Court Issues Production Order for AI Search Prompts
Editor’s Note: As generative AI (GenAI) becomes more deeply embedded in discovery workflows, legal teams are confronting a new question: Can the prompts used to search, analyze, and review data become discoverable themselves? In this article, Phil Favro examines a first-of-its-kind court order requiring the production of AI prompts used by an expert witness, highlighting how courts are beginning to evaluate AI-assisted methodologies. While the decision is narrowly focused on expert discovery, it raises broader questions about relevance, proportionality, privilege, and work product protections in an AI-enabled legal environment. The ruling also examines the importance of documenting AI use, understanding how prompts fit within discovery obligations, and carefully structuring agreements that govern the exchange of AI-related information. As organizations increasingly rely on AI to accelerate legal and investigative workflows, governance and defensibility are becoming critical considerations alongside efficiency. Favro provides a timely analysis of a case that may shape future disputes over AI-generated work product and the discoverability of prompts themselves.
Court Issues Production Order for AI Search Prompts
By Phil Favro, Contributing Author for HaystackID
The eDiscovery landscape is navigating a paradigm shift. While parties have spent decades fighting over the use of traditional search terms—and later, technology-assisted review (TAR)—the rise of generative artificial intelligence (AI) has introduced a new frontier in ESI search disputes: whether AI prompts prepared and used to identify responsive documents should be produced to litigation adversaries.
A recent decision from a Connecticut federal court, Conservation Law Foundation, Inc. v. Shell Oil Company, addressed this issue, at least in the limited context of expert discovery. In Conservation Law Foundation, the court ordered a party to produce the specific prompts an expert witness developed and used in an AI application to identify source material to support her conclusions in an expert report. [1]
The order from Conservation Law Foundation is fact-specific and applies directly to issues involving expert discovery and party stipulations to limit discovery on specific topics. Nevertheless, beyond these topics, parties should familiarize themselves with Conservation Law Foundation. This decision appears to be the first which has touched tangentially on a fundamental issue regarding generative AI use in connection with the search, identification, and review process for discovery. Parties can be certain that Conservation Law Foundation will become a staple citation in negotiations and motion practice over the discovery of AI prompts, especially prompts that parties used to identify relevant information for production in discovery.
Conservation Law Foundation—Discovery of AI Application Prompts
In this longstanding litigation involving environmental claims, the defendants sought production of information on which an expert witness from the plaintiff apparently relied in connection with preparing her expert report. Among other things, the defendants sought “the prompts [the expert] used in conducting her AI analysis and outputs.” The plaintiff resisted discovery of its expert’s AI prompts on multiple grounds, each of which the court rejected.
Scope of Discovery
First, the plaintiff argued that the prompts were beyond the scope of discovery set by Federal Rule of Civil Procedure (Rule) 26(b). Under Rule 26(b)(1), parties may obtain discovery of nonprivileged information that is both relevant to the claims or defenses in the litigation and proportional to the needs of the case. Proportionality is determined by balancing the application of six different standards delineated under Rule 26(b)(1), which include (among others) factors such as the amount in controversy, the importance of the requested discovery, and the burden or expense of producing the sought-after information.
In addressing the plaintiff’s arguments, the court did not specifically weigh the relevance of the requested prompts or consider the proportionality standards. Instead, the court generally observed that an “expert witness’s methodology is fair ground for discovery” and that the process the plaintiff’s expert used to identify pertinent documents from the defendants’ production of documents is a discoverable part of that “methodology.”
Rule 29 Agreements
Next, the plaintiff asserted that the production of AI prompts was beyond the scope of a stipulation the parties reached governing expert discovery and that the court should honor the parties’ Rule 29 agreement. Under Rule 29, parties to litigation may enter into agreements to modify procedures “governing or limiting discovery.”
While Rule 29 does not limit the discovery procedures that parties may decide to modify, the court indicated that such agreements—to be enforceable—“must be quite clear.” In this instance, the Rule 29 agreement at issue forbade the parties from obtaining adversarial experts’ “notes, drafts, or communications” prepared or sent during the expert report “drafting process.” The plaintiff argued that this provision necessarily included its expert’s prompts; they were arguably “notes” and therefore outside the scope of discovery pursuant to the parties’ stipulation.
The court disagreed, finding that the disputed provision was not “quite clear.” While expressing its willingness to generally honor Rule 29 stipulations—“courts should enforce those agreements in appropriate cases”—the court ultimately felt that the instant agreement should not bar the defendants’ discovery request when the sought-after prompts were “otherwise within the scope of Rule 26(b).”
Prompts or Search Terms
Lastly, the plaintiff argued that their expert did not have any AI prompts to produce in discovery. Because the expert used search terms to identify information and considering that the plaintiff had already produced all of its search terms to the defendants, the plaintiff argued that a production order in this instance would be inapplicable.
