How to Cite AI-Generated Content

To cite AI-generated content, identify the tool, describe or name the prompt, note the model/version and date of generation, and choose the right style rule for retrievability. Use in-text only for non-recoverable conversations; create a full reference/entry when the exact output is publicly accessible or reproducible.

What Counts as AI-Generated Content in Academic Writing

AI-generated content includes any material produced substantially by a conversational model (e.g., ChatGPT, Gemini) or generative system (image, code, summaries) in response to a user instruction. It can range from single-sentence clarifications to multi-page drafts. For citation purposes, the crucial factor is how a reader could verify or recover the exact wording you used. If your text came from a private chat that can’t be accessed again by others, that output is non-recoverable. If the content lives at a persistent URL, has a unique share link, is embedded in a repository (e.g., supplemental materials), or is precisely reproducible via a logged prompt, model, and date, then the output is recoverable.

This distinction drives everything that follows. Styles differ in how they record non-recoverable versus recoverable sources, but they all share the same goal: credit the tool, disclose your role (prompting), and give readers enough information to understand and assess the origin of the words, ideas, or structure. Another shared principle is transparency about versioning. Models are updated frequently; noting the model name and version (or month/year) plus the date you generated the response helps future readers situate your source historically.

Finally, keep in mind that AI output often summarizes other works. If a specific claim, statistic, or quotation originates in a third-party source that the AI surfaced, cite that original source directly. The AI is not a substitute for proper attribution to underlying materials; it is the means of generation. Your methods section, acknowledgments, or a brief note in the text can clarify how you used AI during drafting, separate from formal citations.

Core Principles You Can Apply Across Styles

1) State the tool clearly. Use the product name (e.g., ChatGPT or Gemini) and the developer/publisher. Treat the model as a nonhuman authoring system rather than a person.

2) Capture the prompt or task. Provide a short, descriptive prompt label (e.g., “Summary of Smith (2022) on urban resilience”) or quote the prompt if concise. When the exact wording matters, include it in the note or parentheses.

3) Record model/version and generation date. Models change. Indicate the version or release window if available (e.g., “ChatGPT, August 2025 version”), plus the date you generated the content.

4) Decide based on retrievability.

• Non-recoverable chat: cite in text or as personal communication/footnote, usually no reference-list entry.

• Recoverable output: include a full entry with all elements you would expect for a software/web source.

5) Don’t misattribute secondary sources. If the AI paraphrased a report, cite the report, not the AI.

These five principles keep your documentation consistent even as style manuals evolve.

APA 7th: In-Text and Reference Examples for AI Content

How APA treats non-recoverable AI chats. When the exact conversation cannot be accessed by your reader, APA treats the response like non-recoverable content (similar to classroom lectures or emails). You include an in-text citation only identifying the AI system and the year, and you do not add a reference entry. The citation signals that the wording originated from an AI tool at a particular time without implying public access.

In-text example (parenthetical):

(ChatGPT, 2025)

In-text example (narrative):

ChatGPT (2025) generated a draft definition that we refined for clarity.

To increase transparency, you may describe the prompt in the sentence or in brackets when succinct:

Using ChatGPT (2025) with the prompt “Draft a 150-word abstract about…”, we compared the output to our manual summary.

How APA treats recoverable AI outputs. If a reader can retrieve the same output (e.g., a published model card, a persistent share link, or an archived appendix), create a reference-list entry. Format it like software or webpage content authored by the organization behind the tool, with the model/tool name, version, and a descriptor. If your institution asks for a prompt note, add it to the entry’s title element as a bracketed description.

Reference entry template (recoverable):

Organization. (Year, Month Day). Tool or model name (version) $Largelanguagemodeloutput$. Site/Platform. URL

Reference entry example (recoverable):

OpenAI. (2025, August 15). ChatGPT (August 2025 version) $Largelanguagemodeloutputrespondingtotheprompt“Explainthecarboncyclein120words.”$. OpenAI.

In-text citations for this entry:

(OpenAI, 2025) — or — OpenAI (2025)

Quoting and paraphrasing. Short AI quotations can be integrated like other quotations with page/paragraph indicators replaced by a prompt or response identifier if your share link provides one. Otherwise, indicate that the quote is from AI output generated on [date].

Common APA pitfalls to avoid.

• Listing AI as a human author. The author is typically the organization behind the system.

• Providing a reference entry for a non-recoverable chat. For private sessions without a shareable record, use in-text only.

• Omitting the generation date or version when they matter for replicability.

MLA 9th: Entries, In-Text, and When to Use a Descriptive Title

Core MLA approach. MLA centers on containers and titles of source. For AI outputs, two decisions simplify your entry: (1) is the output recoverable? (2) what should count as the title? A practical MLA practice is to give a descriptive title that names the AI and summarizes the prompt, then treat the platform as the container.

Works Cited template (recoverable):

“Descriptive title of the AI response.” Tool name, version, Publisher, Day Month Year of generation.

Example:

“Response to the prompt ‘Compare qualitative vs quantitative reliability.’” ChatGPT, August 2025 version, OpenAI, 15 Aug. 2025.

