Discovering AI: The Relevance of Relevance and Proportionality

Discovering AI is a series that explores the effect—or lack thereof—that artificial intelligence (as a discoverable object) has had on the discovery process.  Does AI-generated material have a heightened threshold of discoverability?  Should it? Do courts need new AI-specific guidelines to regulate discovery?  And what should organizations do if, during the course of litigation, an opposing party seeks AI-generated or AI-based materials?  Answers to these questions and more continue to develop as courts and rules committees across the United States are confronted with the proliferation of AI through workplaces, law firms, and just about every aspect of daily life.  The goal of this series is to provide helpful hints to clients of any industry about a topic that is exciting in its novelty while far-reaching and consequential in its impact. Join us as we navigate the evolving intersection of AI and discovery in the articles to come.

As courts resolve more discovery disputes centered on artificial intelligence, a trend toward relying on existing discovery statutes and rules is taking shape.  It is important, then, to emphasize that the threshold inquiries of discovery remain the threshold.  For instance, rules of procedure governing discovery require that requested information be relevant to a party’s claim or defense and proportional to the needs of the case.  See, e.g., Fed. R. Civ. P. (“FRCP”) 26 (b)(1).  Similarly, rules of evidence require that evidence only be introduced if the evidence tends to make a fact more or less probable than it would be without that evidence.  See, e.g., Fed. R. Evid. 401.  As discussed below, the proliferation of AI has not changed these threshold inquiries: rules of procedure and evidence have handled other technological advances and can do the same with AI. 

This article acts as an introductory roadmap, pointing to a variety of paths for subsequent discussion on the relevance and proportionality of AI-related discovery materials in ongoing AI-oriented lawsuits. 

Relevance is—Still—the Threshold Inquiry 

As AI-based discovery case law develops, it is useful to note that this is not the first time existing procedural and evidentiary guidelines have had to account for rapidly expanding and relevant technology.  Consequently, there does not need to be a complete shakeup of discovery norms and guidelines in the face of AI.  Courts had to similarly scramble to address discovery and internet usage, which, even now, is a relationship that continues to be defined by state court decisions applying existing rules of procedure and evidence.

Recent AI discovery scenarios focus more on work product and attorney-client privilege, not necessarily relevance.  (For a discussion of those decisions, see our recent article, Are AI Interactions Discoverable? Courts Continue to Draw the Lines).  But at least one decision explicitly refers to the threshold inquiry of all evidence in a decision that included discovery of AI-related materials: whether those materials are relevant to the matter at hand.  Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 820 F. Supp. 3d 629, 636 (E.D. Mich. 2026) (“Moreover, [all documents and information concerning use of third-party AI tools] is not relevant, or, even if marginally relevant, is not proportional.  See Fed. R. Civ. P. 26(b)(1).”).  The idea that AI-related discovery should not become “untethered from Rule 26 relevance,” id. at 637, will be increasingly important as courts continue to detangle variations on the AI-discovery theme.  It is a concept that appears to have been implied by numerous decisions, but it is worth stating specifically. 

Relevance When AI is Central to the Lawsuit

To emphasize the relevance of relevance, discovery disputes focusing on internet usage and other electronically stored information may serve as a helpful precedent to different scenarios of AI-based discovery.  This is because internet searches and results, while not exactly the same, provide analogous privacy concerns and function similarly to AI inputs and outputs in which a user provides a prompt and receives a result.  Decisions involving the internet also show ways in which courts have applied longstanding rules to a technology that authors of those rules probably did not foresee. 

For instance, beyond disputes centered on work product and other privilege concerns, what if AI usage is central to the lawsuit?  Turning to internet usage as helpful precedent, courts have found web searches and browser history to be discoverable in civil cases when the internet activity is directly at issue—such as when unauthorized computer use was a reason for employment termination and central to the lawsuit (Helget v. City of Hays, No. 13-2228-KHV-KGG, 2014 WL 1308893 (D. Kan. Mar. 31, 2014)), or when allegations under the Computer Fraud and Abuse Act provided notice that electronic evidence residing on computers would be relevant to the litigation.  (Nacco Materials Handling Grp., Inc. v. Lilly Co., 278 F.R.D. 395 (W.D. Tenn. 2011)).  In other words, when internet-related activities play a central role in a lawsuit, that is a good indicator that much of that internet-related activity will be discoverable.  

This central-to-the-lawsuit scenario is being played out right now in real time as several blockbuster lawsuits in the copyright space are based wholly on AI usage.  In these cases, discovery disputes necessarily involve disagreements about and regarding AI, and relevance is a key topic.  

