Can ChatGPT Conversations Be Used in Court?

Can Someone Get Your AI Chats in a Lawsuit?

Yes. Courts treat AI prompts and outputs as electronic records, so the other side can ask for them in discovery. Singapura, a trial lawyer at Bell Nunnally in Dallas, compares them to email: a message you send today can be demanded by a lawyer two years from now, and the questions you type into a chatbot are now being treated the same way.

The provider is a second route in. In a government investigation, Singapura says, investigators go directly to the AI vendor for the records. The consumer terms at issue in the New York case let the provider collect what users typed and disclose it to third parties, government authorities included, and the court leaned on that in finding the chats were never confidential.

Deleting a conversation isn't a fix. Singapura never advises a client to delete anything, and his experience is that “it's hard to ever, I've learned, get rid of anything electronic.”

Watch the Full Conversation

This article draws on a Navigating Wealth conversation with Kartik Singapura, where we discuss how AI is changing litigation, what clients should know before using chatbots on a legal problem, and whether legal bills will fall. Watch the full episode for the broader discussion.

Subscribe on YouTube · Get weekly episode updates

Kartik Singapura is a partner at Bell Nunnally, a Dallas law firm, where he tries commercial cases in Texas and beyond. He has practiced for about fifteen years and describes the courtroom as his passion. In the conversation with Tad Fallows and Matt Shechtman, he explains where AI helps a legal team, where it misleads clients, and what the early court rulings mean for anyone heading into a dispute.

What Courts Have Ruled About AI Chats So Far

Three rulings in 2026 went in different directions, and the deciding questions were who directed the AI use and whether the work was prepared for litigation.

Source: Rulings as summarized by Gentry Locke (Heppner), Proskauer (Warner) and Haynes Boone (Tate Group).

In United States v. Heppner, Judge Jed Rakoff found that a chatbot isn't a lawyer, that the consumer terms meant the chats weren't confidential, and that the defendant used the tool without his attorneys' direction. The court suggested the result might differ if counsel had directed the use. It also held that the documents didn't become privileged when the defendant later shared them with his legal team.

The Michigan ruling protected a self-represented plaintiff's ChatGPT work, reasoning that an AI program is a tool, not a person, so using it isn't a disclosure that waives protection. In Tate Group, the Texas Business Court read Texas's work-product rule, which covers material prepared “by or for a party,” as broader than the federal standard and expressly rejected Heppner. It still ordered production of pages that weren't prepared for the litigation.

None of these rulings binds courts elsewhere. Singapura's summary is that every state is treating disclosure differently, and courts haven't settled on one approach.

What to Do Before You Research a Business Dispute With AI

Talk to a lawyer first. The protection courts have recognized so far attaches to work prepared for litigation, and in federal court it leans on whether a lawyer directed it.

Partnership disputes and disputes inside closely held companies, Singapura says, carry a layer of emotion from whatever happened at the time, and that's when people reach for a chatbot to test their side of the story. The rulings and the episode point to three practices:

• Ask your lawyer how they want you to use AI on the matter before you start, which is Singapura's core advice.
• Don't paste your lawyer's emails or advice into a consumer chatbot. Commentators read Heppner as holding that doing so can waive privilege over those communications.
• Keep everything. Once a dispute is reasonably expected, deleting records can create a legal problem of its own.

How Litigators Are Using AI on Your Case

As a fast first pass that a lawyer then checks against the sources. Singapura describes document review as reversed: “AI can create the roadmap for you now. You use the roadmap to go review the documents.”

He's equally direct about the limit. He calls AI “an average intelligence,” useful for sorting volume but not for strategy, and he says a lawyer who hands a judge a fabricated or hallucinated case loses the motion and their credibility with it. He also uses NotebookLM to turn explanatory legal articles into a 30-minute audio briefing for a client, in place of a two-hour call the client would pay for.

If you're hiring a litigator, ask how their team verifies AI output before it reaches a court. The same verification question applies to founders building agentic AI tools for business operations.

Will AI Lower Your Legal Bill?

Not measurably yet, in Singapura's view. He sees faster work and easier client updates, but his litigation practice still bills by the hour, because the scope of a case is hard to underwrite in advance.

Fixed pricing is more plausible where the scope is fixed. He points to appeals, where the record is closed and the issues are known, as work a firm could price as a flat fee. Other legal work already bills that way: in Long Angle's 2025 High-Net-Worth Professional Services Report (114 member respondents, fielded August to September 2025), 57% of clients of trust and estate attorneys were billed on a project basis and nearly a third paid hourly.

Where AI does save money, in his telling, is the start of a case. Give it the emails with the two or three people at the center of a dispute and it can map the cast: who did what and when, and who's likely friendly. That lets a client separate who they need to talk to from who they want to talk to, which he calls “a big difference and a big cost difference,” instead of filing nine causes of action and sending discovery in every direction. Some plaintiffs fund that cost through litigation finance.

Where Doing It Yourself With AI Works, and Where It Backfires

It works in forums built for people without lawyers, such as property tax protests and small claims court. Singapura used AI on his own property tax protest.

It backfires in full litigation. Filings drafted by a chatbot for people representing themselves are, in his account, among the most frustrating things lawyers and in-house counsel now handle, because each error creates work to answer. Clients who use AI to understand the framework of their case are a different story: he finds an informed client a better partner, as long as the AI's take is treated as a starting point for the conversation.