New Bluebook citation rules for AI content and LLMs are generating controversy.
Here at FindLaw, we’ve spent a lot of time covering the increasingly relevant topic of using artificial intelligence in the law. As you likely already know, citing LLMs or any AI-generated content should be done sparingly, if at all, and never relied on for legal accuracy. Yet, the Bluebook (the legal world’s playbook for citing your sources) has tackled the problem head-on (sort of) in its 22nd edition, creating rules regarding how to properly cite to AI content.
These rules have been generating some controversy since they were first revealed. Cullen O’Keefe, the Director of Research at the Institute for Law & AI, started the online conversation over on Jural Networks. O’Keefe and others argue that the rules are inconsistent and unclear.
Fear not, law students and diligent legal scholars. We’ll cover the controversy and walk you through the new rules.
Wait … I Can Cite AI?
Hopefully, that was your first question when reading the intro paragraph. The Bluebook does not really address when it is appropriate to cite to AI. This is one of the main criticisms being levied at the new rules.
Generative AI’s primary use is in broad legal research or spot-checking your work, and for that, it should be cited cautiously, if ever. That’s not to say research can’t start with querying an AI, or that legal research should never be aided by it. But once AI has done some legwork, it’s easy to directly refer to the primary sources from which AI draws its responses (and which you have been able to check for accuracy). Additionally, there is a concern that citing certain queries could potentially reveal protected attorney-client information. If you’re trying to find the appropriate amount of pain and suffering damages in a personal injury case, for example, citing to prompts where you discuss the client’s state of mind could be protected information.
That means it should be a very rare occurrence to cite AI at all. It’s an aid to research, and just like you wouldn’t directly cite your law clerk, you wouldn’t need to cite AI in almost all cases.
So, What’s the Point?
There may be some extremely limited situations in which citing a generative AI tool is appropriate. Essentially, the Bluebook's new rules are to treat AI-generated content as "unpublished material," similar to an internal memo. In those situations, the key is to provide enough information for a reader to understand what AI tool was used, when it was accessed, and what specific output was generated.
The Rules
There are three separate rules depending on the type of AI-generated content you are using. They are:
Rule 18.3(a): “Large language models”
Rule 18.3(b): “Search results”
Rule 18.3(c): “AI-generated content”
What’s the difference between LLMs and AI-generated content? This is another issue some lawyers have with the Bluebook's attempt. People typically use LLMs to generate text, which is often referred to as “AI-generated content.” Most AI-generated content comes from LLMs. Absent further clarification, Rule 18.3(c) would appear to refer to non-textual AI output, such as images, audio recordings, and video.
If you are citing AI-generated text, the Bluebook does have a recommended citation format. It’s this: Author, Title of AI Model (Version/Date), at [Specific Instruction/Prompt], Date of Access, parenthetical to where the pdf is stored.
Importantly, Bluebook rule 18.3 requires authors to save screenshots of AI output as PDFs. Well and good, but critics point out that this can be a bit of a technical challenge. It may help to know how to create scrolling screenshots. To do this, you generally need either a tool like ShareX or a browser extension, as conversations with ChatGPT, Claude, or other AI assistants often go beyond the visible screen. Citing that would go something like this: (Joe Fawbush, Perplexity AI (Claude Sonnet 4.5), "How can you create a scrolling screenshot of AI-generated content?" (Oct. 7, 2025) (on file with [Joe Fawbush/FindLaw]).
As most people use ChatGPT and other generative AI conversationally, with several follow-up questions to the initial query, you can at least capture full conversations with generative AI in a pdf, even if the correct citation might be unclear.
Changes to Rule 18.3 Likely
While the controversy over Bluebook citations to AI isn’t likely to pop up frequently in your work, the discussion around Bluebook Rule 18.3 does show how lawyers are using LLMs in their daily lives. Artificial intelligence may not be able to write a law review article worth reading anytime soon, but it appears to be here to stay as a research tool in both our work and personal lives. It seems likely that these rules will be at least somewhat revised in the next Bluebook edition, but in the meantime, you have some guidance if you encounter a rare instance where citing an AI tool is appropriate.
Related Resources
- Why Judges Are Getting Creative with AI-Abusing Attorneys (FindLaw’s Practice of Law)
- Does the Slowdown of AI Adoption by Large Companies Indicate Changing Attitudes? (FindLaw’s Practice of Law)
- District Judge’s Order Reeks of AI (FindLaw’s Practice of Law)
