INTRODUCTION
Criticism of The Bluebook is not new.[1] The Bluebook is an arguable form of hazing, with a long and storied history of making first-year students cry.[2] Judge Posner’s denouncement of the text is perhaps the most famous; he wrote: “I have not read the nineteenth edition. I have dipped into it, much as one might dip one’s toes in a pail of freezing water. I am put in mind of Mr. Kurtz’s dying words in Heart of Darkness—‘The horror! The horror!’—and am tempted to end there.”[3]
The fact that the text is widely disliked makes it an easy scapegoat. That is not to say, however, that the criticism is undeserved.
Past criticism has typically focused on aspects of the book that relate to its usability, not whether the citation format it prescribes is appropriate.[4] A notable exception to this pattern was the widespread condemnation of the sixteenth edition of The Bluebook, which changed the meaning of the signal “see” and the absence of a signal.[5] The response to this change “was so
*Jessica R. Gunder, Associate Dean of Faculty, Assistant Professor of Law University of Idaho College of Law. The idea for this Book Review came from a thoughtful conference discussion, and I am grateful to Rachel Croskery-Roberts, Dyane O’Leary, and Carolyn Williams for their many insights. I am indebted to Carolyn Williams and Rebekah Hanley for their useful feedback on this project. Finally, thank you to Traynor Underwood for his able research assistance (even though I may have called him a “tool” on page 6).
[1]. See, e.g., A. Darby Dickerson, An Un-Uniform System of Citation: Surviving with the New Bluebook (Including Compendia of State and Federal Court Rules Concerning Citation Form), 26 Stetson L. Rev. 53, 57 (1996) (“[B]ecause of its complexity and insularity, the Bluebook has attracted challengers . . . .”); Carol M. Bast & Susan Harrell, Has The Bluebook Met Its Match? The ALWD Citation Manual, 92 L. Libr. J. 337, 342 (2000) (“The rules concerning case names are overly complex.”); Richard A. Posner, Reflections on Judging 96–97 (2013) (noting complexity and stating that “there are declining marginal returns to complexity”).
[2]. Alex MacDonald, Citation Style Is a Cruel Mistress: A Review of the 21st Edition of The Bluebook, 20 Scribes J. Legal Writing 167, 169 (2022).
[3]. See Richard A. Posner, The Bluebook Blues, 120 Yale L.J. 850, 852 (2011).
[4]. See, e.g., Dickerson, supra note 1; Bast & Harrell, supra note 1.
[5]. Christine Hurt, Network Effects and Legal Citation: How Antitrust Theory Predicts Who Will Build a Better Bluebook Mousetrap in the Age of Electronic Mice, 87 Iowa L. Rev. 1257, 1268–71 (2002).
vehement that the House of Representatives of the Association of American Law Schools (AALS) passed a resolution at its January 1997 meeting condemning [the] changes . . . and encouraging its members, and law reviews, to use the signal rules in the Fifteenth Edition.”[6]
Despite its unpopularity and the availability of other citation manuals,[7] The Bluebook remains widely used at many law schools to teach legal citation format to law students, and it is relied on by law reviews and courts.[8]
The twenty-second edition of The Bluebook was released in May 2025.[9] This new edition includes a new rule—Rule 18.3—that crafts a citation format for legal writers to use when citing generative artificial intelligence (“AI”). This Book Review focuses on The Bluebook’s new generative AI rule, and it concludes that the citation format that The Bluebook prescribes within Rule 18.3 is deeply flawed.[10]
This Book Review proceeds in three parts. First, it examines the purpose of citations in legal writing and identifies circumstances in which the citation of generative AI output is appropriate. Second, it considers what The Bluebook requires of authors using generative AI technology and why The Bluebook’s requirements are inappropriate, focusing on: (1) errors within Rule 18.3 itself; (2) the unreasonable burden Rule 18.3 imposes; (3) Rule 18.3’s incompatibility with how generative AI technology is actually used; and (4) how the requirements imposed by Rule 18.3 violate attorney-client confidentiality requirements and work product protections. Third, and finally, it discusses why The Bluebook’s flawed approach matters and how it might be addressed.
I. SHOULD LEGAL WRITERS CITE TO GENERATIVE?
Generative AI technology became widely available to the public in the time period between the publication of the twenty-first edition of The Bluebook in 2020 and the release of the twenty-second edition of the text in 2025.[11][6] The Bluebook has historically sought to provide citation guidance when new types of sources are created, making the editors’ decision to include a new rule regarding generative AI unsurprising.[12]
That does not mean, however, that the topic of whether generative AI warrants citing is uncontroversial. This Part considers when and whether citations are appropriate for generative AI technology.
As in many things, context is key. And context is critically missing from Rule 18.3. Bluebook Rule 18.3 provides detailed guidance regarding how a writer should cite their use of generative AI technology but does not include any discussion or deliberation regarding when a citation is appropriate. The Bluebook is already quite voluminous (the new edition tips the scale at 390 pages), and this sort of deliberation certainly falls outside the scope of the book.
However, the absence of this guidance makes it unclear whether the citation form provided in Rule 18.3 is to be employed (1) when a legal author relies on a generative AI tool as a source of evidence or authority and is crafting a citation to refer to the platform as the source from which the evidence was obtained; or (2) when a legal author employs a generative AI tool as part of their drafting or research process.
It is easy to see why a citation would be appropriate in a situation in which a legal author is relying on a generative AI tool as a source of authority. For example, if an author wanted to highlight the unreliability of a generative AI tool by pointing to the fact that the tool crafted a pizza recipe that included glue as an ingredient to keep the cheese from falling off the slice, a citation—and preservation of the generative AI output—would be appropriate.[13]
The math is different, however, when considering whether citation is appropriate in the second instance: when a legal author employs a generative AI tool as part of their drafting or research process. The simple inclusion of citation rules within The Bluebook without any guidance regarding when Rule 18.3 should be employed ignores this critical distinction.
To evaluate whether a citation is needed, we must consider the many reasons why attorneys include citations in their written work product—and whether a citation to generative AI would serve to advance any of those purposes.[14] Legal scholars and The Bluebook itself provide various rationales for the use of citations, including:
- “The central function of a legal citation is to allow the reader to efficiently locate the cited source.”[15]
- Communicating information regarding the weight or importance of the cited authority.[16]
- Demonstrating the writer’s credibility through properly formatted citations.[17]
- Avoiding allegations of plagiarism.[18]
Each of these rationales—and whether they support the proposition that a legal writer should provide a citation when they use generative AI in their research or writing process—is discussed separately below.
