TL;DR: In Conservation Law Foundation, Inc. v. Shell Oil Co., a District of Connecticut magistrate judge ordered the plaintiff to produce the AI prompts used by its expert in preparing a report, holding that the expert’s AI-driven methodology is discoverable under Rule 26. The May 18, 2026 ruling (ECF 970) is the first federal decision to compel production of AI prompts. The order is currently stayed pending CLF’s Rule 72(a) objection. For trial teams, the decision signals that inputs to machine-assisted analysis can become fair game in discovery, with immediate implications for expert retention, document review workflows, preservation, and cross‑examination.
Background
Conservation Law Foundation (CLF) sued Shell Oil Co. and related entities in the District of Connecticut over environmental compliance and climate-risk issues under the Clean Water Act. CLF’s expert, Dr. Naomi Oreskes, used an AI workflow (GPT-4o accessed through Microsoft Azure) to filter Shell’s discovery production and identify a relevant document subset for her analysis. The defendants sought production of the prompts and inputs that drove that AI-assisted process. On May 18, 2026, Magistrate Judge Thomas O. Farrish granted the motion to compel, concluding that the expert’s AI prompts are part of the methodological underpinnings of the opinion and therefore subject to discovery under Rule 26. The ruling emphasized that if the same prompts shaped which documents were reviewed, they effectively influenced the evidentiary foundation of the expert’s conclusions. The order was entered as an ECF document (reported as 970 in some dockets) and has been the subject of subsequent commentary and analysis. The ruling’s practical takeaway is that prompts used to filter or shape AI-driven work can be disclosed as part of an expert’s methodology, not merely as privileged materials. (dechert.com)
The Ruling and Its Significance
The decision marks the first known federal court ruling that compels an expert to disclose AI prompts used in preparing an expert report. The court treated the AI prompts as part of the expert’s methodology rather than as privileged attorney work product or notes, finding that the solicitation and refinement of AI-driven outputs may be discoverable material under Rule 26(b)(1). The court noted that prompts can shape the universe of documents the expert considers and thus influence the reliability of the resulting opinion, tying the issue directly to Daubert/Rule 702 considerations. Practically, this creates an audit trail for AI-assisted conclusions, enabling opposing counsel to assess how AI influenced a litigant’s evidentiary groundwork. The ruling also forewarns of broader discovery demands for prompt histories, chat logs, and other AI artifacts, potentially expanding the scope of materials that must be preserved and produced in future cases. (dechert.com)
Current procedural posture is that the ruling is stayed while CLF pursues a Rule 72(a) objection to the district court’s ruling. In other words, the immediate effect is restrained pending appellate-type review on the stay motion, but the underlying reasoning already signals a paradigm shift in how AI-assisted methods are treated in discovery. For practicing litigators, the takeaway is clear: anticipate inquiries into the inputs, prompts, and configurations that drive AI-supported work, even when those inputs were not part of the traditional paper trail. (dechert.com)
Practical Implications for Trial Teams
- Expert engagement and disclosure: When an expert relies on AI tools, the prompts and prompts’ history may be discoverable. Counsel should prompt experts to document, at the outset, which AI tools were used, for what tasks, and how prompts were crafted. The CLF decision explicitly links the reliability and defensibility of AI-assisted conclusions to the transparency of the prompts and the process that produced the AI outputs. (dechert.com)
- Preservation and data-management discipline: The ruling highlights the need to preserve AI prompts, search terms, iterations, and outputs as part of the expert’s workflow. Teams should implement prompt-preservation protocols in their standard operating procedures, with clear instructions to the expert and support staff to capture and retain prompts and related logs. Courts will likely treat prompt histories as part of the evidentiary foundation, not as peripheral drafts. (spencerfane.com)
- Distinguishing prompt use: The court’s analysis drew a line between prompts used for document culling and prompts used for substantive analysis. This distinction matters for privilege and work-product considerations, and it underscores the importance of precise disclosures about the role of AI in forming opinions. Counsel should tailor disclosures to reflect where AI influenced the analysis versus where it merely organized the evidence. (spencerfane.com)
- Cross-examination and reliability testing: With the prompts exposed, opposing counsel gains a window into the AI-supported methodology, enabling more rigorous Daubert-style challenges to reliability, reproducibility, and the potential for model drift or prompt-tuning to affect results. This heightens the importance of pre-trial mock examinations and focused questions about AI workflows, data curation, and results. (spencerfane.com)
- Risk management for counsel: The Prompts-as-Methodology framework invites new guardrails to prevent inadvertent disclosure of strategy embedded in prompts. The court’s footing in promoting transparency around prompts can push teams to separate purely strategic notes from methodological prompts, and to shield truly privileged information through carefully drafted privilege logs and containment measures. The same guidance warns that prompt content may reveal strategic thinking if not carefully managed. (spencerfane.com)
Actionable Steps for Counsel
- Create and implement AI-prompt documentation: Require every expert engaged in AI-assisted analysis to maintain a prompt log, including purpose, prompts used, iterations, model version, and rationale for each step.
