How Family Lawyers Can Prevent ChatGPT’s Bad Legal Advice — Inside ChatGPT
The problem isn’t only that ChatGPT’s advice is often wrong. It’s that your client is telling it things that can resurface in court — but you can be the one who warns them, at the moment they’re typing.
You’ve sat across from this client. She’s five minutes into the consult and she’s already correcting you. ChatGPT told her she gets half of everything and primary custody, so she really just needs help with the paperwork. You explain that your state doesn’t divide things in half, that full custody isn’t a presumption, that the advice on her phone assumed facts that aren’t hers.
And you watch the doubt land… not on the chatbot, but on you.
Every family lawyer I work with and talk to has some version of this story now. The client who shows up “already researched.” The one quoting a case that doesn’t exist. The one who pasted your strategy email into a chatbot for a second opinion.
The instinct underneath it is reasonable. People want to understand their own situation before they spend money on a lawyer and they want to participate in their case when they have one. The trouble is where they go to get that understanding, and what it costs them when the machine is confidently wrong.
So the question I get asked, in one form or another, is this: what do I do about it? How do I correct it? How do I get my expertise back in front of someone who trusts a free chatbot more than the person whose license is on the line?
You can’t wait to correct ChatGPT in your office. By then it’s too late.
You correct it inside ChatGPT, at the moment your future client is asking the question.
That’s a real, working channel now, and I’ve run it. But first, the two things you’re correcting.
What ChatGPT gets wrong, and why your office is the wrong place to fix it
There are two problems, and most lawyers only focus on the first.
The first is accuracy. A chatbot gives clean, confident answers because it only ever sees the clean version of the story. It takes the three sentences someone types at 11 p.m. and responds as if those three sentences are the whole marriage. It doesn’t know about the business started during the marriage, the prenup in a drawer, the child with medical needs who reshapes the entire custody question.
And laws vary by state, so the machine will do things like explain community-property rules to someone living in an equitable-distribution state in the same even voice it uses for everything else. It can invent case citations that look real, down to the court and the year, and courts have already sanctioned actual licensed lawyers for filing the fakes.
Your prospect doesn’t walk in unsure. He walks in armed with a position built on something generated to sound right rather than be right, and you spend the meeting taking it apart.
That part is frustrating, and often annoying.
The second problem should concern you more, because your client has no idea it exists.
The part your client doesn’t know: ChatGPT isn’t private
When people pour their divorce into a chatbot, the finances, the custody fears, the hidden assets, the things they’d never say in front of a judge, they assume it’s a private conversation. As a matter of law, that assumption is wrong.
There is no AI privilege. What you tell your lawyer, your doctor, and your therapist is protected. What you type into a consumer chatbot is not. In February 2026, a federal judge ruled that a defendant’s conversations with a consumer AI tool weren’t covered by attorney-client privilege, partly because the company’s own policy let it keep and disclose what he typed.
Courts theoretically treat these chats the way they treat texts and emails: ordinary electronically stored information (ESI) a party can be made to produce in discovery. OpenAI was ordered to hand over 20 million user conversations in a single case earlier this year. Its own CEO has said in public that there’s no legal confidentiality when you talk to ChatGPT the way you’d talk to a lawyer, and called the situation screwed up.
Precision matters here, and it happens to be better positioning too. No published divorce ruling that I’ve been able to find has put a spouse’s chatbot log into evidence yet. The discovery rules that would allow it are already settled, and the first cases are a matter of when, not if. A lawyer who tells the public “your AI chats ARE evidence in your divorce” is overstating it, and the careful version lands harder anyway: what your client types into ChatGPT is generally not privileged, it’s discoverable like any other electronic record, and the safe assumption is that opposing counsel and the judge could one day read it.
Here’s why that’s the warning that matters most:
A survey of 1,000 AI users last year found that about half had no idea their conversations could be subpoenaed, and 51% said that knowing it would make them much more likely to talk to a human lawyer instead.
That’s a behavior change from a single fact. Most legal marketing has to manufacture a reason to call. This reason already exists, and the people you want have told you, on the record, how they respond to it.
