A recent ruling protects an expert’s ChatGPT prompts and drafting materials while requiring disclosure of AI-generated case summaries—underscoring the importance of counsel-guided protocols for generative AI use.

A federal court in Nevada recently rejected an order requiring a testifying expert to produce his entire ChatGPT history, drawing a distinction between AI-generated research considered in forming his opinions and materials reflecting his mental impressions and trial preparation. In Villanueva v. Las Vegas Metropolitan Police Department, No. 2:24-cv-00125-ART-DJA (D. Nev. Sept. 29, 2026), ECF No. 112, the court required production of AI-generated case summaries but protected the expert’s prompts, deposition preparation materials, and draft report content. The decision provides guidance for companies and counsel incorporating generative AI into litigation: discoverability turns on the substance and purpose of the material, not simply the technology used to create it.

Background

The dispute alleged misconduct during the plaintiff’s detention. At his deposition, the plaintiff’s expert, Tom Melton, disclosed that he had used ChatGPT to research and draft his expert report. His use of AI had not previously been disclosed in response to a subpoena seeking the materials considered in forming his opinions. Defendants sought production of Melton’s ChatGPT history, which the plaintiff withheld as protected work product and trial preparation materials. The magistrate judge ordered production of the entire 40-page log, reasoning that Melton had considered it in forming and supporting his opinions and that the log was not itself a draft report. On review, the district court found that requiring wholesale production was clearly erroneous or contrary to law.

Research Outputs Are Discoverable; Mental Impressions and Drafts Are Protected

The district court applied Rule 26’s distinction between discoverable “facts or data” considered by a testifying expert and protected drafts and mental impressions. Significantly, the disclosure obligation encompasses factual materials an expert considered, not merely those the expert ultimately relied upon.

1. AI-generated case summaries must be disclosed. Melton used ChatGPT to summarize excessive-force decisions and consulted those summaries in reaching his opinions. The court held that the summaries constituted “facts or data” subject to disclosure under Rule 26(a)(2)(B)(ii). Producing the actual AI responses also would permit defendants to examine whether the expert had considered allegedly hallucinated citations, quotations, or legal standards. The decision did not establish that hallucinations had occurred; it recognized the opposing party’s ability to investigate that concern.
2. The expert’s prompts and questions were protected. The court distinguished research questions from the factual materials returned in response. Like “online research queries,” the questions were not themselves facts or data considered in reaching an opinion. They revealed the expert’s mental impressions, strategy, and views about which issues mattered to the case, bringing them within the protections for opinion work product.
3. Deposition preparation and draft report materials were also protected. Melton’s use of ChatGPT to anticipate deposition questions reflected preparation for testimony rather than factual support for his opinions. The resulting outline would reveal his priorities and areas of concern. Similarly, AI-assisted drafting concerning the opposing expert’s qualifications and report fell within Rule 26(b)(4)(B), which protects expert report drafts “regardless of the form in which the draft is recorded.” The court rejected the argument that ChatGPT-generated text could not qualify as a protected draft, noting that Melton sometimes incorporated the outputs into his report.

The court also concluded that Melton’s use of ChatGPT did not waive otherwise applicable protection because, in this instance, the application functioned as a research, drafting, and editing program. That conclusion should not be read as a categorical assurance that submitting privileged information to any AI platform is risk-free. The ruling addressed this expert’s use and the particular materials before the court—not every platform, confidentiality arrangement, or legal-team workflow.

The result in Villanueva contrasts with a case opinion from several months ago—Conservation Law Foundation, Inc. v. Shell Oil Co., No. 3:21-cv-00933 (D. Conn. May 18, 2026), ECF No. 970—where the court ordered disclosure of AI prompts used by an expert to cull a document production into a subset for an expert’s review, treating that process as part of her discoverable methodology. Unlike Villanueva, which protected prompts revealing the expert’s mental impressions and litigation strategy, Shell addressed document-selection methodology and rejected reliance on a discovery agreement that did not clearly encompass AI prompts; it did not expressly analyze the work-product protections applied in Villanueva. Together, the decisions caution against treating all expert AI prompts alike and underscore the importance of involving counsel early to distinguish methodological inputs from protected trial preparation and expressly address AI materials in discovery agreements.

Practical Implications for Legal Teams and Experts

Villanueva illustrates why neither blanket production nor blanket withholding is an adequate approach to an expert’s AI history. Companies should involve litigation and eDiscovery counsel before their legal teams and experts begin using generative AI on a matter. Counsel can help:

• Establish permitted uses and confidentiality safeguards. Evaluate platform terms, data retention, training practices, and access controls before confidential or privileged information is submitted.
• Address expert AI use early. Discuss research, analysis, drafting, and preparation workflows at the outset of the engagement, rather than first learning of them during a deposition.
• Preserve and organize the record. Retain relevant prompts and outputs consistent with preservation obligations, while distinguishing research materials from drafts and trial preparation content.
• Validate substantive outputs. Independently check AI-generated authorities, quotations, and factual assertions. Protection from discovery does not establish reliability; the court expressly noted that concerns about the expert’s methodology and ChatGPT reliance could be addressed through impeachment.
• Support specific protection claims. Review AI histories by content and purpose, and prepare privilege logs that explain the applicable basis for withholding particular materials.

The practical lesson is not to avoid generative AI, but to structure its use deliberately. Early consultation with counsel can help clients satisfy expert disclosure obligations while preserving the strongest available protections for legal strategy, mental impressions, and draft work. Companies using—or considering—GenAI in litigation should work with counsel to establish these protocols before a discovery dispute puts their AI history under scrutiny.