Eight in ten family law attorneys at one national firm say a client has sent them something written by AI. In a survey Cordell & Cordell published in May, 80% of 133 attorneys there had received correspondence or draft documents from clients made with AI.
Three months earlier, a federal judge in New York had ruled on what chats like those are worth in court. In United States v. Heppner, a defendant's exchanges with a public AI tool got no protection from attorney-client privilege or the work product doctrine.
Together they make ChatGPT and attorney-client privilege a front-desk problem. The first person to hear "I already asked ChatGPT" usually isn't the attorney. It's whoever answers the phone.
Are a client's ChatGPT chats covered by attorney-client privilege?
In that case, no. It was a criminal fraud prosecution, not a divorce. The defendant used a public AI tool on his own, after he learned he was under investigation, to think through his defense. He later gave the results to his lawyers. Agents seized about 31 of those documents, according to Akin's summary of the ruling.
Judge Jed Rakoff's written opinion is dated February 17, 2026. His reasons were plain. The tool is not a lawyer. Its own privacy policy left no reasonable expectation that the chats were confidential. And the defendant was acting on his own, not at his lawyers' direction.
One phrase from the opinion belongs on a sticky note at every intake desk. Non-privileged communications, the judge wrote, are not "alchemically changed into privileged ones upon being shared with counsel." Forwarding the chat to the firm afterward does not fix it.
I don't want to overstate this, so three cautions.
The tool in that case was Claude, not ChatGPT. As the ruling has been summarized, the reasoning rests on the tool's terms and on who directed the work, not on the brand.
It is one trial court's ruling in a criminal case.
A week earlier, a federal court in Michigan went the other way on a related question. In an employment case, it refused to make a self-represented plaintiff hand over her AI materials and treated them as protected work product. Kirkland & Ellis compares the two rulings here.
I have not found a family court decision that settles the question. Your state's privilege rules are the ones that count.
Why this lands on the legal assistant's desk
Family law attorneys are already drawing the line from that ruling to divorce. Two attorneys at Ward and Smith wrote in April that the reasoning reaches divorce, custody, alimony and property cases. They warn that opposing counsel may ask about AI use in depositions and may subpoena a party's prompts and outputs.
Here is a hypothetical, and I'm labeling it as one. A client gets a letter from your attorney on Friday afternoon. Over the weekend she pastes it into a chatbot and asks whether the advice is right. On Monday she emails you the answer.
That client has not done anything foolish. She was anxious, and the chatbot was awake. But the firm's letter may now sit with a third party, and nobody at the firm knows until someone opens that email.
You are the one who opens it.
What to say when a client says "I asked ChatGPT"
These are scripts for your attorney to approve. They are not legal advice for you to give.
Ask at intake, every time, in a neutral voice: "Have you used ChatGPT or any other AI tool for anything about your case?" Write down the answer and tell the attorney the same day.
When an AI draft arrives, thank the client: "Thank you for sending this. I'll make sure your attorney sees it." Don't say whether it is right. That is the attorney's call.
Ask one follow-up: "Did you put any of our letters or your documents into it?" Note which ones.
Give the line your attorney has approved. Something like: "Please hold off on putting anything about your case into an AI tool until you've talked with your attorney. What you type there may not be private the way talking with us is."
Never tell a client to delete a chat. Whether it has to be kept is a legal question, and it belongs to the attorney.
None of this works if the client feels judged. Assume they were trying to help.
What the attorney should decide this week
Three decisions make the scripts above possible.
One paragraph for the engagement letter or welcome email. What the firm asks clients to do, and not do, with AI tools while the case is open.
Where a client's answer gets recorded. If a legal assistant hears "I asked ChatGPT," who is told, and where it goes in the file.
Whether to ask the other side. If opposing counsel may ask your client about AI use, you may want to ask theirs.
There is also an opening in the New York ruling. The judge noted that if counsel had directed the defendant to use the tool, it "might arguably be said to have functioned in a manner akin to a highly-trained professional" acting as the lawyer's agent. That is not a holding. It does suggest that telling clients how to use AI, under the attorney's direction, may hold up better than a ban nobody follows.
I'd rather a firm give clients a safe way to ask than pretend they will stop asking. This is the same conversation as what your own AI reads in the other side's filings.
I think clients turn to a chatbot when the firm's answer feels far away. That part a firm can change. Del, our AI teammate for family law firms, reads the whole matter file, keeps it current, and answers questions with the source document attached. Del does the first pass and the team reviews it. Client data is never used to train models.
A client who asked ChatGPT first is not the problem. Finding out in the middle of discovery is. The fix costs one question at intake and one paragraph in a welcome email.
If you want to see how Del keeps a matter file ready for those questions, start a free trial. The account is free to start.
This article is practical commentary, not legal advice. Privilege rules differ by state, so check your own state's rules, and have your attorney approve any script before you use it.
Related reading: Hidden Text in a Court Filing Told AI to Take a Side · AI for Family Law: What Works, What Doesn't, Where to Start