Technology’s Effect on Advocacy
Discovery Meets the Prompt Box
That sentence may eventually become as routine as “Do not post about your case on social media.”
California generally permits broad discovery into relevant, nonprivileged matters and expressly encompasses electronically stored information. As AI use becomes more common, litigators should at least be thinking about whether AI-related material could become relevant in discovery depending on the facts of a particular case. Consider a client who uses an AI platform to create a detailed chronology of alleged discrimination before filing suit. Or a personal-injury plaintiff who repeatedly asks an AI system to rewrite a description of an accident. Or a witness who uploads documents and asks the system to identify “the facts that best support my case.” What happens to those prompts and outputs? Are they stored and accessible? Could they be responsive to a discovery request or used for impeachment? Could a particular use implicate privilege or work- product issues? The answers will be highly fact and platform specific, and courts will undoubtedly spend years working through these questions. The practical lesson is available now: we should know whether our clients are creating AI-generated material relating to their cases. The modern case file may include text messages, social-media posts, cloud documents, metadata and conversations with machines.
The AI Polished Witness
Perhaps the most interesting problem arises when AI stops merely advising the client and starts helping the client tell the story. Suppose we ask a client to prepare a chronology of events. Twenty years ago, we might have received handwritten notes. Ten years ago, we might have received a Word document. Today, we may receive a beautifully formatted timeline with headings, bullet points, legal terminology, and suspiciously perfect transitions. The client may have asked AI to “clean it up.” That sounds harmless, sometimes it is, but generative AI does not simply correct commas. Depending on the prompt and system, it may summarize, reorganize, infer, characterize, or embellish. A client might originally write: “My manager seemed annoyed after I complained.” After an AI rewrite, that sentence might become: “Following my protected complaint, my manager’s demeanor changed dramatically, and I experienced increasing hostility and retaliation.” Those sentences are not equivalent. The second may be more persuasive; it may also be less true. Now imagine that language migrates from an AI- generated timeline into an intake questionnaire, then into discovery responses, then into deposition preparation. Six months later, defense counsel produces the original contemporaneous text message: “Manager seemed kind of annoyed today lol.” The problem is no longer grammar, it is credibility. For plaintiffs’ lawyers, credibility is often the case.
Clients Will Continue to Use AI Regardless
Given all of this, the easiest solution might seem obvious: tell clients never to use AI. This is not realistic, and AI can be extraordinarily useful. A worker who suspects pregnancy discrimination can learn the vocabulary necessary to begin researching her rights. An injured person can generate questions
42
Consumer Attorneys of California
FORUM September/October 2026
Made with FlippingBook - Online catalogs