Several points remain unsettled. The courts have not agreed on how AI conversations fit into litigation. In February, a federal court in Michigan held that a self-represented litigant's ChatGPT conversations about her case were work product. The same day, a federal court in New York held that documents a criminal defendant generated using Claude were not privileged attorney-client communications or work product. Courts have remained split since. The reported effect on settlement is also anecdotal. Magistrate Judge Allison Goddard of the Central District of California has said unrepresented parties often get poor advice from ChatGPT when valuing their claims, and Outten & Golden partner Dave Jochnowitz says chatbots tend to be overly encouraging. Those observations are not backed by data. A mediator's blog cites MIT and Stanford research on repeated AI interactions reinforcing certainty, but the source is secondhand and should be treated as context, not evidence.
For practitioners, the immediate task is managing expectations early. Ask new clients what AI tools they have used on their matter and what they asked them, because those exchanges may be discoverable depending on the jurisdiction and the court's reasoning. When a chatbot has given a client an optimistic valuation, deliver your own assessment with the reasoning spelled out, since a measured view can land as bad news after a positive one. Watch for further privilege and work-product rulings on AI chats, particularly any appellate decision that would resolve the split.