What to Know First
Citing AI tools in academic papers is no longer optional—it’s a requirement for most institutions as of 2026. Over 70% of universities now have formal AI-use policies that mandate disclosure and proper citation when AI-generated content appears in your work. But here’s the critical detail most students miss: each major citation style treats AI differently, and using the wrong format can flag your paper as a formatting error or, worse, as undisclosed AI use.
This guide covers the exact citation formats for APA 7th Edition, MLA 9th Edition, Chicago 17th Edition, and Harvard referencing—with copy-paste examples for each, plus practical guidance on when citation is required versus when a disclosure statement alone suffices.
When Do You Actually Need to Cite AI?
Before formatting citations, understand the core rule that applies across all styles:
Cite AI when you use its output directly in your work: quoted text, paraphrased arguments, generated data, summaries you incorporate, or content that shaped your final text.
Don’t cite AI when you use it as a process tool—like a spell checker, grammar improver, or brainstorming aid that doesn’t contribute directly to your published text. Running your draft through an AI tool for editing doesn’t require a citation, but asking ChatGPT to generate three paragraphs that you include in your essay does.
The gray area is where AI helped you develop ideas or refine phrasing. Most institutions now recommend at least a disclosure statement, even when formal citation isn’t required.
Practical rule: If you can point to specific AI-generated text that appears in your paper—quoted, paraphrased, or substantially adapted—you need a citation. If the AI only helped you think, not produce words you included, a disclosure statement is typically sufficient.
APA Style 7th Edition: AI Citations (2026 Update)
APA treats AI-generated text as algorithmic output, with the AI company listed as the author. This is the most widely used format in social sciences and research.
Reference List Format
For a specific chat with a shareable URL (preferred):
OpenAI. (2026, February 15). Comparison of renewable energy policies [Generative AI chat]. ChatGPT. https://chatgpt.com/share/[unique-id]
For a general reference when no shareable URL exists:
OpenAI. (2026). ChatGPT (Feb 15 version) [Large language model]. https://chat.openai.com/
For Claude:
Anthropic. (2026). Claude (3.5 Sonnet version) [Large language model]. https://claude.ai/
In-Text Citations
- Parenthetical: (OpenAI, 2026)
- Narrative: OpenAI (2026) noted that…
Key APA-Specific Rules
- Author is the company, not the tool. It’s “OpenAI,” not “ChatGPT.” It’s “Anthropic,” not “Claude.”
- Date is the generation date, not the model release date. If you used GPT-4o on February 15, 2026, cite February 15, 2026.
- Include the exact model version in parentheses after the title.
- APA recommends including full AI-generated text as an appendix if you directly quote it, since chat outputs are not permanently retrievable.
- If no shareable URL exists, use the general homepage URL of the AI tool as the source element.
APA guidance source: APA Style AI References (Updated April 2026)
MLA Style 9th Edition: AI Citations (2025-2026 Update)
MLA’s approach differs fundamentally from APA. MLA does not treat the AI tool as an author. Instead, the prompt or a description of generated content comes first. MLA updated its guidance in August 2025 and continues to refine the format.
Works-Cited Format
With a shareable URL:
“Identify the main themes in The Great Gatsby movie trilogy” prompt. ChatGPT, model GPT-4o, OpenAI, 15 Feb. 2026, https://chatgpt.com/share/[unique-id]
Without a shareable URL:
“Summarize the key arguments for universal basic income” prompt. ChatGPT, model GPT-4o, OpenAI, 15 Feb. 2026.
For Claude:
“Explain quantum entanglement for a general audience” prompt. Claude, model 3.5 Sonnet, Anthropic, 15 Feb. 2026.
In-Text Citations
The AI suggested that “the trilogy’s central theme is the corruption of power” (“Identify the main themes”).
Key MLA-Specific Rules
- The prompt text is the “title” of the entry. Put it in quotation marks.
- Italicize the AI tool name as the container title.
- Include both the model version and the date you generated the content.
- MLA does not treat AI as an author. This follows publisher policies, including those of Cambridge University Press and MLA’s own journal PMLA.
- If you quote or paraphrase AI output, use a shortened version of the prompt in your in-text citation.
- For AI images, use the prompt description followed by the AI tool name, model version, and date in a figure caption.