First, if the central function of a citation is to aid a reader in finding a source and learning more about a particular topic, that basis does not support a legal author’s citation regarding their use of a generative AI tool. A “hallmark of GenAI is that it trains itself.”[19] Users do not receive identical output from a generative AI tool even if they replicate the same prompts.[20] Accordingly, possessing a citation for generative AI output would not aid a reader in locating that source, and this justification does not support the proposition that a citation is needed.
The second rationale underpinning the use of citations in legal writing is that a citation communicates important information to readers regarding the value of the authority the author is relying on. For example, “[t]he jurisdiction and date are important factors in weighing the precedential value of a case.”[21] Interestingly, a citation to generative AI as a research or writing tool would—in the current climate—have the effect of devaluing the writing and making it appear less credible,[22] irrespective of any steps the author took to verify the information that is being presented. The many instances of attorneys facing sanctions for their use of generative AI have been widely reported and shared across the legal profession, and as one judge noted in 2025, “[a]t this point, to be blunt, any lawyer unaware that using generative AI platforms to do legal research is playing with fire is living in a cloud.”[23] From the perspective of an advocate who is seeking to persuade their reader,[24] this rationale would not justify providing a citation to inform readers that a generative AI tool was used as part of the drafting or research process.
The third rationale for a legal writer to provide a citation is that the use of perfectly crafted citations lends credibility to the author of the piece. As stated by K.K. DuVivier: “[P]roper citations illustrate both your knowledge of the rules of etiquette for legal writing and your precision in following those rules. If your citations are sloppy, some readers may presume that your research and reasoning were done in a similarly uninformed and careless fashion.” The logic underpinning this sentiment is not particularly applicable to the citation of generative AI, however. This is because generative AI’s capacity to hallucinate[25] has given the technology such notoriety within the legal profession that a citation—even a perfectly formatted citation—is likely to have the opposite effect and reduce an author’s credibility with their reader.
Finally, citations are employed to avoid allegations of plagiarism and give credit to the source of an idea. Generative AI tools, however, are created by training the machine-learning model on large data sets. For example, one large language model was trained on a dataset that contained over 170,000 pirated and copyrighted books.[26] The origin stories for many generative AI tools smell like theft, which unsurprisingly has resulted in the filing of a number of lawsuits.[27] Ultimately, providing a citation to the generative AI tool that produced the output is not going to give credit where it is due: to the people from whom generative AI sourced the information.[28] Similarly, the citation to a generative AI source is unlikely to protect a legal writer from a plagiarism allegation related to the underlying source.[29]
Ultimately, generative AI is a tool, not an authority in and of itself. Much like a professor would not cite their research assistant, a partner at a law firm would not cite the legal assistant who put together the first draft of a motion, or a judge would not cite a law clerk who crafted the first draft of an order, one would not expect a legal writer to cite a generative AI tool.
An important line can be drawn between the two generative AI uses discussed at the start of this section: (1) when a legal author is relying on a generative AI tool as a source of authority and is crafting a citation to refer to the platform as the source from which the proposition was obtained; and (2) when a legal author employs a generative AI tool as part of their drafting or research process.
Professor Jayne Woods describes this distinction as between items that are “authority” and those that are not.[30] Another way to conceptualize this distinction is that a citation is appropriate when a writer is pointing to generative AI as the source of some sort of evidence. An example of the type of evidentiary use that would justify citation and preservation of the generative AI source would be: “It said to put glue on my pizza!” Conversely, when the author is employing generative AI as a research or writing tool, the rationales for legal citation do not support the use of citation.
While The Bluebook’s new rule is silent regarding when citation is appropriate, it will be critical that students, practitioners, and judges learn and recognize this distinction.
II. BLUEBOOK RULE 18.3 IS FLAWED[31]
This Part begins with a brief discussion of the differences between the Whitepages and Bluepages to establish the applicability of this new rule. It then considers problems with how the new rule is organized and what each subpart is attempting to capture. Third, this Part considers the specific citation formats specified within Rule 18.3, highlighting errors and inconsistencies in the rule. Finally, this Part considers how users will implement the specified formats and determines that they are inappropriate because of (1) the unreasonable burden Rule 18.3 imposes; (2) the Rule’s incompatibility with how generative AI technology is used; and (3) how the requirements imposed by the rule will conflict with attorney ethics requirements.
A. UNDERSTANDING THE DIFFERENCE BETWEEN THE WHITEPAGES AND THE BLUEPAGES
The Bluebook’s pages are color-coded. The Whitepages—a long section of the book, consisting of white-colored pages with a blue bar across the top —are geared toward legal scholarship and how to properly format a footnote in a law review article.[32] Conversely, the Bluepages—a brief section at the front of the book—are oriented around the proper citation format to be used when citing legal authority in non-academic legal documents.[33]
That delineation is not the end of the story, however. The Bluebook specifies that: “Where the Bluepages . . . are silent regarding the citation of a particular document, you may use the other rules in The Bluebook, referred to as the ‘Whitepages,’ to supplement the Bluepages.”[34]
The result is that the materials in the Whitepages cannot be viewed as siloed and specific to just legal scholarship, as they become widely applicable when a topic is not addressed within the Bluepages. Relevant here, the Bluepages in the twenty-second edition of The Bluebook do not address the citation of generative AI technology.[35] Accordingly, users are to turn to the Whitepages for guidance on this topic, and Rule 18.3 is applicable to both legal scholars and practitioners.
B. HOW IS BLUEBOOK RULE 18.3 STRUCTURED?
The text of Rule 18.3 is divided into three subparts and, as Cullen O’Keefe, Director of Research at the Institute for Law & AI, immediately noted, the differences between the content covered within the subparts are unclear.[36] The heading structure of Rule 18.3 is as follows:
18.3 AI-Generated Content
- (a) Large language models.
- (b) Search results.
- (c) AI-generated content.[37]
The fact that the third subheading mirrors the rule title is confusing but arguably the least of the problems here. Much more confusing is distinguishing between subpart (a) and (c) of this rule. As O’Keefe notes, “[c]ontent generated by LLMs is a type of AI-generated content!”
