Did Your Witness Use ChatGPT? Better Find Out Right Away.
AI chats have become a litigation discovery and privilege issue. For defense counsel, the practical approach belongs in the same intake and preservation process already used for notes, emails, texts, and other ESI.
Defense lawyers already know what to ask when a new case comes in.
Does the insured have emails about the incident? Text messages? Notes? Photographs? Recorded statements? Internal reports? Where are these documents? Has anything been deleted?
Those questions are routine because lawyers understand that seemingly informal communications can later become important evidence. They identify what exists, tell the client not to delete it, and arrange for collection in due course.
There is now another set of questions to be added to that familiar inquiry:
Have you used ChatGPT, Claude, Copilot, Gemini, or another generative-AI tool to discuss this accident, claim, or lawsuit? Has anyone else done so at your request?
Because of the way these chat services function, a witness may have created a detailed written record without thinking of it as a written record at all. And the time to find out is at the beginning of the representation, not months later when the witness is getting ready for their deposition.
Ask About AI at Intake
Consider an insured driver who receives suit papers and asks ChatGPT:
“What questions is the plaintiff’s lawyer going to ask me?”
That may lead to:
“Here’s what happened. How should I explain this so it’s not my fault?”
A store manager might ask whether the company could be liable for a customer’s fall. A supervisor might upload an incident report and ask for an analysis of the company’s exposure. An employee might paste allegations from the complaint into an AI tool and ask what they mean.
To the witness, this may feel more like a Google search than the creation of a statement.
But the resulting conversation may contain another version of the witness’s story, potentially including facts, omissions, admissions, concerns, or suggested wording that differs from an incident report, recorded statement, discovery response, email, or later testimony. That makes asking about AI use a logical extension of the document and ESI inquiry defense counsel already conducts.
Preserve What Exists. Caution Against Further Use.
If the witness tells you they used a generative AI tool, just like a litigation hold, they should be advised not to delete anything. Potentially relevant chats should be preserved while counsel determines what exists, when it was created, what information or documents were supplied to the AI system, and where the material is stored.
Counsel should also advise the client or witness not to continue independently using generative-AI tools in connection with the case such as analyzing potential defenses, rehearsing testimony, summarizing confidential documents, or generating interrogatory answers.
The principle is not new: preserve what exists and avoid creating new potentially discoverable material relating to the litigation.
Best practices are to document preservation efforts and non-use requests in appropriate client correspondence, the same way other preservation instructions are commonly handled.
If there are chats that predate your retention, the next step is to evaluate whether they are privileged. Whether they are will turn on the specific facts and applicable law, which is constantly evolving. Here are a few cases to get you started.
Heppner, Warner, and Shealy: Why the Facts Matter
In United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), a represented criminal defendant independently used Anthropic’s Claude to generate materials concerning possible defense strategies. His lawyers had not directed him to use Claude. He later shared the materials with counsel and asserted attorney-client privilege and work-product protection. The Southern District of New York rejected those arguments. Among other things, the court emphasized that Heppner created the materials on his own initiative rather than at counsel’s direction.
A different result reached in Warner v. Gilbarco, 820 F. Supp. 3d 629 (E.D. Mich. 2026). There, a pro se plaintiff used ChatGPT in connection with her civil litigation. The Eastern District of Michigan treated the material as protected work product and rejected the argument that using ChatGPT itself necessarily waived that protection.
More recently, Shealy v. Seaside Investments, No. 2684CV00799-BLS2 (Mass. Super. Ct. June 16, 2026), provided yet another warning for represented parties acting outside counsel’s direction. There, a plaintiff sent litigation-related materials to his romantic partner, who uploaded them to ChatGPT and generated responsive materials. The Massachusetts court ordered production, concluding that neither the partner nor the AI system supplied the kind of litigation-representative relationship necessary for work-product protection.
These cases arise in different settings and under different procedural rules. They do not establish a simple rule that AI chats are privileged or that they are not. They show something more useful for practicing defense lawyers: protection may turn on who used the tool, why it was used, whether counsel directed the activity, and the circumstances in which the material was created. That uncertainty is itself a reason to get control of the issue early.
Individual Insureds Are Easier. Corporate Defendants Are Not.
For many insurance-defense matters, the inquiry is relatively straightforward. If counsel represents an individual insured, say the driver in an auto accident case, the questions can be asked right at the outset of representation. If the insured used AI concerning the accident or lawsuit, counsel can address preservation and future use immediately.
Corporate defendants present more of a challenge. There may be managers, supervisors, safety personnel, HR employees, engineers, drivers, investigators, or others with relevant knowledge. Counsel may not even know all their identities when the case first arrives.
That means engaging them takes more effort. But it needs to happen as soon as practical once the representation begins to minimize the potential for use of AI tools by them. As additional witnesses are identified, the inquiry can follow them.
Ask Again Before the Deposition
Even if counsel raises AI use at intake, the question should be asked again during deposition preparation.
Generative AI services are now embedded practically everywhere, from browsers and office software to search tools and phones. A witness who had never used ChatGPT before the lawsuit began may forget or ignore your advice and ask it to explain the complaint, analyze documents, identify likely deposition questions, or suggest answers without appreciating the potential consequences.
Defense lawyers already ask witnesses what they reviewed before deposition, whom they spoke with, and whether they revisited prior statements. The inquiry should now include whether they used generative AI. By then, however, it should ideally be a confirmation rather than a surprise.
The worst time for defense counsel to learn about case-related AI use is when opposing counsel asks:
“What did you do to prepare for your deposition?”
And the witness answers:
“I used ChatGPT.”
Ask. Preserve. Instruct. Ask Again.
Generative AI does not require defense lawyers to reinvent case intake. But it does require them to expand an inquiry they already make.
Ask whether the client and witnesses have used generative AI concerning the incident, claim, lawsuit, or anticipated testimony.
Preserve existing conversations and uploaded materials.
Instruct appropriate clients and witnesses against further case-related AI use, and document that instruction.
Ask again when preparing witnesses for deposition.
Litigation files have always contained multiple versions of events: incident reports, recorded statements, emails, medical records, discovery responses, deposition testimony, and more. Generative AI conversations by clients and witnesses now create one additional version. The defense job remains the same: learn what has been said, know where the accounts differ, identify those differences before opposing counsel does, and inform the client to fairly consider it in formulating defense and resolution strategies.
That is also where tools such as esumry can help. As the litigation record grows, the challenge is not simply collecting more information. It is understanding how it all fits together—and finding important inconsistencies before they become surprises.
Schedule a demo to see how esumry helps defense teams turn case documents into structured, accessible litigation knowledge that supports trusted, faster, consistent pretrial work.
About the Author
James Chapman is a co-founder of esumry and a defense litigator. He writes about the intersection of AI, litigation strategy, and legal operations.
Using esumry, privilege is protected with ZDR (zero data retention), and case analysis is fast, strategic, and secure. Create timelines, tag testimony, assess credibility, and get ahead of how the other side will use the record—before they do.
