515.218.7888 (Iowa) 469.200.4077 (Texas)
Case Briefs

Conservation Law Foundation, Inc. v. Shell Oil Co.

No. 3:21-cv-00933 (D. Conn. Nov. 20, 2026) · No. 3:21-cv-00933 (VDO) · Decided November 20, 2026 · 2026 WL 2664341 (D. Conn. May 18, 2026) (Farrish, M.J.), affirmed

Presented by John Goodhue

Conservation Law Foundation, Inc. v. Shell Oil Co. — video case brief

Are the prompts an expert witness feeds to a generative-AI tool discoverable? In Conservation Law Foundation, Inc. v. Shell Oil Co., No. 3:21-cv-00933 (D. Conn. Nov. 20, 2026), Judge Oliver overruled the plaintiff's objections and affirmed a magistrate judge's order compelling production of the AI prompts and queries the plaintiff's expert and her research assistant used with ChatGPT to cull the defendants' document production. Reviewing under Rule 72(a), the court held that the prompts comprise the basis for the expert's opinions under Federal Rule of Civil Procedure 26(a)(2)(B) and fall within the scope of discovery under Rule 26(b), that a stipulation shielding drafts, communications, and notes did not reach them, and that sworn testimony the prompts once existed warranted an order of production.

Read the opinion (PDF)

Transcript

Conservation Law Foundation v. Shell Oil Company, decided November 20th, 2026. Judge Oliver, in the District of Connecticut, held that the artificial intelligence prompts and queries an expert witness and her team used in producing her report are discoverable. The court overruled the plaintiff's objections and affirmed, in all respects, a magistrate judge's order compelling their production. Here's the brief.

Conservation Law Foundation, a nonprofit, brought a citizen enforcement action under the Clean Water Act and the Resource Conservation and Recovery Act over a bulk storage and fuel terminal in New Haven, Connecticut. The remaining defendants are three companies that include Equilon Enterprises, doing business as Shell Oil Products US.

The plaintiff retained Dr. Naomi Oreskes, a Harvard professor of the history of science, as an expert. To review the documents produced in the litigation, she worked with a computational scientist, Dr. Alexander Kaurov, as a research assistant. He processed the documents in a dedicated cloud environment and used ChatGPT to search them and to identify documents for Dr. Oreskes to review. At her deposition, Dr. Oreskes testified that the team used ChatGPT on a secure server to identify the documents. Dr. Kaurov did not preserve complete logs of the prompts and outputs.

After the expert's deposition, the defendants moved to compel production of the AI reliance materials. The parties narrowed the dispute to the prompts and outputs. On May 18th, 2026, Magistrate Judge Farrish granted the motion. The plaintiff objected, and the district judge stayed the order and heard argument.

The question was whether that order was clearly erroneous or contrary to law, the standard under Rule 72(a) for a magistrate judge's ruling on a nondispositive matter. A district judge may not modify such a ruling simply because he would have decided the matter differently.

Two parts of Rule 26 framed the merits. Rule 26(b) sets the general scope of discovery. And Rule 26(a)(2)(B) requires a retained expert's report to state the expert's opinions, the basis and reasons for them, and the facts or data the expert considered.

The court first rejected three procedural objections. The discovery requests had been sent by email, but the plaintiff had answered them without objecting to the manner of service, so that objection was waived. Any lateness in the motion was excused by the defendants' diligence: they sent a deficiency letter ten days after the deposition and conferred with the plaintiff at least seven times before filing. And the parties' Rule 29 stipulation, which shielded draft reports, expert communications, and notes, did not reach the prompts. Drawing on ChatGPT's own description, the court explained that a prompt is how a user tells the tool what the user wants to know, make, or change, and it found no clear error in distinguishing prompts from an expert's notes.

On the merits, the court held that the result was the same under either part of Rule 26. Under Rule 26(b), the process by which the expert culled the document production down to a subset was relevant, nonprivileged, and proportional.

Under Rule 26(a)(2)(B), the court concluded that the prompts and queries comprise the basis for the expert's opinions. Dr. Oreskes had testified that she used AI as a sieve to identify the documents relevant to a question, including every document that mentioned sea level rise, and that she and her assistant decided together what the prompts would be.

Finally, the plaintiff argued, on counsel's representation, that the materials sought did not exist. The court acknowledged that a motion to compel is ordinarily denied when a party avers in good faith that the materials do not exist. But here there was sworn testimony that prompts had been used. The court lifted its stay and directed the plaintiff to comply by December 18th, 2026.

Conservation Law Foundation v. Shell Oil treats an expert's use of generative AI as part of the expert's method. Where prompts determine which documents an expert reviews, the court regarded them as the basis for the resulting opinions under Rule 26(a)(2)(B), and as relevant matter under Rule 26(b). A stipulation protecting drafts, communications, and notes did not cover them. The decision is a district judge's review of a discovery order under a deferential standard.

Conservation Law Foundation, Inc. v. Shell Oil Co., number 3:21-cv-00933, District of Connecticut, decided November 20th, 2026. I'm John Goodhue. Thanks for watching.

These videos are educational case briefs, not legal advice, and watching them does not create an attorney-client relationship with the presenter or the firm. Case law and its interpretation evolves, always check a decision's subsequent history. Do not rely on these case briefs, but read the case yourself or have your attorney read them. Videos are presented via an AI avatar and voice clone of John Goodhue, created with his participation and consent.