Students and researchers increasingly ask how to cite AI-generated content in APA, MLA, and Chicago style. ChatGPT, Claude, Gemini, and other tools now appear in brainstorming notes, draft paragraphs, data summaries, and even image captions. Style guides have responded, but the rules differ by citation system and by how you used the tool.
This guide explains current conventions for citing generative AI in academic work, when disclosure is required versus optional, and how university policies intersect with style manual rules. It is written for human editors who will refine examples for their audience, but every section reflects guidance published by APA, MLA, and Chicago as of 2025–2026.
Before you submit, run your draft through an AI detector only if your institution requires it—not because detection replaces citation, but because some syllabi ask for both disclosure and verification. Pair that with a grammar checker after you rewrite AI-assisted sections in your own voice. The full Proofly platform supports that pre-submission workflow without replacing your responsibility to cite honestly.
Why Citing AI Matters Now
Generative AI produces text that reads like human prose. Without citation, readers cannot tell whether an argument, definition, or statistic came from you, from a published scholar, or from a model trained on unknown sources. That ambiguity undermines reproducibility—the core expectation in academic and professional writing.
Citation also protects you. If an AI tool hallucinates a case study, journal title, or historical date, your reference list should show that the claim originated with the tool, not with a peer-reviewed source you never verified. Instructors increasingly treat undisclosed AI use as an integrity issue separate from plagiarism, because it misrepresents authorship and process.
Style organizations updated guidance quickly after ChatGPT's public release. APA published explicit instructions in 2023 and refined them in subsequent FAQ updates. MLA added a template entry for AI in its ninth edition handbook supplements. Chicago addressed generative tools in its online Q&A. These are not identical: APA treats the tool as a non-recoverable source; MLA emphasizes the prompt and container; Chicago distinguishes chat transcripts from published AI outputs.
Employers and graduate programs are watching, too. A methods section that says "I used GPT-4 to code open responses" signals transparency. A paper that reads like polished AI with no attribution signals the opposite. Citing AI is less about punishing tool use and more about documenting how knowledge was produced.
Finally, citation trains good habits for a hybrid future. You will cite datasets, software, lab instruments, and collaborative platforms the same way you cite AI: name the tool, describe your use, and separate your analysis from machine output. Starting now prevents retroactive cleanup when policies tighten further.
Accessibility and equity add another layer. Some students use AI for drafting support permitted by disability accommodations; others use it for translation or structure help. Citation and disclosure make those legitimate uses visible so instructors can evaluate learning outcomes fairly rather than guessing at process.
APA 7th Edition Format for AI
The American Psychological Association instructs writers to treat generative AI as an author when quoting or paraphrasing its output, with important caveats. Because chat content is often not retrievable by other readers, APA recommends describing the exchange in the text or in a note rather than implying a traditional recoverable URL.
In-text citation pattern: When you quote or paraphrase text produced by ChatGPT or similar tools, name the AI model and version in the narrative, give the date of the conversation, and describe the prompt briefly if it clarifies your use.
Example narrative citation: "When prompted to define working memory, ChatGPT (OpenAI, 2024) described it as a limited-capacity system for holding information temporarily (July 15 conversation)."
Reference list entry: APA's official FAQ provides a template for the reference list. The author field is the name of the AI model. The year is the year of the version you used. The title describes what the model is. The source is the company, and you note that the output is non-recoverable.
Example reference list entry:
OpenAI. (2024). ChatGPT (Mar 14 version) [Large language model]. https://chat.openai.com
If you only used AI for editing—grammar, clarity, tone—and did not quote or paraphrase its language, APA generally does not require a reference entry. Instead, disclose in a methods or acknowledgments paragraph: "Grammar and clarity were improved with assistance from Grammarly; no generative text was inserted."
For AI-generated images, APA treats them similarly: credit the model, describe the prompt, and note non-retrievability. If you publish the image, include it as a figure with a note under the figure caption explaining AI generation.
APA also distinguishes between AI as a search assistant and AI as an author. If you asked a tool to locate sources and it returned fake citations, you still must verify every reference independently. Citing the AI does not transfer responsibility for accuracy to the model—it documents where the error chain began.
