AI Expert Witnesses and Court-Ordered Disclosure: What Litigators Need to Know

A federal magistrate judge in Connecticut has done something that no federal court appears to have done before: ordered an expert witness to hand over the prompts she used when querying an AI tool in the course of preparing her expert report. The May 2026 decision in Conservation Law Foundation, Inc. v. Shell Oil Co. is the first federal court ruling requiring production of expert AI prompts as part of discoverable methodology, and it signals a major shift in how expert witness work product will be scrutinized in litigation.

This is not an isolated development. Courts are increasingly alert to the fact that AI tools are reshaping how experts do their work — from literature review and document analysis to quantitative modeling and opinion drafting — and that the current disclosure rules were not designed with AI in mind. The result is an emerging body of court-ordered practice that litigators on both sides need to understand before they retain an expert, prepare an expert report, or take an expert’s deposition.

The Foundation: What Rule 26 Requires

Federal Rule of Civil Procedure 26(a)(2) requires parties to disclose expert witnesses who will provide opinion testimony, and to produce a written report signed by the expert. The report must contain:

• A complete statement of all opinions the expert will express and the basis and reasons for each;
• The facts or data considered by the expert in forming those opinions;
• Any exhibits to be used to summarize or support the opinions;
• The expert’s qualifications;
• A list of all other cases in which the expert has testified as an expert within the preceding four years; and
• A statement of the compensation the expert is to be paid.

The “facts or data considered” element has historically been one of the most litigated aspects of expert discovery. Rule 26(b)(4) provides a qualified protection for draft expert reports and certain communications between retaining counsel and the expert, but facts or data provided to the expert by counsel that the expert considers in forming her opinions must be disclosed.

The Conservation Law Foundation decision creates a new question: when an expert uses AI tools to analyze data, filter documents, or generate analytical frameworks, do the AI prompts — the specific queries the expert entered — constitute part of the “facts or data considered” that must be disclosed? And if not “facts or data,” are they part of the expert’s “methodology” that is independently subject to discovery?

Conservation Law Foundation v. Shell Oil Co.: The First AI Prompts Ruling

Background

Conservation Law Foundation, Inc. v. Shell Oil Co. is an environmental enforcement action pending in the United States District Court for the District of Connecticut (Case No. 3:21-cv-00933). Conservation Law Foundation (CLF) alleges that Shell and other operators of a bulk petroleum storage terminal in New Haven, Connecticut failed to account for climate-change-related risks — including sea-level rise, storm surge, and severe weather — in the design and operation of the facility.

CLF retained Dr. Naomi Oreskes, a prominent historian of science with expertise in climate change research, as an expert witness. Dr. Oreskes’s report addressed the history of climate change science and the state of scientific knowledge about climate-related risks at the terminal.

How AI Was Used

In preparing her expert report, Dr. Oreskes and her research assistant, Dr. Kaurov, used a commercially available AI tool to process Shell’s document production. Specifically, they used OpenAI’s GPT-4o, accessed through a private Microsoft Azure server, to upload the defendants’ produced documents and filter them with AI prompts into a relevant subset. This AI-assisted filtering was a data management step: the expert used AI to identify which documents from the defendants’ production were worth closer examination.

The AI prompts were used to narrow a large document collection to a manageable set — not, apparently, to generate opinions or conclusions. The prompts themselves were instructions to the AI about what to look for and how to categorize or select documents.

The Magistrate Judge’s Ruling

On May 18, 2026, Magistrate Judge Thomas O. Farrish ordered CLF to produce the AI prompts that Dr. Oreskes and Dr. Kaurov used in this process.

The court’s reasoning had two components.

First, expert methodology is fair game for discovery. The court found that Rule 26 and the discovery rules generally allow parties to explore how an expert reached her opinions — not just what those opinions are. The process by which Dr. Oreskes used AI to narrow the defendants’ document production into a workable subset was, in the magistrate judge’s view, an aspect of her methodology. If you can depose an expert about which search terms she used in a Westlaw search, or which databases she queried, the analogous inquiry for AI use is what prompts she used.

Second, the court rejected CLF’s argument that the prompts were protected as “notes” under a Rule 29 discovery agreement the parties had entered. The court interpreted the agreement narrowly, finding that it “must be quite clear” for the agreement to protect material as notes, and that the AI prompts did not meet that standard.

