Table of Contents
1. What This Guide Covers and Why Citing AI Is Different
Picture this: You're a lawyer putting together a brief with the help of an AI assistant. It drafts a persuasive paragraph or suggests an on-point case. Can you just copy-paste and move on? Do you have to tell the court or your client that AI was involved?
Citing generative AI tools is not just about "giving credit." It's mainly about transparency: letting your reader (or client, or judge) understand what was produced by you versus produced (or materially shaped) by a tool. Transparency also means giving enough detail that another person could evaluate, replicate, or at least contextualize what happened.
Two things make AI citation unusual compared to citing a book or case law:
Unstable, Non-Retrievable Content
Many AI outputs are not inherently stable or publicly retrievable. The same prompt can yield different outputs later, and not every platform provides a shareable record.
"Authority" Concerns
AI tools often generate text that looks authoritative while being partially wrong, biased, or completely fabricated. Remember those attorneys sanctioned for citing ChatGPT-invented cases?
"The MLA explicitly warns that AI tools can hallucinate sources; it recommends consulting and citing the sources directly rather than treating the AI's response as an authoritative reference."
2. Key Definitions: AI Tool, Generative AI Output, and "Assistive" Tools
AI Tool (Generative AI Tool)
A software or platform that uses artificial intelligence models (often large language models or image generators) to produce content in response to user prompts. Examples: ChatGPT, Bard, Midjourney, DALL-E, GitHub Copilot.
Generative AI Output
The content produced by the AI tool — e.g. the paragraph of text, the code snippet, the image, the translated segment, or the data table the AI generated for you.
"Assistive" vs. Generative Tools
Tools that only check or refine formal criteria (spelling, grammar, style, formatting) are considered aids, whereas truly generative AI tools actually create new content in a substantive way. Running your draft through Grammarly = assistive. Having ChatGPT write a paragraph from scratch = generative.
What Counts as "Use" of AI
Modern guidance tends to interpret "use" broadly. You don't have to directly copy-paste AI output. If you used an AI tool to brainstorm ideas, outline your argument, suggest sources, rewrite a section, or "massage" your prose — that is a form of use that may need to be disclosed or documented.
3. The Three Transparency Buckets: Cite vs. Acknowledge vs. Disclose
A. CITE
As a source-like contribution
Use a formal citation (in bibliography, Works Cited, footnotes) when you quote, paraphrase, or incorporate content generated by AI in your work.
B. ACKNOWLEDGE
As a tool that helped you
If AI materially assisted you in the process but no verbatim content appears in your document, a full citation may not be necessary — but an acknowledgment is recommended.
C. DISCLOSE
To meet institutional requirements
Some cases require disclosure because of policy or ethical rules, even if you're not formally citing the AI as a source. Many journals and courts now mandate this.
4. When You Should Cite AI Tools or AI Output
If AI output appears in your work as content, cite it.
If you directly include text that ChatGPT wrote, a graph that an AI tool generated, code that GitHub Copilot suggested, or an image that DALL-E created, you need to cite the AI as the source of that content.
If AI shaped your thinking but you didn't use its words, acknowledge it.
If you used ChatGPT to help outline your argument or improve the style of a paragraph you wrote, and the final text is yours, acknowledge this help (e.g., "Drafting assistance was provided by ChatGPT").
If you used AI only to discover other sources, cite those sources instead.
If ChatGPT tells you about a case or article, always verify it from an official source. If you end up using that case or article, cite the case or article, not ChatGPT. (And if ChatGPT gave you a case that doesn't exist, definitely don't cite it!)
If your venue has explicit AI-use policies, follow their rules.
IEEE's publications require disclosure in the acknowledgments. Some courts require an attorney certification that sources were verified. Even absent a formal rule, consider the expectations of your audience — transparency safeguards your credibility.
5. What to Document So Your Use Is Reproducible and Defensible
To properly cite or disclose AI use, you need to have kept records of what you did. Multiple policies and guides converge on the same advice: record everything important about your AI sessions.
AI Interaction Checklist
Which AI system are you using? (e.g., ChatGPT, Bard, Midjourney)
Record the specific model (GPT-4 vs GPT-3.5, Claude Instant v1.2, etc.)
Document when you ran the prompt. AI outputs can change as models update.
Save the text of your question or prompt, word for word.
Save the full response, or at least the portion you're incorporating.
Temperature, creativity mode, plugins, web browsing, etc.
Did you quote it verbatim? Edit it heavily? Use it as inspiration?
Note if you corrected factual errors or added your own analysis.
Shareable link, appendix, or archived copy location.
Remember: WHAT you used (the output), WITH WHAT (which tool/model), WHEN, and HOW (with what prompt)
6. Core Citation Elements and a "Universal" Citation Template
Across all the styles, the solutions are converging on a set of core elements. If you know these elements, you can assemble a citation in any style:
Description of Content
Since AI outputs have no traditional title, describe what it is (e.g., "Response to the prompt 'Explain laches'")
Name of AI Tool
ChatGPT, Bard, DALL-E 3, GitHub Copilot, etc.
Model/Version
GPT-4, Midjourney v5, or date if that's how the tool identifies versions
Company/Developer
OpenAI, Google, Microsoft, Stability AI
Date of Generation
The date you got the output (full date is most useful)
Location or Link
URL to shareable version, or reference to appendix/archive
Universal Template
Description of content or prompt. Name of AI Tool, version/model (if known), Company/Publisher, Date of output. Location (URL or Archive).
