cost to you.Common mistakes
Mistake 1: Thinking citation makes prohibited use acceptable
If the course bans a use, citation does not fix it. Ask your instructor before submitting.
Mistake 2: Letting AI create citations from memory
A chatbot's citation examples reflect whatever it absorbed during training, and it has no way to know which of APA's two reference targets your use calls for — that decision depends on facts about your own work that are not in the chat. Use the current official style page for the structure, decide the target yourself, and use the model to interrogate your draft rather than to write it.
Mistake 3: Citing an AI summary instead of the source
If your claim depends on a source, cite the source you read, not the AI assistant that mentioned it.
Mistake 4: Forgetting non-text AI output
Images, tables, charts, code, and generated data may also need citation, captioning, or disclosure.
FAQ
Do I always need to cite ChatGPT?
Not always. If AI only helped you plan your study schedule, your class may not need a formal citation. If AI-generated content appears in your submitted work, you should expect to cite or disclose it.
Should ChatGPT be listed as an author?
MLA says not to treat the AI tool as the author. Other styles handle this differently. Follow the required style and your instructor's policy.
Can I cite sources that AI recommended?
Only after opening and reading those sources yourself. If you cannot verify a source, remove it.
My chat has no share link. What do I do?
In APA this is a case the guidance anticipates: where citing a specific chat is unhelpful or inappropriate, the general tool-level reference is the intended alternative, so use that and disclose the use in your text. Either way, record everything you can see — tool, model, date, chat title. Do not paste a private URL and hope it resolves, and do not invent one. A reader who cannot open your link is in a worse position than a reader told plainly that the exchange is not retrievable.
Tool-level or chat-level in APA — how do I decide?
Ask what you are actually pointing the reader at. If a particular conversation is the source of an idea, a phrase, or a piece of information in your paper, cite that chat. If the tool did work that is not the source of anything — editing, organising, translating, generating a figure or code, part of your method — or if the chat cannot appropriately be linked, cite the tool. Both entries can appear in the same reference list, because they are citing different things. What you should not do is describe one kind of use with the other kind of entry.
Final recommendation
Before submitting, do a three-part check: course policy, official style guidance, and your AI-use log. If any part is unclear, ask your instructor early instead of guessing the night before a deadline. And remember what the whole exercise is for — the formatting is downstream of an honest account of what you did and what you actually read.
Disclosure: AI Study Pilot may add affiliate links later. We recommend free-first tools where possible and never promise guaranteed grades or outcomes.