For example, the ongoing consolidated In re OpenAI, Inc. litigation, which includes authors like George R.R. Martin and news agencies like the New York Times, is in its discovery phase. This litigation is based on a number of lawsuits alleging that OpenAI and other AI companies copied authors’ works without a license in order to train ChatGPT models.  In January of this year, the court overseeing this litigation ordered OpenAI to produce 20 million de-identified ChatGPT logs.  The Court reasoned that such a decision sufficiently “balanced ChatGPT users’ privacy interests against the relevance of the documents in light of the privacy protections already in place.”  In re Openai, Inc., No. 25-md-3143, 2026 WL 21676, at *2 (S.D.N.Y. Jan. 5, 2026).  Decisions like this one reiterate the point that courts are not applying a separate set of rules to AI.

Another example of central-to-the-lawsuit AI outside the IP space is found in defamation cases where AI-based chatbots or deepfakes have been allegedly used to tarnish the reputation of certain individuals.  At this point, pending defamation cases have not resulted in publicly litigated or published discovery disputes, but monitoring lawsuits like LTL LED, LLC v. Google, No. 25-cv-02394 (D. Minn., filed June 2025 and alleging that a Google “AI overview” informed the public that a business was being sued by a state attorney general), and Starbuck v. Google, No. N25C-10-211 (Del. Super. Ct., filed October 2025 and alleging that Google’s large language models generated false accusations including claims of sexual abuse), should eventually result in the respective courts having to weigh the relevance of requested evidence during discovery. 

Turning to the workplace, a number of companies are implementing AI-usage guidelines, and just like improper use of the internet at work can get an employee in trouble, improper use of AI tools could do the same.  Furthermore, state and Federal legislatures continue to pass or contemplate laws regulating AI and its usage, with new legislation rolling out on a consistent basis.  With these trends, inappropriate AI usage in the workplace and violations of state or nationwide AI statutes may present more central-to-the-lawsuit AI scenarios that will require the violating use of AI to be scrutinized in court, mimicking Helget and Nacco Materials above, and placing the relevance of AI-based or adjacent material squarely into discovery dispute consideration.

Relevance When AI is Not Central to the Lawsuit

On the other hand, what about situations where AI is not necessarily the main thrust of a lawsuit? Turning back to internet usage, courts have found internet activity can still be relevant when it is not the central focus of the case.  This can be in instances when browser or search history will help prove or disprove an alleged state of mind, subjective understanding, or knowledge about facts relevant to the case.  This is apparent in criminal matters, where courts have held that browser history is relevant, and admissible, as evidence of or against a defendant’s state of mind or motive.  One example is when threats or actions in cyberspace mirror counterpart threats or actions in the real world.  (United States v. Segui, No. 19-CR-188(KAM), 2019 WL 8587291 (E.D.N.Y. Dec. 2, 2019)).  Additionally, in physical and sexual abuse cases, internet history can also be discoverable and may serve as evidence that supports motivation for committing particular criminal acts. (State v. McGrath, 501 P.3d 346 (2021)).  

One particular AI case appears to have been confronted with a situation adjacent to Segui in which actions suggested by ChatGPT mirrored actions taken in real life. In Fortis Advisors v. Krafton, No. 2025-0805-LWW, (Del. Ct. Ch. Mar. 16, 2026), a CEO requested and followed legal advice from ChatGPT to avoid making an earnout payment to employees of an acquired company.  ChatGPT provided step-by-step instructions, including recommendations to terminate certain employees and then stage a company takeover, suggesting formation of an internal task force, name for the project, and providing detailed talking points, letters, and other communications to be used at each stage of this takeover.  After following ChatGPT’s advice to avoid payments, the CEO was eventually sued for breach of contract, among other claims. The terminated employees obtained discovery of the CEO’s exchanges with ChatGPT and introduced them in evidence.  The exchanges with ChatGPT became critical evidence that the court considered and relied on to conclude that the motivation for termination of the employees was purely financial and not for cause.  Although not the subject of the lawsuit, AI evidence was relevant to support improper motivation and ultimately assisted in proving wrongdoing. 

In sum, whether internet searches or AI prompts, specific searches can, potentially, provide evidence for or against elements of civil or criminal cases—whether the subject of the dispute or not.  And AI used either to perpetrate a crime or plan for one, or to violate some right or contract, may be admissible evidence and relevant to helping prove or disprove causes of action.  

