Extract from Emory & Grossman’s article, “GenAI Prompts in eDiscovery: Protected Work Product or Not?”
Generative AI (“GenAI”) is reshaping how attorneys search for responsive documents in eDiscovery. In GenAI technology-assisted review (“GenAI TAR”), attorneys write natural language prompts instructing a large language model about what makes a document responsive (or not). Properly developed, these prompts are refined iteratively against a review population and can reflect precisely what the work-product doctrine has long sought
to protect, including an attorney’s factual investigation, mental impressions, and case strategy. This scenario raises a question that courts will increasingly confront: Should GenAI TAR prompts be disclosed, like the search terms that parties are routinely required to exchange—or protected, like attorney-selected seed or training sets for TAR, document compilations assembled to prepare witnesses for testimony, and document-review protocols?
This article argues that absent party agreement or a demonstrated deficiency in the production (or other discovery misconduct), prompts should ordinarily not be subject to compelled disclosure, and that prompts reflecting strategic judgment or case theory should be protected as opinion work product. It also posits that routine disclosure may incentivize producing parties to generate generic prompts that mirror the requests for production,
resulting in a weaker, less effective outcome for both parties. Transparency is better served, in most cases, by validation of results through sound sampling methods, proper calculation of recall and precision, and careful examination for document types or categories that may have been missed, rather than by exposing—and haggling over—the prompts themselves.