Current Status: Stay and Pending Objection

On June 3, 2026, CLF filed a Rule 72(a) objection with the district court, arguing that the magistrate judge’s ruling was clearly erroneous on multiple grounds. The district court has issued a stay of the production order pending resolution of the objection. As of this writing, the district court has not ruled on the objection, and the issue of whether the AI prompts must ultimately be produced remains open.

The stay means the ruling does not yet have the finality of settled precedent, but the magistrate judge’s analysis has already been widely cited and discussed, and its logic is available for courts in other cases to adopt or adapt.

Why the Ruling Matters Beyond This Case

The Conservation Law Foundation decision has implications well beyond climate change litigation and well beyond document filtering. The underlying principle — that AI-assisted steps in an expert’s methodology are part of that methodology and therefore discoverable — applies across practice areas and industries.

The Scope of AI Use in Expert Work

Experts across litigation contexts are using AI in ways that range from document filtering (as in Conservation Law Foundation) to far more substantive analytical functions:

Quantitative analysis. Economic and financial experts use AI tools to model damages, analyze market behavior, and process large datasets. The prompts and parameters used to configure AI analysis of financial data are methodologically significant.

Medical and scientific literature review. Medical experts routinely face enormous volumes of published literature. AI tools that filter or summarize literature for expert review select what information the expert is exposed to. If an AI system systematically excluded relevant literature, the expert’s opinion may rest on an incomplete picture — and opposing counsel is entitled to explore that.

Document analysis in complex commercial cases. Experts in intellectual property, securities, and commercial contract disputes routinely review large document productions. If AI is filtering which documents an expert examines, the prompts that control that filtering are methodologically relevant.

Opinion drafting assistance. Perhaps the most sensitive use case: some experts may use AI to generate initial drafts of their opinions, which the expert then reviews, edits, and adopts. The extent to which AI-generated reasoning shapes the final opinion is directly relevant to whether the opinion represents the expert’s independent judgment — a Daubert requirement.

Daubert and the Reliability of AI-Assisted Methodology

Under Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), trial courts act as gatekeepers for expert testimony, admitting only opinions that rest on sufficient facts, a reliable methodology, and a reliable application of that methodology to the facts. AI use in expert analysis raises reliability questions that existing Daubert precedents do not squarely address.

The reliability inquiry focuses on whether the methodology is scientifically valid and has been tested, whether it has been subject to peer review, whether there is a known or potential error rate, and whether it is generally accepted in the relevant scientific community. AI-assisted methodology has at most a nascent literature on reliability, error rates vary significantly by tool and application, and peer review of specific AI-assisted analytical approaches is largely undeveloped.

The Conservation Law Foundation ruling creates the discovery mechanism that will allow these Daubert questions to be pursued. If AI prompts are discoverable, opposing counsel can examine: what exactly did the AI do; what was it asked to find or exclude; what were the AI tool’s known limitations; and did those limitations systematically bias the document subset or analytical output the expert worked with? Those questions feed directly into a Daubert reliability challenge.

Strategic Implications for Litigators

For Retaining Experts: Disclose Early and Document Well

The clearest lesson from Conservation Law Foundation for attorneys who retain experts is that AI use must be disclosed early and documented thoroughly. Waiting until deposition for AI use to surface — or worse, having it surface through opposing counsel’s discovery motion — is far worse than proactive disclosure.

If your expert intends to use AI tools, discuss it before the report is written. Understand exactly what tasks the AI will perform, what tools will be used, and how the AI output will be reviewed and validated. Document the AI use in the expert’s files, and consider including a disclosure of AI use in the expert report itself.

Early disclosure also allows you to frame the AI use in its most favorable light. If the AI was used for a limited, clerical task — like filtering a large document collection to identify relevant documents for closer review — presenting that clearly in the report, along with the validation steps taken, prevents opposing counsel from depicting it as something more significant.

For Opposing Experts: New Discovery Tools Are Available

For counsel challenging an opposing expert, Conservation Law Foundation provides a new discovery vehicle: a targeted request for production of AI prompts, AI interaction logs, and any records of how the AI tools were configured and used. These requests can be tailored to the specific tasks the expert performed and the specific AI tools used.

Deposition questions should probe:

• What AI tools did the expert use?
• At what stages of analysis?
• What prompts were entered?
• Did the expert review the AI’s selection or filtering before accepting it?
• What steps were taken to validate AI outputs?
• Is the expert aware of the AI tool’s known limitations, including hallucination rates and training data gaps?