The fact that the third subheading mirrors the rule title is confusing but arguably the least of the problems here. Much more confusing is distinguishing between subpart (a) and (c) of this rule. As O’Keefe notes, “[c]ontent generated by LLMs is a type of AI-generated content!”[38]
While it is unclear from the headings themselves, the illustrative examples within each subpart lead to the conclusion that each subpart is concerned with providing a citation form for the different types of items that generative AI technology may create. Specifically, subpart (a) provides the following example:
Luke Cronin, Google Gemini Advanced, “Who would make a better Supreme Court Justice: Beyoncé or Taylor Swift?” (Mar. 29, 2024) (on file with the Columbia Law Review).[39]
Accordingly, it can be inferred that this subpart governs output in the form of text that is produced by a large language model. As detailed in Section I(A) above, however, it is unclear when this rule should be applied: Is this citation form appropriate only when the text output is being cited as evidence? Or is the citation form to be used whenever a legal author uses generative AI as a writing or research tool?
Subpart (b) gives users the following examples of proper citation form for “Search results”:
Bing, “The Bluebook”, 6,050,000 results (May 22, 2024) (on file with the Harvard Law Review).
Westlaw, +”Disclosure Controls” +”SEC”, 1 result (Mar. 30, 2024) (on file with the Yale Law Journal) (filtered by “Cases”, “2d Cir.”).[40]
These examples demonstrate that this portion of The Bluebook is geared around search engines and seeks to provide a citation form for instances in which a generative AI tool is involved in searching and providing responsive results in response to a prompt.[41]
What is less clear, however, is how widespread the use of Rule 18.3(b) should be. For example, is Rule 18.3(b) to be employed for anything the author searches for? On any platform (such as Google, Bing, Westlaw, or Lexis+)?
To highlight how broad this application might be, one only needs to look as far as my own experience writing this Book Review. I have used Google to search for alternate ways to phrase an idea (for example, “what is another way to say . . .”), a more precise release date for the twenty-second edition of The Bluebook, and other relevant queries. Which of these searches—if any—warrant citation under Rule 18.3(b)?
Finally, The Bluebook provides two examples within subpart (c)—subtitled “AI generated content”—of the proper format to be used:
Pablo Xavier, Photograph of Pope Francis in a Puffer Jacket, in This is Not a Real Photo of the Pope in a Puffy Coat, Snopes (Mar. 26, 2023), https://www.snopes.com/fact-check/not-real-photo-pope-in-puffy-coat/ [https://perma.cc/3EN7-BP4P] (generated by Midjourney AI).
Illustration of a Tornado on the Moon (on file with the Columbia Law Review) (generated by DALL-E 3).[42]
These examples are of a photograph and an illustration, indicating that this portion of the rule is concerned with generative AI output that is in the form of a picture, chart, or some other non-textual content.[43] This conclusion regarding subpart (c) is reinforced by language within the subpart that instructs users to employ the relevant Bluebook subrule for the particular content[44] and add certain details to reference that it was generated by AI.[45]
C. WHAT CITATION FORMAT DOES RULE 18.3 REQUIRE?
Each subpart of Rule 18.3 imposes different requirements for how those specific kinds of generative AI output should be cited. This Section considers those requirements and notes several errors or inconsistencies within Rule 18.3. Ultimately, these errors undercut The Bluebook’s goal of creating a uniform citation system and make it impossible for legal writers to discern whether they are complying with the citation format specified in Rule 18.3.
Subpart (a) of the new rule is titled “Large language models.” The rule starts by requiring authors who employ the technology to “save a screenshot capture of that output as a PDF to be stored on file.”[46] Additionally, it directs that the generative AI output should be cited by listing “the author of the prompt,[47] the name of the model used (including version number if the model is identified according to version), the exact text of the prompt submission in quotation marks, the date when the prompt was submitted, and a parenthetical indicating where the PDF is stored.”[48]
Problematically, however, the examples The Bluebook provides to accompany Rule 18.3(a)—which give guidance as to how the citation should look in practice—do not align with that description in two ways. First, while the text of Rule 18.3(a) specifies that it should include “the name of the model used,” the examples sometimes also include the name of the company that made the model and sometimes do not.[49] Second, the requirement that the citation contain “the exact text of the prompt submission in quotation marks” is followed in some instances and ignored in others.[50] Specifically, The Bluebook provides the following as an example of how to properly cite an AI source pursuant to Rule 18.3(a) in the “Basic Citation Forms” table at the start of Rule 18:
Akesh Shah, OpenAI GPT 3.5, Constitutional Confiscatory Takings (Jan. 15, 2024) (on file with the Columbia Law Review).
This exemplar certainly does not comply with the rule’s requirement that the citation contain “the exact text of the prompt submission in quotation marks.” These inconsistencies make it impossible for legal writers to discern whether they are complying with the parameters set by Rule 18.3(a).
Subpart (b) of the rule applies to “Search results” and specifies how users are to cite the results of search engine inquires. Such citations are to include “the name of the search engine . . . [and] the exact text of the query (including any Boolean operators employed) in quotation marks . . . and a parenthetical indicating where the PDF is stored.” The discussion of a PDF is a reference to subpart (a)’s requirement that users “save a screenshot capture of that output as a PDF.”
More detail is necessary, however, regarding what search results should be captured in the PDF. Should a user be capturing the first page of results? Or every page? This problem is highlighted by one of the examples The Bluebook employs for this subpart, which specifies that a Bing search for “The Bluebook” returned “6,050,000 results.” Presumably Rule 18.3(b) does not expect generative AI users to create a PDF cataloging all those results, but what is expected is unclear.
Finally, subpart (c) focuses on “AI-generated content.” It specifies that content “generated by AI should be cited” in keeping with the relevant Bluebook rule for that type of content, with a “parenthetical indicating that the content has been generated by AI, and the AI model used to generate the content.” Additionally, when the citation form would list the author, that spot should list “the name of the individual who submitted the prompt to the AI.” It contains no date requirement, which is inconsistent with the other Rule 18.3 subparts.
D. THE BURDEN IMPOSED BY RULE 18.3 IS UNCLEAR AND UNREASONABLE
The Bluebook’s citation requirements reflect an optimistic view of the technological skills possessed by attorneys. However, the reality is significantly bleaker. Empirical studies of individual attorneys, law students, and law firm staff have found that those groups are unable to fully use programs like Microsoft Word, PDF, and Excel.[51]
For example, recent assessments of law students—who have grown up as digital natives—have found that only around one-third of law students can perform the following tasks in Microsoft Word:
Accept/Turn-off track changes.
Cut & Paste.