7. Style-by-Style Guidance
7.1MLA (Modern Language Association)
MLA has explicitly addressed AI citation. Key point: there is no single author to credit for AI outputs, so the citation focuses on the tool as a source.
"Describe the nature of the protest rhetoric used in the 1920 Palmer Raids" prompt. ChatGPT, GPT-4 version, OpenAI, 8 Feb. 2024, chat.openai.com/chat.
If the output is longer than typical, you may include full prompts/responses in an appendix. The emphasis is on transparency and reproducibility.
7.2Chicago (and Bluebook for Legal Writing)
Chicago style suggests treating AI output as personal communication (not listed in bibliography, but cited in notes). However, if the output is essential, the 18th edition now allows bibliography entries.
1. Text generated by ChatGPT, OpenAI, March 7, 2024, https://chat.openai.com/chat.
For Bluebook (Rule 18.3): Include prompt author, model, exact prompt in quotes, date, and where output is saved (e.g., "on file with author").
7.3APA (American Psychological Association)
APA treats AI-generated text as algorithm output. The organization is the "author," and you describe the tool in brackets.
OpenAI. (2025). ChatGPT (GPT-4, January 2025 version) [Large language model]. https://chat.openai.com
In-text citation: (OpenAI, 2025). If the content isn't retrievable, you may cite it as personal communication instead.
7.4IEEE (Institute of Electrical and Electronics Engineers)
IEEE has a formal policy requiring authors to disclose use of AI tools in the acknowledgments. This is in addition to any references.
[1] ChatGPT response to "How does a solar panel work?" (GPT-4 model), OpenAI, generated Sept. 2, 2025. [Online]. Available: URL or DOI if available.
Acknowledgments: "The authors acknowledge the use of OpenAI's ChatGPT (GPT-4 model) in generating the initial draft of Section IV."
7.5Bluebook (Legal Citation for Attorneys)
The Bluebook's 22nd edition introduced Rule 18.3 — the first attempt at addressing citations of generative AI content.
Jane Doe, Prompt to ChatGPT (GPT-5.2) "Explain the doctrine of laches" (Mar. 1, 2026) (output on file with author).
⚠️ Practical Advice for Lawyers
Don't treat AI as a source of legal authority. Verify all case law or statutes with traditional sources and cite those primary sources. AI is a tool, not a source of law.
8. Non-Text Outputs: Images, Code, Tables, Data, Audio/Video
AI-Generated Images
Credit the AI just as you would a photographer. Caption: "Figure 1. Image of a fantasy city. DALL-E, version 3, OpenAI, generated 2 Feb. 2026."
AI-Generated Code
Add a footnote or code comment: "Generated with GitHub Copilot." Take responsibility for correctness and license compliance.
AI-Generated Data/Tables
Describe how the data was created (e.g., "Table 2 was generated by ChatGPT based on the prompt…"). Include prompt and response in an appendix.
AI-Generated Audio/Video
Provide a caption or note: "Audio generated using ElevenLabs AI voice software." Include length, tool, date, and link if available.
9. Academic Publishing and Professional Ethics Rules
Authorship and Accountability
AI tools cannot be listed as authors. They don't take responsibility for content, can't sign copyright agreements, and can't answer questions in peer review. Humans must disclose their use and are fully responsible for any inaccuracies.
Peer Review and Confidentiality
Do not feed confidential work into an AI tool without permission. Many AI platforms store inputs for training, which could violate confidentiality or copyright. Uploading client documents to a public AI could breach attorney-client privilege.
Emerging Rules in Legal Ethics
Bar associations say lawyers must understand the technology (competence), not rely blindly on AI for legal conclusions, and preserve client confidentiality. Some jurisdictions require lawyers to disclose AI assistance for certain filings.
Plagiarism and Originality
Presenting AI-generated text as if you wrote it is a form of academic dishonesty, unless allowed by the assignment. The safest route: always disclose and cite as needed, avoiding any suggestion that you tried to pass off AI's work as your own.
10. A Practical Workflow for Logging Prompts and Generating Citations
Create an AI Use Log
Every time you use an AI tool, jot down: Date, tool, model, what you asked, and what it gave you. Save a copy of any output.
Decide: Cite vs. Acknowledge vs. Disclose
Using the rules from Section 4, determine how to credit each AI use. Highlight portions of your draft that came from AI.
Preserve the Evidence
Save shareable links, copy text into documents, take screenshots or PDF prints. Keep files with meaningful names including tool and date.
Integrate Citations as You Write
Add citations immediately, even in rough form. Write placeholder footnotes like "{ChatGPT, OpenAI GPT-4, output from 2026-01-05}".
Check Consistency and Completeness
Before finalizing, ensure every AI use has appropriate citation/disclosure. Reconcile tool names, dates, and versions. Verify factual claims.
Include a Methodology Note
Add: "Note: We used [Tool] to assist with [task]. All AI-generated content was reviewed and verified by the authors."
Quick Reference Checklist
Before Using AI
- Check your venue's AI policy
- Set up an AI use log
- Understand confidentiality implications
After Using AI
- Record tool, model, prompt, output, date
- Save shareable link or archive output
- Verify any factual claims independently
- Cite or acknowledge appropriately
- Include methodology disclosure if required
By following a workflow like this, you transform AI from a possible liability into a well-documented asset. Transparency is your friend — and your shield.