In response, the court reasoned that it could not issue a production order to compel a producing party to turn over responsive information where the party does not have the sought-after information. Moreover, courts generally accept a party’s representation of no responsive documents unless faced with “solid evidence” that the requested information indeed exists. In this instance, the court found that such evidence existed:
In this case, the defendants have an evidence-backed reason for doubting [the plaintiff’s] representation, because [the expert’s] assistant, Dr. Alexander Kaurov, referenced “prompt[s]” in his declaration. (emphasis added)
Accordingly, the court overruled the plaintiff’s objections and ordered the production of the expert’s AI prompts to the defendants. In particular, the court ordered the plaintiff to update its responses to any interrogatories or document requests that sought the production of AI prompts that the expert used in connection with preparing her expert report. If the plaintiff continued to assert that it did not have responsive prompts from its expert, then it could so indicate in its amended interrogatory or document request responses. However, the court cautioned that it would consider imposing sanctions on the plaintiff under Rule 37(b) (for violating a discovery order) if the plaintiff represented that it did not have responsive AI prompts and “that representation is later revealed to be untrue.”
Takeaways from Conservation Law Foundation
Conservation Law Foundation is instructive on the preparation of Rule 29 agreements that restrict the scope of discovery in litigation. As Conservation Law Foundation makes clear, courts typically honor those agreements to the extent the provisions at issue are sufficiently clear and lend themselves to judicial enforcement. Whether in expert discovery or fact discovery, parties who wish to limit the production of AI prompts, outputs, or related information should (to the extent possible) specify that information in their Rule 29 stipulations. This could very well lessen the possibility that a court may sidestep the stipulation and order the production of prompts or outputs that are “otherwise within the scope of Rule 26(b).”
Nevertheless, Conservation Law Foundation is focused on a limited issue, i.e., the discovery of AI prompts that a party’s expert apparently used to identify documents on which she could rely in developing her expert report. Given the context in which the parties’ dispute arose, Conservation Law Foundation did not have the occasion to consider specific notions of relevance or the application of proportionality standards. Nor did the court address arguments that the AI prompts were protected from disclosure by the attorney-client privilege or as work product. Under the circumstances, such a narrow holding should not be the basis for extrapolating “best” practices or developing sweeping pronouncements for handling issues regarding generative AI use in connection with the search, identification, and review process for discovery.
And yet, because Conservation Law Foundation is apparently the first order in which a court has addressed the production of prompts used for identifying particular documents, parties should become familiar with this decision, as it could be cited to support an argument demanding the production of AI prompts and outputs. Like the recent order in United States v. Heppner, — F. Supp. 3d —, 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026)—which addressed the interplay between AI outputs and the attorney-client privilege and work product doctrine—expect Conservation Law Foundation to generate a fair degree of attention among commentators, given its role as the first decision touching on (albeit indirectly) the AI prompt production issue. What impact—if any—that the opinion has on actual discovery practices remains to be seen. However, what is reasonably certain is that a court that tackles this issue directly should anticipate vigorous arguments on relevance, proportionality, privilege, and work product—fundamental issues that do not appear to have been addressed in Conservation Law Foundation.
[1] Conservation Law Foundation, Inc. v. Shell Oil Company, No. 3:21-cv-00933-VDO (D. Conn. May 18, 2026), ECF No. 970 (ORDER granting Motion to Compel Production of Reliance Materials).
About Phil Favro
Phil Favro is the founder of Favro Law PLLC, where he counsels clients on ESI, AI, and discovery issues and serves as a special master, mediator, and expert witness. Phil is nationally recognized for his expertise on ESI, discovery, and information governance, with courts acknowledging his credentials. See, e.g., Oakley v. MSG Networks, Inc., No. 17-CV-6903 (RJS), 2025 WL 2061665 (S.D.N.Y. July 23, 2025). This background makes Phil particularly well-suited to counsel clients and advise courts on information-related issues. As a special master, Phil is acclaimed for his collaborative approach, working with parties to find stipulated solutions to complex issues. For disputes that require adjudication, he is renowned for the clarity and vigor of his written dispositions, which are available on legal search engines.
About HaystackID®
HaystackID® solves complex data challenges related to legal, compliance, regulatory, and cyber requirements. Core offerings include Global Advisory, Cybersecurity, Core Intelligence AI™, and ReviewRight® Global Managed Review, supported by its unified CoreFlex™ service interface and eDiscovery AI™ technology. Recognized globally by industry leaders, including Chambers, Gartner, IDC, and Legaltech News, HaystackID helps corporations and legal practices manage data gravity, where information demands action, and workflow gravity, where critical requirements demand coordinated expertise, delivering innovative solutions with a continual focus on security, privacy, and integrity. Learn more at HaystackID.com.
Assisted by GAI and LLM technologies.
SOURCE: HaystackID