Replace text.
Format font and paragraph.
Fix footers.
Insert hyperlink.
Apply/Modify style.
Insert/Update cross-references.
Insert page break.
Insert non-breaking space.
Clean document properties.
Create comparison document (that is, a redline).[52]
A study that looked at attorneys and their staff found similar technological deficiencies.[53] For example, while greater than 85% of lawyers and staff are able to convert an item to a PDF file, less than 5% can perform that task successfully when asked to create a PDF file that maintains active hyperlinks and converts internal headings into bookmarks.[54]
Ultimately, the citation requirements Rule 18.3 imposes are likely setting a bar that exceeds the technological skills possessed by the individuals who employ this technology. Specifically, subpart (a) directs individuals to “save a screenshot capture of that output as a PDF to be stored on file.”[55] As Professor Jayne Woods notes, for attorneys who do “not know how to take a scrolling screenshot (one that covers more than just the content visible on a single page) or convert files to PDF, the process can quickly become cumbersome, given that most conversations with a generative AI model likely span more than one visible page.”[56] As discussed in Section II(C) above, this problem is also relevant to subpart 18.3(b), which concerns search results.
The problem with the creation of a citation form that is too complex or challenging for users to employ is that it can “encourage[] practitioners and scholars to deviate from the standard norms.”[57] This technological hurdle will undercut a primary goal of The Bluebook: to create a uniform citation system.[58]
E. THE REQUIREMENT OF RULE 18.3 ARE INCOMPATIBLE WITH HOW GENERATIVE AI IS ACTUALLY USED
To prompt well,[59] an individual must tell AI the persona it should take on. They should explain the tone and the audience. They should specify the genre or the structure that the response should take. Ultimately, this may occupy several paragraphs. Additionally, the prompting process might include uploading documents: which would appropriately be considered part of the prompt as well. The use of a generative AI tool may also be iterative—where the program responds, and the user refines their questions and asks for additional information or specific changes—and that entire series would constitute one prompt.
The Bluebook’s new rule does not accord with this. It presumes a simple, one-sentence prompt, as demonstrated by the sample citations it provides.
Individuals who use generative AI tools recognize that output “can be the result of multiple rounds of prompting, generation, and refinement.”[60] For example, an attorney might be served with a motion for extension of time in one of their cases. If they turn to a generative AI tool for assistance with creating their response, they would presumably not just ask the program to “draft a response to a motion for extension of time.” Instead, a skilled prompter would upload the motion that was filed. They may upload sample responses they have employed in other cases so that the generative AI tool can emulate the style and tone of those documents. They would review the initial output for errors or omissions and ask the tool to craft a revised response that addresses those problems. They might ask the generative AI tool whether there are other arguments that could be raised in opposition to the motion. And so on. Ultimately, the final prompt may be something like, “Please rewrite that to capitalize the word Court.” While it would be the final prompt that was entered related to this work product, it surely would not be the prompt that The Bluebook is seeking. At the same time, however, full inclusion in a citation of the uploaded documents and the numerous, iterative prompts that were entered to create the final product would be unworkable.
Conversely—and similarly problematic—the generative AI output may not be the result of an iterative, back-and-forth process between the writer and the generative AI tool. Instead, the output may be the result of one single—albeit paragraphs long—prompt. This would be similarly unworkable in a citation format.[61] Ultimately, the prompts in Rule 18.3’s example citations are short and simple to a degree that is not representative of how generative AI is actually employed.[62]
Additionally, as noted in Section II(C) above, The Bluebook does not even follow its own specified format within the sample citations the text provides, as it fails to provide the text of the prompt in one example.[63] Between prescribing a citation format that is unworkable and failing to follow that format in the examples themselves, The Bluebook has adopted a flawed approach to citing generative AI output.
F. COMPLIANCE WITH BLUEBOOK RULE 18.3 WOULD VIOLATE AN ATTORNEY’S ETHICAL OBLIGATIONS
Context is key. Unfortunately, however, context is critically missing from Rule 18.3. As discussed in Part I, it is unclear whether the citation form provided in Rule 18.3 is to be employed (1) when a legal author is relying on a generative AI tool as a source of authority or “evidence” of some sort; or (2) when a legal author employs a generative AI tool as part of their drafting or research process. Should Rule 18.3 be interpreted to require citation when a legal author has chosen to employ a generative AI tool in their drafting or research process, the result could both violate an attorney’s duty of confidentiality and eviscerate work-product protections.
1. The Duty of Confidentiality
Attorneys owe certain duties to their clients, and primary among them is the duty of confidentiality.[64] While there are several exceptions to the duty to maintain client confidences, Model Rule of Professional Conduct 1.6(a) specifies that “[a] lawyer shall not reveal information relating to the representation of a client.”[65]
The duty of confidentiality is broad.[66] It covers more than the attorney-client privilege.[67] It exists even before a client decides to hire a particular attorney.[68] And it continues: extending beyond the termination of the attorney-client relationship,[69] and beyond even the client’s death.[70]
The duty of confidentiality is rooted in the idea that confidentiality is needed to encourage “client[s] to communicate fully and frankly with counsel.”[71] Ultimately, what is protected is not limited to what the attorney learns from the client. Instead, this broad duty applies to all “information relating to the representation of a client,” including information obtained from other sources as well.[72]
It is easy to see how someone employing the prompting techniques described in Section II(E) above might find themselves providing a generative AI tool with confidential information related to the representation. Indeed, this is an effective way to use a generative AI tool: by providing the tool with factual details about a case, the legal writer is more likely to obtain useful output. Nonetheless, these factual details—how a client described their pain, information about the client’s mental state, details regarding profits—would constitute information that relates to the representation.
Concerns have been raised regarding the incompatibility of an attorney’s duty of confidentiality and the use of generative AI.[73] However, an attorney does not automatically violate this duty if they choose to use such tools. For example, to avoid raising issues related to client confidentiality, some law firms have invested in in-house generative AI systems, in which client data is stored on a secure internal server, and some generative AI services are offering versions that include data protection features.[74] Use of those types of generative AI systems would not inculpate confidentiality concerns.[75] Alternatively, some clients may consent to the limited disclosure of confidential information in order to reap the cost-savings benefits that the generative AI tool may provide.[76]
What would run afoul of the duty of confidentiality, however, would be the disclosure of the factual details a legal writer employed in their generative AI prompts as a component of their citations in a motion filed with a court. While a client may be enthusiastic about their attorney’s use of generative AI, they likely would not be so enthusiastic if they then found their case subjected to a citation requirement that resulted in the disclosure of confidential information alongside their attorney’s arguments, as specified by Bluebook Rule 18.3(a).