- Update discovery responses to reflect AI workflows: If AI inputs influenced the analysis, disclose them with specificity in Rule 26 disclosures or expert reports, and be prepared for follow-up requests for prompts, logs, and related materials.
- Separate prompts from attorney work product where necessary: Draft clear language in engagement letters and protective orders to distinguish prompts created by the expert, prompts created by vendors, and any attorney-directed prompts that may be more sensitive.
- Prepare for prompt-related questions at deposition: Develop a set of grounded, non-revealing answers that explain how AI contributed to the analysis without exposing trial strategy or confidential methodologies beyond what is necessary for fairness and accuracy.
- Use training to stay ahead: Objection Academy offers scenario-based drills and evidence-focused training that help trial teams practice objections and foundational challenges around AI-generated evidence and expert methodology. For ongoing readiness, see Objection Academy’s materials on FRE 702-style reliability, AI-evidence considerations, and practical objection strategies in 2026. (objectionacademy.com)
Objection Academy in Practice
As AI becomes a routine element of litigation workflows, training in how to handle AI-generated evidence and prompt-driven methodologies becomes essential. Objection Academy provides a structured platform for drills and simulations that align with the practical realities of AI-assisted expert work and the evolving evidentiary gatekeeping around machine-generated outputs. For trial teams, integrating such practice into pre-trial preparation helps sharpen objections, foundation questions, and cross-examination tactics in light of AI-driven analyses and the growing likelihood that AI prompts themselves may surface in discovery. See Objection Academy’s resources and related analysis on AI-evidence themes and FRE 707 discussions for context on how practitioners are approaching these developments. (objectionacademy.com)
What to Watch Next
The Shell CLF decision is part of a broader, steadily evolving landscape in which courts are testing the boundaries of AI in litigation. In the near term, expect:
- More district courts to consider whether AI prompts and AI-derived outputs constitute discoverable methodology or protected work product, potentially leading to varied standing on these issues across jurisdictions.
- Further guidance from the Advisory Committee on Evidence Rules and state bar analyses as practitioners seek clear rules around AI-generated evidence, reliability testing, and disclosure obligations.
- Ongoing coverage of AI-evidence developments and their practical impact on trial strategy, which can be anticipated to intersect with evolving rules on AI-generated outputs, deepfake content, and the admissibility of machine-assisted conclusions at trial. For litigators, keeping a close eye on these developments and maintaining robust AI-use documentation will be essential.
Sources
- Magistrate Judge Farrish, Conservation Law Foundation v. Shell Oil Co., No. 3:21-cv-00933 (D. Conn. May 18, 2026) (order compelling production of AI prompts) and contemporaneous coverage. (dechert.com)
- Court and law-firm commentary detailing the May 18 2026 order and its implications for discovery of AI prompts. (law360.com)
- Spencer Fane, Court Orders Disclosure of Expert Witness’s AI Prompts: What Litigators Need to Know, July 8, 2026. (spencerfane.com)
- Mayer Brown / JDSupra coverage summarizing the holding and its practical consequences, June 2026. (jdsupra.com)
- Objection Academy resources on AI-evidence themes and best practices for trial teams. (objectionacademy.com)