When reproducing a long AI response in an appendix for transparency, label the appendix clearly and cross-reference it in the body. Reviewers and instructors increasingly appreciate appendices that show prompts and raw outputs when AI shaped your analysis.
MLA 9th Edition Format
MLA style centers the "container" concept. For generative AI, the MLA Handbook supplemental guidance treats the AI tool as the title of source, the prompt as a descriptor, and the software version or access date as part of the entry.
Works Cited entry pattern: "Title of Source" (often the prompt or a short description), Title of Container (the AI tool name), version, publisher, date of interaction, general URL if applicable.
Example Works Cited entry:
"Describe the theme of isolation in Frankenstein" prompt. ChatGPT, 13 Mar. 2024 version, OpenAI, 15 July 2024, chat.openai.com.
In-text citation: MLA prefers concise parenthetical references tied to your prose. Because page numbers do not exist for chats, use a short description or refer to the tool in your sentence: "According to a response generated by ChatGPT using the prompt listed in the Works Cited entry, Shelley links isolation to..."
MLA emphasizes that you are responsible for verifying facts AI provides. The Works Cited entry documents the tool; it does not certify truth. If you paraphrase AI output, cite it. If you use AI only to suggest synonyms after you wrote the sentence yourself, many instructors accept a disclosure statement instead of a Works Cited entry—check course policy.
For humanities papers that analyze AI text itself—as in digital humanities or rhetoric courses—MLA allows quoting the AI in block format like any primary source, with the prompt preserved so readers understand context.
When AI contributes a translation, MLA treats the translation as AI output requiring citation unless your instructor accepts a blanket disclosure for language support. Document the source language text separately from the machine translation you relied on.
Chicago and Turabian Style
The Chicago Manual of Style Online addresses generative AI in its Q&A section. Chicago generally recommends citing AI in a note or bibliography entry when you quote or substantially paraphrase model output, similar to personal communications or non-archived electronic material.
Footnote/endnote pattern: Include the prompt, tool name, version, date of access, and a statement that the content may not be reproducible.
Example note:
1. ChatGPT, response to "Summarize Marshall's theory of citizenship," OpenAI, March 14, 2024 version, July 15, 2024.
Bibliography entry: Chicago lists the model as author when the AI produced the cited language.
ChatGPT. OpenAI, March 14, 2024 version. Response to "Summarize Marshall's theory of citizenship." July 15, 2024.
Turabian, used widely in undergraduate theses, follows Chicago's note-bibliography system. Student papers should ask advisors whether AI citations belong in notes only or also in the bibliography. Many departments prefer a single "AI Use Statement" on the title page plus notes for direct quotations from the tool.
Chicago draws a useful line between using AI and citing AI. Using AI to brainstorm, outline, or check spelling may require disclosure per institutional policy even when Chicago does not mandate a bibliographic entry. Citing is required when AI language enters your argument.
For author-date Chicago variant common in sciences, treat the model as author with year and a "personal communication" style note if the exchange is not archived. Consistency within your document matters more than forcing AI into a template designed for journal articles.
When to Cite AI vs When Not To
Not every interaction with AI requires a formal citation. The dividing line is whether AI language or AI-generated ideas appear in your submitted work in a way a reader would want to trace.
Cite or document when:
- You quote AI output verbatim or near-verbatim.
- You paraphrase AI explanations of concepts, events, or data.
- You use AI-generated images, tables, code, or audio in the project.
- Your instructor or journal requires an AI use statement listing tools and prompts.
- AI produced a literature summary you relied on before reading primary sources—you should cite the primary sources after verification, and may also disclose AI's role in discovery.
Disclosure without formal citation may suffice when:
- AI checked grammar or spelling without inserting new sentences.
- You used AI to generate interview questions you then asked real participants—disclose the assistance in methods, cite participants, not the model.
- AI helped format references from metadata you supplied, and no AI prose appears in the body.
Neither citation nor disclosure is a substitute for prohibited use. If your syllabus bans generative drafting, citing ChatGPT does not make a banned workflow acceptable. Policy and citation are separate layers.
When in doubt, over-disclose in a short paragraph at the end of your introduction or in an acknowledgments section. Editors and instructors prefer transparency they can trim over omissions they must investigate.