MLA guidance source: MLA Style Center — Citing Generative AI (Updated August 2025)
Chicago Manual of Style 17th Edition: AI Citations
Chicago takes a notably different approach. It treats AI-generated content like a personal communication (similar to citing a phone call or private conversation). This means no bibliography entry—only footnotes or endnotes.
Footnote Format
If the prompt is included in your text:
1 Text generated by ChatGPT-4o, OpenAI, May 18, 2026, https://chat.openai.com/.
If you need to include the prompt in the note itself:
1 ChatGPT-4o, response to “Explain the historical significance of the Magna Carta,” OpenAI, May 18, 2026.
Bibliography Entry
Chicago generally advises against including AI sources in the bibliography. However, if you generated a shareable, permanent link to the conversation, a bibliography entry is acceptable:
OpenAI. Response to “Explain the historical significance of the Magna Carta.” ChatGPT-4o. May 18, 2026. https://chatgpt.com/share/[unique-id]
Key Chicago-Specific Rules
- Do not include AI sources in your bibliography unless you have a publicly accessible, shareable URL.
- Use footnotes or endnotes only for in-text attribution.
- Include the model name and model version in the note.
- Include your prompt if it’s relevant to the reader.
- Chicago Manual of Style official FAQ confirms this treatment: Citation, Documentation of Sources #422
Harvard Referencing Style: AI Citations (2026 Update)
Harvard follows a structure similar to APA but has its own distinctive formatting. Since there is no single official body governing Harvard style, university-level guidance varies—so always check your institution’s specific requirements.
Reference List Format
OpenAI (2026) ChatGPT (Feb 15 version) [Large language model]. Available at: https://chat.openai.com/ [Accessed: 15 February 2026].
Anthropic (2026) Claude (3.5 Sonnet version) [Large language model]. Available at: https://claude.ai/ [Accessed: 15 February 2026].
In-Text Citations
- Parenthetical: (OpenAI 2026)
- Narrative: OpenAI (2026) noted that…
Key Harvard-Specific Rules
- Include the access date since AI outputs are not static and may change over time.
- Specify the model version in the title if possible.
- Harvard author-date format credits the developer as author with the release year.
- Some universities using Harvard style have published their own AI citation addenda, so check your institution’s library guides first.
Harvard guidance source: UCD Harvard Style Guide — Generative AI
Quick Comparison Table
| Element | APA 7th | MLA 9th | Chicago 17th | Harvard |
|---|---|---|---|---|
| Author | AI company | Not an author | N/A (footnote) | AI company |
| Title | Chat title or model name | Your prompt | Description in note | Model name |
| In bibliography? | Yes | Yes | No (footnote only) | Yes |
| Prompt included? | Optional (appendix) | Required (as title) | Optional | Optional |
| Date used | Generation date | Version + generation | Generation date | Access date |
| In-text style | (Company, Year) | (“Shortened prompt”) | Footnote | (Company Year) |
When to Cite AI vs. When to Disclose Only
The line between formal citation and disclosure-only can be blurry. Here’s a practical decision framework:
| Use Case | Citation Required? | Recommendation |
|---|---|---|
| Quoting AI-generated text directly | Yes | Formal citation in required style |
| Paraphrasing AI content | Yes | Formal citation in required style |
| Using AI to brainstorm ideas | Disclosure only | Note in acknowledgments or methods |
| Using AI for editing/proofreading | No citation | Standard process tool |
| Using AI to find secondary sources | Cite original source | Never cite AI as the source of evidence |
| Using AI to generate images/data | Yes | Citation in required style |
| Using AI as a research conduit | Cite original source | Click through to cited sources and cite them directly |
AI Image and Figure Citations
If you used an AI image generator (DALL-E, Stable Diffusion, Midjourney) to create figures for your paper, each style handles it differently:
APA:
OpenAI. (2026). A watercolor painting of a sunset over mountains [AI-generated image]. DALL-E 3. https://openai.com/
MLA:
“Create an expressionist-style image of two people looking at the ocean” prompt, DALL-E, version 3, OpenAI, 23 Sept. 2024.
Chicago:
1 Image generated by DALL-E 3, OpenAI, Sept. 23, 2024.
Harvard:
OpenAI (2024) DALL-E 3 [Image generator]. Available at: https://openai.com/ [Accessed: 1 Oct. 2024].
The AI Disclosure Statement
Beyond formal citations, many institutions now require a separate AI disclosure statement. This is a paragraph—usually placed at the end of your paper, before references—explaining exactly how you used AI.