2. Work Product Protections
The work product doctrine is codified in Federal Rule of Civil Procedure 26(b)(3).[77] The Rule specifies that a party may not ordinarily “discover documents and tangible things that are prepared in anticipation of litigation or for trial.”[78] Additionally, in the odd instance in which discovery of these materials is ordered, the Rule specifies that a court must “protect against disclosure of the mental impressions, conclusions, opinions, or legal theories.”[79]
The conversational and iterative nature of generative AI tools lends itself to the transmittal of an attorney’s mental impressions, conclusions, opinions, and legal theories. For example, an attorney may provide the program with details about a case, say that they are considering causes of action X and Y, and ask what other causes of action they should consider. Alternatively, an attorney might respond to generative AI output, indicating that they were concerned that a particular witness was not credible and asking it to rewrite the statement of facts with less focus on the statements from that witness.
This use of a generative AI tool would not result in the waiver of work-product protections[80] in and of itself because waiver is limited to situations in which “there is a significant likelihood that an adversary or potential adversary in anticipated litigation will obtain it.”[81] Nonetheless, while the generative AI use itself would not pose a significant risk concerning the disclosure of attorney work product, imposing the requirements of Rule 18.3(a) on practicing attorneys would have that effect.
III. WHY DOES THIS MATTER AND WHAT SHOULD BE DONE?
The Bluebook’s widespread adoption—as a textbook for 1L students and as a style guide for law reviews and courts—makes its creation of a flawed rule for generative AI particularly concerning. How legal writers should respond to this new rule varies based on whom is impacted.
A. STUDENTS AND PROFESSORS
Law schools are entrusted with ensuring that students graduate possessing many skills, and within the long list of items that law schools must impart is the ability to produce accurate citations. Accordingly, from a teaching perspective, faculty will need to teach students how to use Rule 18.3—including drawing a firm line around when it is appropriate to cite to generative AI output. Law professors simply should not be teaching students to violate their duty of confidentiality or expose work product. Accordingly, they will need to help their students learn to distinguish between when generative AI is being relied on as some sort of authority or evidence and requires a citation, and when it is merely being used as a tool. This responsibility is not limited to the instruction of first-year students, but extends to the supervision of students in clinics, faculty advising the editorial board of the school’s law review or moot court team, and professors teaching students in upper-division writing courses.
Students who serve as editors for one of their school’s journals will find themselves in the position of needing to adopt a generative AI policy. They will have to make a decision regarding when their journal will require authors to provide citations to generative AI tools in the scholarship they publish.
Additionally, given the contradictions and errors within the format prescribed by Rule 18.3, faculty (in both their roles as teachers and as legal writers) and students (in both their roles as legal writers and as law review editors) will also have to choose a citation format from the different variations shown in The Bluebook.
B. COURTS AND PRACTITIONERS
This rule is potentially more fraught for practitioners because of the standing orders issued by some judges in response to instances of attorney misconduct involving generative AI. The standing orders are varied, containing their own nuances and features.[82] While they are not specifically concerned with the citation of a generative AI tool, many of them require disclosure of some sort.
For example, Northern District of Illinois Magistrate Judge Jeffrey Cole issued an order containing a broad certification requirement, directing litigants to submit certifications if a generative AI tool was used for research.[83] The order entered by Federal District Court for the District of New Jersey Judge Evelyn Padin required a certification that identified “the ‘portion of the filing’ for which AI assistance was employed.”[84] Finally, the certification requirement imposed by Judge Michael M. Baylson of the Eastern District of Pennsylvania did not limit its application to generative AI tools but required disclosure when an attorney “used Artificial Intelligence (‘AI’) in the preparation of any complaint, answer, motion, brief, or other paper filed with the Court.”[85] This requirement is particularly problematic given the ubiquity of AI technology.[86]
The requirements these judges imposed on litigants are quite similar to a citation requirement, and—because of The Bluebook’s vagueness concerning when its proposed citation format should be used—it is possible that a judge might interpret the rule as requiring attorneys to provide citations that comply with Bluebook Rule 18.3 when using generative AI tools.
This potential use of Rule 18.3 is particularly concerning because some jurisdictions have rules naming The Bluebook as the official citation authority in that jurisdiction. For example, Florida’s Rules of Appellate Procedure include the requirement that citations “be in the form prescribed by the latest edition of The Bluebook: A Uniform System of Citation,”[87] and other courts have similar requirements.[88]
Accordingly, practitioners in jurisdictions that have standing orders concerning generative AI and practitioners whose jurisdictions have already adopted the twenty-second edition of The Bluebook will need to guard against any suggestion that Rule 18.3 applies to instances in which generative AI is used as a tool in order to avoid violating their duty of confidentiality and work-product protections.
CONCLUSION
Unfortunately, the citation format The Bluebook prescribes within Rule 18.3 is deeply flawed. These flaws range from (1) errors within Rule 18.3 itself; (2) the unreasonable burden Rule 18.3 imposes; (3) Rule 18.3’s incompatibility with how generative AI technology is actually used; and (4) how the requirements imposed by Rule 18.3 violate attorney-client confidentiality requirements and work-product protections.
These problems are not insignificant and will have to be addressed by legal writers. From a teaching perspective, faculty will need to teach students how to use Rule 18.3—including drawing a firm line around when it is appropriate to cite to generative AI output. Law review editors will need to set a citation policy for when a citation to generative AI is appropriate for the scholarly articles they publish. Practitioners will need to be prepared to defend against any suggestion that Rule 18.3 applies to instances in which generative AI is used as a tool in order to avoid violating their duty of confidentiality and work-product protections. Judges in jurisdictions with local rules that adopt the “current edition” of The Bluebook should consider changing that approach; and judges in jurisdictions that are considering adopting the twenty-second edition should evaluate Rule 18.3’s flaws when deciding whether adoption is appropriate. Additionally, given the contradictions and errors within the format prescribed by Rule 18.3, all legal writers who find themselves in the position of citing a generative AI tool will also have to choose a citation format from the different variations shown in The Bluebook—undercutting The Bluebook’s intended purpose of crafting a uniform citation system.
Editor’s Note – This article is republished with permission of the author with first publication, Southern California Law Review, April 2026, in Volume 99.
