Jones v. Delta Air Lines, Inc., No. 2:24-cv-11224 (E.D. Mich. Apr. 22, 2026)

‍

Why This Case Is Important

This landmark ruling establishes that generative AI prompts used by testifying experts are discoverable methodology rather than protected work product. As legal teams increasingly integrate AI into evidentiary analysis, they must explicitly address AI workflows in discovery protocols.

Overview Text

In an environmental enforcement action pending in Connecticut, the plaintiff, Conservation Law Foundation, Inc. (CLF), alleged that Shell Oil Co. failed to adequately protect a critical terminal from climate change risks. To prepare her testimony, CLF’s expert witness, historian Dr. Naomi Oreskes, utilized GenAI tools to review Shell’s voluminous document production. The expert and her research team used AI to sift through millions of records and distill the data into a workable, narrower subset for closer analysis.

Following months of meet-and-confer sparring, the defendants moved to compel the production of the AI prompts, queries, and inputs used by the expert team. Shell argued that these AI inputs represent standard Rule 26 discovery because they reflect data selection, filtering, and core methodological choices that bear directly on the reliability of the expert’s final report. Without reviewing the precise prompts, the defendants claimed they would be deprived of the ability to determine whether the AI tools were fed biased inputs, whether critical evidence was improperly excluded, and whether the generated outputs reliably supported the expert’s conclusions.

CLF resisted the motion on three primary grounds. First, they argued that AI prompts fall outside the scope of Rule 26(b) because they were used solely to cull documents and do not constitute "facts or data considered" by the expert. Second, CLF asserted that a prior Rule 29 discovery stipulation shielded the material, characterizing the AI queries as protected "expert notes or drafts" made during the report-drafting process. Finally, the plaintiff maintained that no additional responsive materials existed, claiming the expert had utilized ordinary "search terms"—which had already been produced—rather than distinct "prompts."

Ruling Summary

• AI Prompts Constitute Discoverable Expert Methodology Rejecting CLF's claim that AI prompts fall outside Rule 26(b), the court held that "[a]n expert witness's methodology is fair ground for discovery," and "the process by which [the expert] culled down the defendants' document production into a subset to be worked with is an aspect of that methodology." Because the interactive prompts directly shaped the filtered dataset upon which the expert relied, they are a fully discoverable component of the analytical framework.
• ‍Rule 29 Agreements Must Explicitly Address AI The court dismissed CLF's argument that a stipulation shielding "expert notes, drafts, or communications" covered the prompts. The ruling emphasized that "before a court denies otherwise-relevant discovery based on a Rule 29 agreement, that agreement 'must be quite clear.'" Much like the precise waiver-displacement language evaluated in FTC v. Amazon.com, Inc., labeling AI prompts as "notes" was an ambiguous categorization that failed to meet the rigorous standard required to override Rule 26 obligations.
• ‍Evidence Defeats Conversational Denials CLF contended that no additional "prompts" existed beyond previously disclosed "search terms." The court noted that a party’s good-faith representation usually resolves production disputes, but this principle gives way when an adversary provides an "evidence-backed reason" for doubt. Mirroring Muslims on Long Island, Inc. v. Town of Oyster Bay, where deposition contradictions destroyed previous non-use claims, an earlier declaration from Dr. Oreskes' own research assistant explicitly referencing the use of "prompt[s]" provided a solid reason to compel a thorough production search.

Expert Analysis

Hon. Andrew Peck (ret.), Senior Counsel, DLA Piper

While the case law on the discoverability of AI prompts is in its early stages, I think it is becoming clear that subject to proportionality concerns, client use of AI in the regular course of business is discoverable, and expert use also is discoverable. Attorney use is likely to be privileged or work product.  Any FRCP 29 agreement for a different result must be very clear (which it was not here).  Finally, once litigation is reasonably anticipated (as it clearly is when experts are involved), one must preserve AI prompts and outputs until the court resolves the matter of discoverability.

Tip Text

This decision highlights the critical need to mandate strict AI preservation protocols for retained experts and to explicitly address generative AI tools in initial Rule 29 and ESI protocols. To ensure your litigation strategy aligns with rapidly shifting judicial expectations surrounding electronic discovery, consider downloading Exterro’s Modern eDiscovery Checklist.