The answers to these questions may reveal that the expert’s analysis rested on AI-generated inputs that were not independently validated, or that the AI’s filtering systematically excluded documents that would have been relevant to the analysis. Either finding can support a Daubert motion or an effective cross-examination at trial.

Witness Preparation: New Conversation Areas

Expert witnesses now need to be prepared to answer questions about their AI use in deposition and at trial. Many experts have adopted AI tools quickly, without thinking carefully about the discovery implications. Preparing an expert for deposition must now include a conversation about every AI tool used in the matter, the specific tasks those tools performed, and what the expert can and cannot say about how those tools work and what their limitations are.

An expert who cannot explain what AI did in her analysis, or who adopted AI-generated filtering without reviewing it, is a weak witness on cross-examination. Preparation conversations should identify gaps in the expert’s AI knowledge and either fill them (by helping the expert understand what the tools did) or address them explicitly (by documenting validation steps that compensate for limited expert knowledge of AI internals).

How Courts Are Approaching AI in Expert Testimony More Broadly

Beyond the Conservation Law Foundation decision, courts have begun paying attention to AI in expert testimony through several additional channels.

Disclosure in report requirements. Some courts are beginning to require, through standing orders or case management orders, that expert reports identify any AI tools used in preparing the report or the underlying analysis. This is a narrower version of the general AI disclosure trend in attorney filings, applied specifically to expert testimony.

Court-appointed technical advisors. In complex AI-related cases — patent disputes involving AI, products liability cases involving AI systems, antitrust cases involving algorithmic pricing — courts are increasingly relying on court-appointed technical advisors to help evaluate the validity of AI-related expert testimony. These advisors assess whether the AI methodology the expert used was appropriate and reliable.

Pending FRCP amendments. The Judicial Conference Advisory Committee on Civil Rules has been examining whether amendments to Rule 26 are needed to expressly address AI use in expert analysis. Proposed amendments that would require disclosure of AI tools used in expert analysis, or that would address the discoverability of AI prompts and logs, are under discussion. Any amendment would go through the standard notice-and-comment process and would not be effective for at least two to three years, but the Committee’s attention to the issue signals that the current lack of a specific rule will not persist indefinitely.

Practical Recommendations

Based on the Conservation Law Foundation decision and the broader trajectory of courts’ approach to AI in expert work, the following practices represent current best practice for litigators in 2026.

Create an AI use log for every expert engagement. Before the expert’s work begins, establish a practice of documenting AI tool use: what tools, what tasks, what prompts (or at minimum, what categories of prompts), and what validation steps were taken. This log is potentially discoverable, so it should be created with the awareness that it may be reviewed, but its existence also demonstrates responsible AI use.

Include AI use in the expert report. Voluntarily disclosing AI use in the expert report — in the methodology section, with a brief description of what the AI did and how its outputs were validated — is increasingly the standard of care. Courts and opposing counsel will notice its absence; its presence frames the issue favorably.

Conduct Daubert analysis of your own expert’s AI use. Before the report is served, critically examine whether the AI methodology used by your expert would survive a reliability challenge. If the answer is uncertain, either redesign the methodology or prepare a strong defense of it.

Tailor expert retention agreements. Consider including provisions in expert engagement agreements that specify requirements for AI use: that AI tools must be disclosed to retaining counsel, that AI outputs must be independently validated before being incorporated into the analysis, and that AI interaction logs must be preserved.

Conclusion

The Conservation Law Foundation ruling marks the beginning of a new era in expert witness practice. AI-assisted methodology is discoverable under existing rules, and courts are beginning to require experts to account for every step of their analytical process — including the AI-assisted steps. For litigators, this means rethinking expert retention strategy, witness preparation, and the disclosure standards that govern expert reports.

The underlying principle — that juries and courts are entitled to scrutinize the full methodology behind an expert opinion, including any AI-assisted steps — is a straightforward application of Daubert’s reliability requirement to a new technology. The practitioners who adapt their approach now, before a court orders production of their expert’s AI prompts in a high-stakes matter, will be better positioned than those who treat AI use as a private implementation detail.

This post is for general informational purposes only and does not constitute legal advice. Reading this post does not create an attorney-client relationship. If you have questions about your specific situation, consult a qualified attorney.