* * *
[1]. Id. at 1270–71 (citing Pamela Lysaght & Grace Tonner, Bye-Bye Bluebook?, 79 Mich. B.J. 1058, 1058 (2000).
[2]. See, e.g., Ass’n of Legal Writing Dirs. & Carolyn V. Williams, ALWD Guide to Legal Citation (7th ed. 2021); The Indigo Book: A Manual of Legal Citation (Christopher Jon Sprigman et al. eds., 2d ed. 2022).
[3]. See C. Edward Good, Will The Bluebook Sing the Blues?, Trial, Jan. 2001, at 78, 79 (discussing widespread adoption of The Bluebook). While total sales volume is unknown, in fiscal year 2020, The Bluebook “made $1.2 million in net profits” and the text yielded $16 million in net profits from 2011–2020. Dan Stone, Harvard-Led Citation Cartel Rakes in Millions from Bluebook Manual Monopoly, Masks Profits, Substack: Daniel Stone (June 9, 2022) https://danielstone.substack.com/p/legal-bluebook-profits-havard-yale-columbia-penn [https://perma.cc/WHG7-LFN9].
[4]. See The Bluebook: A Uniform System of Citation, at VIII (Columbia Law Review Ass’n et al. eds., 22d ed. 2025) [hereinafter The Bluebook].
[5]. The author is cognizant that The Bluebook is the work product of second and third-year law students and appreciates their efforts toward improving the book. For an insider’s perspective regarding the drafting process, see M. Burke Craighead, The Bluebook: An Insider’s Perspective, 124 Mich. L. Rev. (forthcoming 2026) https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5271305 [https://perma.cc/DY37-S666].
[6]. ChatGPT is not the only generative AI product that became available in this timeframe, however, the release of a demo of the program on November 30, 2022, “went viral” and led to much more widespread use of the technology. Bernard Marr, A Short History of ChatGPT: How We Got to Where We Are Today, Forbes (May 19, 2023, at 01:14 ET), https://www.forbes.com/sites/bernardmarr/2023/05/19/a-short-history-of-chatgpt-how-we-got-to-where-we-are-today [https://perma.cc/8RLJ-D4P3].
[7]. Craighead, supra note 10, at 18.
[8]. Jack Kelly, Google’s AI Recommended Adding Glue to Pizza and Other Misinformation—What Caused the Viral Blunders?, Forbes (May 31, 2024, at 06:00 ET), https://www.forbes.com/sites/jackkelly/2024/05/31/google-ai-glue-to-pizza-viral-blunders [https://perma.cc/E6M5-43VH].
[9]. Carolyn V. Williams & Margie Alsbrook, Presentation at the 2025 Association of Legal Writing Directors Biennial Conference: AI, Authorship, Accuracy, and Acknowledgment: Addressing New Challenges in Legal Citation (July 17, 2025) (thoughtfully discussing this topic).
[10]. The Bluebook, supra note 9, at 1.
[11]. Alexa Z. Chew, Citation Literacy, 70 Ark. L. Rev. 869, 879–80 (2018) (citing Ass’n of Legal Writing Dirs. & Coleen M. Barger, ALWD Guide to Legal Citation, at xxiii (5th ed. 2014)); see The Bluebook, supra note 9, at 4–5, 66–70.
[12]. Id. at 880 (citing Alexa Z. Chew & Katie Rose Guest Pryal, The Complete Legal Writer 366 (2016)); Jonathan Su, Thoughts on the Law School Experience, 80 U. Det. Mercy L. Rev. 535, 537 (2003); Bryan A. Garner, The Redbook: A Manual on Legal Style 147 (3d ed. 2013)).
[13]. Chew, supra note 16, at 880 (citing Christine Coughlin, Joan Malmud Rocklin & Sandy Patrick, A Lawyer Writes: A Practical Guide to Legal Analysis 127–28 (2d ed. 2013)).
[14]. Amanda K. Stephen, Citing GenAI: If, When, and How to Cite Generative Artificial Intelligence in Your Legal Writing, Wash. State Bar News (Mar. 7, 2024), https://wabarnews.org/2024/03/07/citing-genai [https://perma.cc/2KWF-5XVN] (citing Michael D. Murray, Artificial Intelligence and the Practice of Law Part 1: Lawyers Must Be Professional and Responsible Supervisors of AI, SSRN (July 19, 2023), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4478588 [https://perma.cc/7FPD-TDJE]).
[15]. Id. (citing Amanda K. Stephen, ChatGPT-3.5, “do you always provide an identical response for an identical prompt?” (Jan. 29, 2024)).
[16]. K.K. DuVivier, Legal Citations for the Twenty-First Century, Colo. Law., May 2000, at 45, 45.
[17]. Varun Magesh, Faiz Surani, Matthew Dahl, Mirac Suzgun, Christopher D. Manning & Daniel E. Ho, AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries, Stan. Univ. Hum.-Centered A.I. (May 30, 2024), https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries [https://perma.cc/D3LB-7KR4].
[18]. In re Marla C. Martin, 670 B.R. 636, 647 (Bankr. N.D. Ill. 2025). More recently, judges have been in the news for their mistaken reliance on false case law and citations created by generative AI. See, e.g., Shahid v. Esaam, 918 S.E.2d 198, 200 (Ga. Ct. App. 2025).
[19]. A different result may be reached if this is considered from a different perspective. For example, a court that does not want to be persuaded by hallucinated authority might want a citation requirement, believing that it could provide a useful “red flag” to warn the court. Importantly, however, that citation requirement would not satisfy any of the goals of citations from the legal writer’s perspective but would be imposed to undercut the validity of the information conveyed.
[20]. “AI technology ‘hallucinates’—which is to say that it lies about facts, or invents them,” ultimately creating incorrect, albeit convincing-sounding, answers. Jessica R. Gunder, Why Can’t I Have a Robot Lawyer? Limits on the Right to Appear Pro Se, 98 Tul L. Rev. 363, 406 (2024) (citing Jonathan H. Choi, Kristin E. Hickman, Amy B. Monahan & Daniel Schwarcz, ChatGPT Goes to Law School, 71 J. Legal Educ. 387, 394 (2022)).
[21]. See Lauren Leffer, Your Personal Information is Probably Being Used to Train Generative AI Models, Sci. Am. (Oct. 19, 2023), https://www.scientificamerican.com/article/your-personal-information-is-probably-being-used-to-train-generative-ai-models [https://perma.cc/QA7X-FFZ7].
