If you have ever pasted a hostile text from your ex into ChatGPT and asked it to "write a calm reply," you are far from alone. Family-law attorneys report that clients now routinely pre-clear messages with a chatbot before sending them, and social media is full of parents calling it the best use of AI they have found. For a stressed co-parent, it genuinely helps: you get to vent, and the other house gets the composed version.
But in the last year, lawyers went from recommending this trick to warning about it. Here is what changed, in plain English, and how to get the same help without the new risk.
AI chats are not private in the way people assume
Conversations with a general chatbot feel like a diary. Legally, they are closer to an email account. In 2026, a federal court treated a defendant's AI chatbot conversations as ordinary discoverable material, and family-law firms have since published client warnings that divorce and custody chats with general AI tools can be requested in litigation. There is no attorney-client privilege with a chatbot, and no therapist privilege either.
That matters in a custody case for a specific reason: it is not just the polished message that can surface. It is the prompt you typed to get it. "Rewrite this so I sound reasonable" reads very differently to a judge than the reasonable message itself. In at least one reported matter, a parent's drafted-by-AI messages were undermined when the original prompts came out in discovery.
The second problem: no record where it counts
Even when nothing goes wrong legally, a general chatbot leaves your co-parenting record worse than it found it. The hostile message from your ex sits in one app. Your drafted reply sits in another. The version you actually sent lives in text messages. When you later need to show a pattern, whether to a mediator, an attorney, or a judge, the story is scattered across tools that were never meant to be evidence, with no timestamps that anyone can verify.
What to do instead
The answer is not to give up the help. It is to move the same help inside a system that was built for custody communication:
- Draft help that stays in the record. In CoreParent, when your reply reads hot, Companion offers a calmer rewrite that keeps every fact and request. You send it or you do not, and either way the thread stays in one certified, timestamped record.
- Incoming hostility, pre-read. Conflict Shield reads hostile incoming messages first and hands you the part that matters, while the original stays preserved, word for word, for the record. It is free on every CoreParent account.
- Privacy built for children. For everyday Companion features, CoreParent's servers replace your children's profile names with neutral placeholders before anything reaches the AI provider. We publish exactly what our AI sees and how long anything is kept at Your Data and AI.
- Nothing sent without you. Companion never messages your co-parent. Every suggestion is a proposal you confirm, edit, or dismiss.
If you keep using a general chatbot anyway
Some habits are hard to break, so harm-reduce: never paste your children's full names, school, or medical details into a general chatbot; assume anything you type could be read aloud in a courtroom someday; and keep the messages you actually send inside a documented channel, not loose text threads. If you are in active litigation, ask your attorney before using AI tools on case-related material at all, and know that some courts now require disclosing AI use in filings.
The calm-reply trick is real. The mistake is doing it in a tool that remembers everything and protects nothing. Do the venting where it cannot be subpoenaed into your case, and keep the record where it can actually help you: in one place, certified, and yours. See how CoreParent handles this end to end at CoreParent Companion.