Why it’s hard to correct any of this after the fact
By the time someone is sitting in your conference room, three things have already happened. The belief is set, and now you’re the one introducing complications. The conversation is logged on a server, potentially discoverable. And sometimes the mistake is already made, the email sent, the money moved, the kids told, because a confident chatbot suggested it and nobody warned them first.
You can’t un-ring those bells. The correction has to reach people earlier, while they’re still typing, before the wrong belief hardens and before the risky message goes out. Which means it has to happen in the one place you’ve never been able to reach them: inside the AI conversation itself.
Until recently, this was impossible. But I’ve proven it is very possible now.
How to correct ChatGPT, inside ChatGPT
ChatGPT opened to advertising on May 5, 2026. The policy bans ads for legal services and names personal injury, immigration, claims, and document prep. So the traditional paid ads move, “experienced divorce attorney, free consultation,” is off the table, and most people concluded law firms simply can’t advertise there.
They missed the one carve-out, or couldn’t figure out how to use it. OpenAI permits ads for general legal education, and correcting AI misinformation and knowledge gaps is education by definition. An ad that says “ChatGPT can give you false advice, and what you type into it isn’t private” offers no legal service and makes no consultation pitch. It corrects a public misunderstanding, which is what the education carve-out is for.
I wrote about that carve-out before any family law firm had used it. Then, on June 1st, I placed what I can document as the first confirmed family law firm ad inside ChatGPT. My first attempts were rejected, and the reason taught me the whole game: the educational offer was fine, but hosting it where a law firm obviously sells legal services was not. The fix was a stripped, education-only property, with no branding, no consultation offer, nothing but the resource. Everything that connects a reader to the firm happens on the back end, after they opt in to communication.
In its first week, that ad ran at $2.33 a click with about 22% of visitors opting in for a free educational resource. Those are opt-ins for a guide, not signed clients, but they’re names, emails, and demonstrated divorce intent, captured inside the platform where people are already confiding their situation. Legal keywords are famously the most expensive clicks Google sells. This reached the same people for a fraction of that, at the moment of highest intent.
That first ad offered plain education. The version I’d run now is sharper, because the warning we just walked through is stronger than any generic tip.
Picture it serving while someone is mid-conversation about their divorce: your AI chat about your divorce isn’t private and can be used against you in court, here’s a free guide on what to know and what to do. They click. They read a plain-English explanation of what discovery is, why not to delete anything, why their state’s rules are specific to them, and where the one conversation the law does protect lives. The guide naturally points to their need for a real attorney. The correction and the introduction happen in the same motion.
The caveats worth stating plainly
I’d be doing you a disservice if I sold this as a sure thing.
Ad approval involves reviewer discretion, and the carve-out is narrow, so the educational property has to stay clean. No court has ruled on whether a divorce litigant’s chatbot history is admissible, so the warning has to be framed as the documented risk it is, not settled law. State bars regulate attorney advertising differently, which matters for how this is actually executed. And this is a new channel. The right way to treat it is as a learning line item while it matures, not a faucet you turn on for instant clients.
None of that changes the core opportunity:
Legal lead clicks for under $3 on a platform that more of your potential clients are looking for legal advice on than Facebook or YouTube, and that your competitors can’t access.
But it does mean it should be run by someone who understands both the platform policy and the bar rules.
The position goes to whoever takes it first
There’s a difference between the firm that warned its market before the headline and the dozen firms that pile on after. The first reported divorce case built on a chatbot confession is coming, and when it lands, every legal marketer in the country finds this angle at once. Right now, in almost every market, the warning is fresh, true, and unclaimed.
Correcting ChatGPT in your office means arguing with a belief that’s already set. Correcting it inside ChatGPT, at the moment your future client is typing, is how you get there first, with the accuracy they’re missing and the warning they’ve never heard and the trust that will make them want to hire you.
If you want to think through whether this fits your firm and your state, I run a limited number of strategic assessments. You can book one at https://calendly.com/joshuadlittle/aquila-assessment, or reach me directly at joshua@aquila.law.