[22]. Id.; see also Kate Knibbs, Every AI Copyright Lawsuit in the US, Visualized, Wired (Dec. 19, 2024, at 13:41 PT), https://www.wired.com/story/ai-copyright-case-tracker [https://perma.cc/UC4U-JL7B].
[23]. A small caveat is appropriate because some generative AI tools—like Lexis+ AI—will provide source information. See generally Serena Wellen, How Lexis+ AI Delivers Trustworthy Linked Legal Citations, LexisNexis (May 3, 2024), https://www.lexisnexis.com/community/insights/legal/b/product-features/posts/how-lexis-ai-delivers-hallucination-free-linked-legal-citations [https://perma.cc/3WLD-6CMB] (describing Lexis+ AI’s programmatic structure).
[24]. Some may argue that that citation of generative AI output is needed because “the general rule of thumb for legal memoranda and court documents is that you need a citation for every fact, thought, or opinion that comes from another source.” DeCarlous Y. Spearman, Citing Sources or Mitigating Plagiarism: Teaching Law Students the Proper Use of Authority Attribution in the Digital Age, 42 Int’l J. Legal Info. 177, 203 (2014) (citing Tracy L. McGaugh & Christine Hurt, Interactive Citation Workbook for The Bluebook: A Uniform System of Citation (2009)). Importantly, however, nothing generative AI creates is a fact, thought, or opinion. This accords with guidance from the U.S. Copyright Office regarding works produced with generative AI tools. See Stephen, supra note 19 (noting that the U.S. Copyright Office and courts have concluded that generative AI is not an author).
[25]. Jayne T. Woods, The New Bluebook Rules for Generative AI, App. Advoc. Blog (July 8, 2025), https://www.appellateadvocacyblog.com/2025/07/the-new-bluebook-rules-for-generative-ai.html [https://perma.cc/TE78-AL7R] (asserting that “generative AI platforms are not authoritative on anything, not even themselves”). For the purpose of this Book Review, the term “authority” refers to generative AI as the source of the information, not to indicate that it is accurate or correct.
[26]. Part II discusses many different flaws and problems with The Bluebook Rule 18.3. The cause of these specific issues is unknown, however, they may be an inevitable result of the sorts of problems discussed in Craighead, supra note 10 (reporting missed deadlines, contributing journals that did not turn in work or turned in their work late, and tight timelines).
[27]. The Bluebook, supra note 9, at 3.
[28]. Id.
[29]. Id. (citation modified).
[30]. See id. at 28–29.
[31]. Cullen O’Keefe, Citing AI in the New Bluebook, Substack: Jural Networks (June 14, 2025), https://juralnetworks.substack.com/p/citing-ai-in-the-new-bluebook [https://perma.cc/CPN8-XP2L].
[32]. The Bluebook, supra note 9, at 191.
[33]. O’Keefe, supra note 36.
[34]. The Bluebook, supra note 9, at 191.
[35]. Id. (errors in original).
[36]. But see O’Keefe, supra note 36 (criticizing Rule 18.3(b) as being misplaced, because while “modern search engines frequently incorporate AI-generated elements (for example, summaries), there is no indication that Rule 18.3(b) is uniquely concerned with that, and the examples given are not especially AI-forward”).
[37]. The Bluebook, supra note 9, at 191.
[38]. Alternatively, O’Keefe asserts that Rule 18.3 could be interpreted as requiring the use of Rule 18.3(c)—and not Rule 18.3(a)—in instances in which “an alternative citation format for the content is available.” O’Keefe, supra note 36 (emphasis omitted). While I find this interpretation of Rule 18.3 to be less plausible, the fact that it is an arguable interpretation demonstrates the confusing nature of this new rule.
[39]. As O’Keefe notes, this is likely a reference to rules like 18.7: Videographic Material; 18.8: Audio Recordings and Streaming; and 18.9: Images. Id.
[40]. The Bluebook, supra note 9, at 191.
[41]. Id.
[42]. While this is not a focus of this Book Review, it is likely that the requirement that the citation list the author of the prompt will become problematic. For example, the Wharton GENAI prompt library specifies that individuals who use the library’s prompts “give credit” to the creators. Attribution 4.0 International, Creative Commons, https://creativecommons.org/licenses/by/4.0 [https://perma.cc/B297-VUQP].
[43]. Id.
[44]. O’Keefe, supra note 36 (noting that the Shah example includes “OpenAI,” while the company name is not included in the Illustration of a Tornado on the Moon example).
[45]. Id. (highlighting that the Shah example does not contain the text of the prompt).
[46]. Casey Flaherty, Developing Technological Competency as a Lawyer, Mich. B.J., June 2017, at 70, 70.
[47]. Darth Vaughn & Casey Flaherty, Tech Comes Naturally to ‘Digital Native’ Millennials? That’s a Myth, A.B.A. J.: Legal Rebels (Oct. 13, 2016, at 08:30 CT), https://www.abajournal.com/legalrebels/article/tech_comes_naturally_to_digital_native_millennials_thats_a_myth [https://web.archive.org/web/20250907180909/https://www.abajournal.com/legalrebels/article/tech_comes_naturally_to_digital_native_millennials_thats_a_myth].
[48]. Flaherty, supra note 51.
[49]. Id. at 72 (noting other disappointing statistics, including the fact that only 25% of lawyers and staff can correct agreement numbering within a reasonable timeframe and only 5% can update cross-references).
[50]. The Bluebook, supra note 9, at 191.
[51]. See Woods, supra note 30. Additionally, Professor Woods notes that “the process of screenshotting and converting these files is time-consuming,” which is likely to lead to less compliance with Rule 18.3’s prescriptions. Id.
[52]. Margie Alsbrook, Untangling Unreliable Citations, 37 Geo. J. Legal Ethics 415, 419 (2024) (warning that the creation of a too-complex citation system will encourage deviation).
[53]. For a discussion of how The Bluebook’s editors consider changes and various concerns regarding how changes may be accepted by users, see Craighead, supra note 10, at 13–14.