Example Disclosure Statement
This paper was written by the author with assistance from ChatGPT (OpenAI, GPT-4o). The AI tool was used to brainstorm outline structures and suggest phrasing for the literature review section. All AI-generated text was substantially revised and verified for accuracy by the author. The AI was not used for data analysis, original argumentation, or generating citations.
Why disclosure matters: Even if your professor doesn’t require it, including one demonstrates academic honesty and protects you from accusations of undisclosed AI use. Most institutions with AI policies expect at least a disclosure statement when AI contributed to the drafting process.
Common Mistakes When Citing AI (and How to Avoid Them)
1. Citing the AI Model Instead of the Company
Wrong: “ChatGPT (2026)”
Correct: “OpenAI (2026)” — The author is the company that created the model.
2. Using the Model Release Date Instead of Your Usage Date
Wrong: Citing the date GPT-4 launched.
Correct: Using the specific date you generated the content (e.g., “February 15, 2026”).
3. Citing AI as the Original Source
Wrong: Citing AI for factual claims or research evidence.
Correct: Never cite AI as the original source. If AI surfaces a claim, find the primary source and cite that directly. AI tools can hallucinate false references.
4. Forgetting to Save the Conversation
If your citation style requires a URL (APA and MLA with links), save or share the chat before you close it. You cannot reconstruct an AI conversation later, and many styles expect a shareable link.
5. Assuming Your Institution Follows the Standard Style Guide Exactly
Many universities have published their own AI citation amendments that add or modify requirements. Always check your school’s library guides.
How to Verify You’re Citing AI Correctly
Before submitting your paper, run this checklist:
- Is the author the company, not the tool? (OpenAI, not ChatGPT)
- Is the date the generation date, not the model release date?
- Is the model version specified? (GPT-4o, Claude 3.5 Sonnet, Gemini 1.5)
- Is a shareable URL included where the style requires it?
- Is the prompt described in MLA format? (MLA requires prompt as title)
- Is AI excluded from Chicago bibliography? (Footnotes only)
- Is an AI disclosure statement included if required?
Why This Matters More Than Ever in 2026
Citation guidelines for AI are still evolving. APA and MLA have both updated their recommendations multiple times since 2023, and more changes are likely as AI use becomes standard in academic work. The core principles remain stable, but the formatting details shift.
The safest approach is to check your style guide’s official website at the start of each semester and confirm your institution’s AI policy. When in doubt, default to the most cautious standard: cite when uncertain, disclose when unclear, and always verify the accuracy of any information AI helped produce.
Related Guides
- AI in Grant Writing: Ethical Use, Disclosure, and Detection Concerns (2026 Guide)
- AI-Generated References and Citations: Detection and Ethical Use [2026 Guide]
- AI as Co-Author: Guidelines for Transparency in Academic Publishing
- Fair Use in Academia: How to Legally Use AI-Generated Content in Research Papers
- AI-Generated Bibliographies: Why They’re Problematic and How to Verify Sources
Ready to Verify Your Academic Work?
If you want to confirm that your academic papers are fully original and properly cited, try Paper-Checker’s free plagiarism and AI detection tools to scan your documents before submission.
Academic Integrity Back-to-School Checklist: Your Complete Guide for Fall 2026 Semester
In Brief The academic integrity landscape in 2026 is fundamentally different from previous years. Universities have abandoned blanket AI bans in favor of course-specific syllabus policies, and detection tools are increasingly treated as screening instruments rather than definitive proof of misconduct. This checklist walks you through every stage of the semester—from pre-assignment policy review to […]
AI Content Provenance and Watermarking: The New Era of Academic Integrity (2026)
In Brief AI content provenance and watermarking are fundamentally changing how educational institutions verify authentic student work. Instead of relying on detection algorithms that produce 43-83% false positive rates on human writing, universities are shifting to cryptographic proof systems. The Coalition for Content Provenance and Authenticity (C2PA) creates tamper-evident digital credentials, while Google’s SynthID embeds […]
How AI Detectors Actually Work: Understanding Perplexity, Burstiness, and Stylometry Explained
You’ve probably heard that AI detectors can tell whether your essay was written by a machine or a human. But here’s the thing most people don’t understand: these detectors don’t actually “read” your writing at all. They’re measuring the mathematical fingerprints left behind by how text is generated. And understanding those fingerprints—specifically three metrics called […]