MLA in-text citations rely on the shortened title since there is no human author. Use quotation marks and a shortened version of your Works Cited title:

In-text (first mention):

(“Response to the prompt ‘Compare…’”)

In-text (subsequent):

(“Response to the prompt”)

Non-recoverable conversations. If no public record exists, you may document the use in the text (e.g., “In an interaction with ChatGPT on 15 Aug. 2025…”) and omit the Works Cited entry, or include a minimal entry that records the tool and date without a URL if your instructor prefers a complete list of consulted sources. Clarity about the prompt and date remains valuable.

Formatting details that improve MLA clarity.

• Put the tool name in italics only when it functions as the container title; otherwise, keep it roman.

• Use Title Case for the descriptive response title in quotation marks.

• Note version (month/year or named version) after the tool if available.

Common MLA pitfalls.

• Treating the AI as a person with a surname in in-text citations. Prefer a shortened title keyed to Works Cited.

• Leaving out the prompt context, which helps instructors understand what the tool actually did for your work.

Chicago: Notes-Bibliography and Author-Date Options for AI

Chicago provides two documentation systems. Both can accommodate AI with small adjustments centered on transparency and retrievability.

Notes-Bibliography (NB) system. For a non-recoverable chat, use a note to describe the interaction and include the tool, model/version, prompt, and generation date. You typically do not include a bibliography entry in such cases.

Footnote example (non-recoverable):

1. ChatGPT (OpenAI), response to the prompt “Outline the key steps in a cold-brew extraction,” generated August 15, 2025.

If the output is recoverable, create a bibliography entry similar to web content or software, with a descriptive title and the tool as author or corporate author.

Bibliography example (recoverable):
OpenAI. ChatGPT (August 2025 version): Response to “Outline the key steps in a cold-brew extraction.” Generated August 15, 2025.

Corresponding footnote (first):

1. OpenAI, ChatGPT (August 2025 version): Response to “Outline the key steps in a cold-brew extraction,” generated August 15, 2025.

Author-Date system. Treat the corporate author as the author and include a year. For non-recoverable chats, many writers prefer a textual mention rather than a reference-list entry (e.g., “Using ChatGPT (2025), we generated…”). For recoverable outputs, add a reference list entry:

Reference list example (recoverable):
OpenAI. 2025. ChatGPT (August 2025 version): Response to “Explain genetic drift in one paragraph.” Generated August 15, 2025.

In-text author-date:
(OpenAI 2025)

Quoting, paraphrasing, and notes. For NB, long AI quotations may appear in block format like other prose, with a note that attributes the output to the tool and includes the generation details. For Author-Date, reserve block quotation for longer excerpts and provide a parenthetical citation keyed to the reference. In both systems, consider adding an explanatory note the first time you use AI in the paper: what you asked, how you checked the accuracy, and how you revised the wording.

Comparison Table: APA vs MLA vs Chicago for AI Citations

Style-Neutral Writing Tips for Using and Citing AI

Be explicit about your human role. Readers and instructors care about which sentences are your analysis and which originated from an AI tool. Signpost this in your prose the first time you draw on AI. A compact disclosure might say that the first draft of a definition came from an AI response generated on a particular date and that you revised and verified it.

Prefer primary sources for facts. If the AI provides a statistic or cites a study, track down the original and cite that work, not the AI response. Your AI citation indicates the medium of generation, not the origin of third-party information.

Archive what you can. If your institution allows it, save the exact prompt, model name, version/date, and the text you used. If you can put this in an appendix or a project repository, your output becomes recoverable, and your citation can be more standard and useful to future readers.

Name the model precisely. “ChatGPT” and “Gemini” encompass multiple versions. When possible, specify the release (“August 2025 version,” “Gemini Advanced, mid-2025 release”) rather than a generic label. This helps others interpret differences in tone, coverage, or accuracy between iterations.

Avoid over-citation. Cite AI when its wording or structure materially influenced your text or when quoting/paraphrasing its output. If you used AI only to brainstorm synonyms or outline ideas that you then independently developed, a brief methods disclosure may suffice without repeated citations.

Ready-to-Use Citation Patterns (Customize the Words in Italics)

APA — non-recoverable (in-text only):
(ChatGPT, 2025) or ChatGPT (2025) generated an initial summary we later revised.

APA — recoverable (reference + in-text):
Reference: OpenAI. (2025, August 15). ChatGPT (August 2025 version) $Largelanguagemodeloutputrespondingtotheprompt“\ast yourprompthere\ast .”$. OpenAI.
In-text: (OpenAI, 2025)

MLA — recoverable (Works Cited + in-text):
Works Cited: “Response to the prompt ‘your prompt here.’” ChatGPT, August 2025 version, OpenAI, 15 Aug. 2025.
In-text: (“Response to the prompt”)

Chicago NB — non-recoverable (note only):

1. ChatGPT (OpenAI), response to the prompt “your prompt here,” generated August 15, 2025.

Chicago Author-Date — recoverable:
Reference list: OpenAI. 2025. ChatGPT (August 2025 version): Response to “your prompt here.” Generated August 15, 2025.
In-text: (OpenAI 2025)

These patterns are deliberately link-free to match academic environments that restrict URLs or require internal archiving. Replace the italicized parts with your own prompt and dates; adjust capitalization to your style guide.