[54]. For guidance on designing effective generative AI prompts, see How to Create Effective AI Prompts (With Examples), Grammarly (May 2, 2024), https://www.grammarly.com/blog/ai/generative-ai-prompts [https://web.archive.org/web/20250930224434/https://www.grammarly.com/blog/ai/generative-ai-prompts]; Legal ChatGPT: Tips, Prompts, and Use Cases, A.B.A.: L. Prac. Div. (Mar. 21, 2025), https://www.americanbar.org/groups/law_practice/resources/law-technology-today/2025/legal-chatgpt-tips-prompts-and-use-cases [https://web.archive.org/web/20250912164130/https://www.americanbar.org/groups/law_practice/resources/law-technology-today/2025/legal-chatgpt-tips-prompts-and-use-cases]; Prompt Library, Wharton, Univ. Pa.: Wharton Generative AI Labs, https://gail.wharton.upenn.edu/prompt-library [https://perma.cc/MD6L-KCPG]; Williams & Alsbrook, supra note 14.
[55]. O’Keefe, supra note 36; see also Woods, supra note 30 (noting that conversations with a generative AI tool “are iterative, with both questions and responses building on prior discussions”).
[56]. See O’Keefe, supra note 36 (calling the inclusion of a paragraphs-long prompt in a citation “absurd”).
[57]. The Bluebook, supra note 9, at 191.
[58]. See O’Keefe, supra note 36 (discussing this error).
[59]. See, e.g., Commonwealth v. Downey, 793 N.E.2d 377, 381 (Mass. App. Ct. 2003) (“It is axiomatic that among the highest duties an attorney owes a client is the duty to maintain the confidentiality of client information.”).
[60]. Model Rules of Pro. Conduct r. 1.6(a) (A.B.A. 2025).
[61]. In re Bryan, 61 P.3d 641, 656 (Kan. 2003) (noting that the duty of confidentiality is “interpreted broadly, with the exceptions being few and narrowly limited”).
[62]. Douglas R. Richmond, Lawyers’ Duty of Confidentiality and Clients’ Crimes and Frauds, 38 Ga. St. U. L. Rev. 493, 496–97 (2022); Sarah Helene Sharp, On Being a Blab or a Babbler: The Ethics and Propriety of Divulging Client Confidences, 10 Geo. J. Legal Ethics 79, 80 (1997) (“Unlike the attorney-client privilege, the current rule to maintain client confidences applies not simply to what clients say to their attorneys, but to matters that attorneys discover during the course of representation and regarding which clients may not have a reasonable expectation of privacy.”).
[63]. Model Rules of Pro. Conduct r. 1.18 (A.B.A. 2025); Grace M. Giesel, The Attorney-Client Relationship in the Age of Technology, 32 Miss. C. L. Rev. 319, 320 (2013).
[64]. Sharp, supra note 67, at 79 n.2 (citing Model Rules of Pro. Conduct r. 1.6 cmt. 21 (A.B.A. 2025)).
[65]. Patrick Shilling, Attorney Papers, History and Confidentiality: A Proposed Amendment to Model Rule 1.6, 69 Fordham L. Rev. 2741, 2742 (2001).
[66]. Swidler & Berlin v. United States, 524 U.S. 399, 407 (1998).
[67]. Model Rules of Pro. Conduct r. 1.6 (A.B.A. 2025); Restatement (Third) of the Law Governing Lawyers § 59 (A.L.I. 2000).
[68]. See, e.g., Fl. Bar Bd. of Governors, Bd. Rev. Comm. on Pro. Ethics, Op. 24-1 (2024).
[69]. Joe Regalia, Lex Ex Machina: Forging a New Ethical Framework for AI and Technology, 55 Cumb. L. Rev. 53, 98 (2024); Andrew M. Perlman, The Legal Ethics of Generative AI, 57 Suffolk Univ. L. Rev. 345, 348 (2024).
[70]. Regalia, supra note 74, at 99.
[71]. Client consent is an exception to the duty of confidentiality. Model Rules of Pro. Conduct r. 1.6(a) (A.B.A. 2025).
[72]. The doctrine was created by the Supreme Court in Hickman v. Taylor, 329 U.S. 495, 511–12 (1947).
[73]. Fed. R. Civ. P. 26(b)(3)(A).
[74]. Fed. R. Civ. P. 26(b)(3)(B).
[75]. See, e.g., Behnia v. Shapiro, 176 F.R.D. 277, 279 (N.D. Ill. 1997) (noting that the disclosure of work product to a third party can result in a waiver in some instances).
[76]. Restatement (Third) of the Law Governing Lawyers § 91(4) (A.L.I. 2000). Typical use of a generative AI tool is unlikely to present a “significant” risk that an adversary would obtain attorney work product.
[77]. See Jessica R. Gunder, Rule 11 Is No Match for Generative AI, 27 Stan. Tech. L. Rev. 308, 348–60 (2024) (detailing the nuances of many of the standing orders that were entered); Generative Artificial Intelligence (AI) Federal and State Court Rules Tracker, LexisNexis (Jan. 30, 2026), https://advance.lexis.com/api/permalink/9e84d41a-308e-4f79-a03d-e6a9a817f131/?context=1000522 [https://perma.cc/JLL6-X37D].
[78]. Gunder, supra note 82, at 353 & n.244 (citing Magistrate Judge Jeffrey Cole, The Use of “Artificial Intelligence” in the Preparation of Documents Filed Before This Court (July 25, 2023), https://www.ilnd.uscourts.gov/_assets/_documents/_forms/_judges/Cole/Artificial%20Intelligence%20standing%20order.pdf [https://perma.cc/GQ9U-8CGY]).
[79]. Id. at 352 (citing Judge Evelyn Padin, General Pretrial and Trial Procedures (Nov. 13, 2023), https://www.njd.uscourts.gov/sites/njd/files/EPProcedures.pdf [https://perma.cc/MUH2-KVCL]).
[80]. Id. at 353 & n.245 (citing Judge Michael M. Baylson, Standing Order Re: Artificial Intelligence (“AI”) in Cases Assigned to Judge Baylson (June 6, 2023), https://www.paed.uscourts.gov/sites/paed/files/documents/locrules/standord/Standing%20Order%20Re%20Artificial%20Intelligence%206.6.pdf [https://perma.cc/XTD4-6USB]).
[81]. Id. at 356–57 (“[W]hile generative AI is becoming ubiquitous, artificial intelligence has already crossed that threshold, with everyday-use programs like Microsoft Word employing extractive AI technology.”).
[82]. Fl. R. App. P. 9.800(p).
[83]. See Ind. R. App. P. 22; E.D. Mich. Loc. R. 7.1, D. Mont. Loc. R. 1.5(d); N.M. Sup. Ct. Gen. R. 23-112(F).