Group projects need shared agreement: one member's undisclosed AI use affects everyone if the submission is unified. Document tool use in a collaboration log the way you document who drafted each section.
Disclosure Statements That Satisfy Syllabi
Many universities now require a standalone AI disclosure statement even when formal citation is not needed. A good statement names tools, describes tasks, and clarifies what remained the student's own intellectual work.
Minimal template: "I used [Tool Name, version] on [dates] to [specific tasks: outline, define terms, check grammar]. All analysis, conclusions, and final wording are my own unless quoted and cited."
Expanded template for research papers: "Generative AI assisted with preliminary keyword brainstorming and sentence-level clarity edits. No statistics, quotations, or literature claims were taken from AI without independent verification against peer-reviewed sources listed in the reference list. Prompts and raw outputs are available on request."
Place the statement where your instructor expects it—title page footnote, methods section, or honor code appendix. Match the tone of your discipline: STEM papers often use a "Computational Tools" subsection; humanities papers may use an "Acknowledgments" paragraph.
Avoid vague lines like "I used AI a little." Specificity protects you during review. If you used Copilot inside Word, name it. If you used a custom GPT, describe its purpose. If you regenerated answers until one sounded right, say so—revision history is increasingly admissible in integrity hearings.
Disclosure pairs naturally with quality control. After drafting, use a grammar checker on your final human-written version, not on raw AI paste. That sequence shows you treated AI as support, not author.
University Policies and How They Interact With Style Guides
Style manuals tell you how to format attribution; university policies tell you whether attribution is enough. The stricter rule always wins. A perfectly formatted APA reference to ChatGPT does not legitimize AI draft generation on an assignment that requires fully independent writing.
Common policy categories in 2026 include:
- Permitted with disclosure: brainstorming, outlining, grammar support, coding help in specified courses.
- Permitted with instructor approval: translation, paraphrasing support, literature mapping.
- Prohibited: submitting AI text as original analysis, fabricating sources with AI, using AI on proctored exams.
Honor codes increasingly define "unauthorized assistance" to include undisclosed generative tools, parallel to plagiarism definitions. Sanctions range from revision opportunities to course failure, depending on severity and intent.
Appeals often hinge on process documentation. Students who keep prompts, drafts, and revision timestamps fare better than those who claim "I only used AI for grammar" when metadata suggests otherwise. Your citation and disclosure practices should align with records you could present if questioned.
International students face extra complexity when home-country norms differ from U.S. or U.K. institutional rules. Citation standards are global; permission standards are local. Read the syllabus in the language of enforcement, not assumption.
Libraries and writing centers publish local citation handouts that merge APA, MLA, or Chicago with campus AI rules. Use those when available—they reflect how your institution interprets national style guidance.
Best Practices for Ethical AI Citation
Start every project by reading the assignment's AI policy before opening a tool. Record the model version and date in your research log the way you record database access dates. When AI suggests sources, open each source yourself; never cite a bibliography the model invented.
Separate layers in your draft: color-code or comment AI-assisted paragraphs during writing, then rewrite them in your voice before submission. If a paragraph survives mostly intact, cite it. If you transformed it through critical analysis, your prose may no longer need AI citation—but disclosure may still apply.
Use verification checkpoints:
- Fact-check every proper noun, date, and statistic AI mentions.
- Cross-check definitions against course readings or textbooks.
- Run sensitive drafts through an AI detector only as one signal, not as proof of authorship—detectors complement citation; they do not replace it.
- Integrate the Proofly platform tools into a checklist: detect, rewrite, grammar-check, then cite and disclose.
When collaborating with instructors, ask before ambiguous tasks. Email creates a paper trail showing good faith. Most faculty prefer a question before submission over an integrity meeting after.
Teach peers what you learn. Citation norms for AI will keep evolving as models gain multimodal outputs and persistent memory. Students who document process carefully today set standards for tomorrow's policies—and protect their own reputations in a world where "who wrote this?" is a legitimate scholarly question.
Finally, treat citation as respect for your reader's time. A clear AI citation lets someone replicate your inquiry or skip your machine-generated detour entirely. That is the same reason we cite journal articles, datasets, and interviews: academic writing is a conversation, and every voice in the room should be named.