Potential Variations on the Theme of Relevance 

Looking beyond internet-specific examples, there are as many potential opportunities for AI materials to be admitted as critical evidence in a lawsuit as there are ways to use AI. Take for example the drafting of contracts.  The drafting process itself could be subject to discovery and eventually admitted at trial if drafters use AI to generate or revise portions of a contract.  The content of the prompts fed into AI and the various iterations of output may be relevant to show knowledge, intent, or lack of either, as occurred in Fortis Advisors above. 

The legal field adds extra wrinkles to a similar scenario as attorneys draft contracts all the time for high-profile clients that are sure to value privacy of negotiations as contracts progress from version to version.  Courts will have to determine if privilege applies to prompts-based contract revisions that have been handled by attorneys, considering if work product and attorney-client privilege will attach, either blocking discovery or significantly limiting it. So far, courts have been reluctant to extend attorney-client privilege to communications with public AI tools, but more willing to protect AI-assisted work product, considering if work product and attorney-client privilege will attach, either blocking discovery or significantly limiting it.

Additionally, and not only in the realm of attorney use, malpractice lawsuits based on improper/negligent use of AI, or use of AI when a client specifically forbids it, will place that AI use and resulting materials at the very center of that lawsuit to determine when, how, or why AI was used—and if it was used with competence or not.  

The intellectual property space also has more potential for relevance analyses. For instance, plugging proprietary source code or trade secrets straight into an AI portal could itself be discoverable evidence of an unauthorized disclosure or breach of company AI use policies.  Such AI-related actions may even be considered to determine whether a voluntary disclosure or waiver occurred.  Since the AI explosion, companies have tried to crack down on this type of AI use to avoid leaks of sensitive or protected information.  Even still, just one careless copy-and-paste may lead to severe IP-related consequences, and that interaction with AI may be discoverable and admitted into evidence to make, or break, a case. 

The bottom line: if there is some reason that AI materials or usage are indeed relevant to some aspect of a lawsuit, a party may be ordered to produce as much as all AI materials and information.  

Proportionality also Remains a Valid Consideration 

While relevance may take up the lion’s share of discoverability analysis, proportionality is a familiar and necessary consideration when requesting or producing AI materials in discovery.  In all discovery requests, courts, and parties, must ask whether the cost and effort of collecting and/or producing any potentially relevant AI-generated information or preserved AI usage is reasonable and proportional to the needs of the case.

The rules governing discovery indicate that, in today’s technological world, the sheer volume of stored electronic information and the multitude of devices that generate that information may make certain requests disproportionate to the needs of a case. For instance, the expenses and burden to both sides resulting from the preservation, collection, review, and production of AI-generated materials or interactions may completely overshadow any benefit of knowing the contents of that information—in theory. In practice, few litigants will turn down production of all information, preferring to sift through what they can and make their own decisions about what is and is not important.  And lawsuits like the consolidated In re OpenAI, Inc. show that production of 20 million de-identified ChatGPT logs may actually be very proportional in the context of the dispute. 

Nevertheless, as a strategy that may be employed by both sides in a discovery dispute, litigants may try to emphasize the heavy burden of producing that much information, focusing on the volume of AI data, and any technical limitations to preservation, collection, and/or production of that data.  This strategy may not get a request for AI materials completely denied, but the court may see fit to narrow the requests only to what is relevant to the pending claims.  For example, discovery requests for any and all information regarding AI usage in designing marketing visuals will probably be disproportionate to the needs of a case focusing on an alleged breach of contract—unless the breach had something to do with a clause on marketing materials.  In other words, discovery requests must fit the needs of the case—as they always have been required to do.  Discovery requests focusing on AI are no different. 

Conclusion

AI has revolutionized the way the world operates, but in the discovery phase of litigation, AI has provided different angles through old lenses.  What is and is not discoverable material during litigation does not need newfangled discovery procedures or considerations.  Longstanding rules governing relevance and proportionality still apply and have been utilized in a growing number of AI discovery disputes.  

The discussion above is just a snapshot of current AI proceedings, and Womble attorneys look forward to monitoring and analyzing numerous aspects of these cases as they weave their way through state and federal court.  Should you have any questions, comments, or concerns about implementation of AI into your organization, or are currently involved in litigation where AI-generated materials are being requested, Womble attorneys have extensive experience in the AI and e-discovery spaces and are uniquely positioned to guide clients through these complex developments. Please reach out to the authors of this alert or any member of Womble's AI team or Digital Solutions team.