Nearly 41 percent of Americans say they’re comfortable using AI to create or update their estate planning documents. Among Millennials and Gen Z, it’s more than half.
If you plan estates for a living, you probably had a physical reaction to that number. You should have. You know what’s actually in those documents. You’ve seen the trust that was never funded, the power of attorney the bank wouldn’t accept, the successor trustee named twenty years ago who is in no condition to serve today. You know the mistake doesn’t surface until the person who made it is gone and can’t explain what they meant.
So the instinct is to argue with the number, and I understand it. But it isn’t going backward, and your clients aren’t going to stop asking AI questions because you’d prefer they didn’t. The number is a fact about the market you’re practicing in now.
What’s actually worth your attention is a distinction hiding underneath it.
A question and a verdict are not the same thing
Here’s a client using AI exactly right.
She types something like this: “My son is currently named as executor of our estate, but he’s in the middle of a nasty divorce and I’m not sure he can take this on. What should I be thinking about?”
AI gives her a handful of things to consider. Whether the divorce creates exposure she hasn’t thought about. Whether there’s a successor already named. Whether a corporate fiduciary makes sense here. Whether this is a permanent change or a temporary one. She writes them down and brings them to her appointment.
That is one of the best client conversations you’ll have all month. She’s engaged, she’s thought about it, and she’s handing you the exact issues you’d otherwise spend forty minutes drawing out of her.
Now the other version of the same appointment. She arrives having already decided. “ChatGPT told me I should name my daughter instead. Can you just do that today?”
Same client, same underlying situation, completely different meeting. You’re not advising anymore. You’re overruling something she has already accepted as true, which is a much harder job and a worse experience for both of you.
The distinction is simple enough to teach, and it’s worth teaching explicitly: ask AI what to consider, not what to decide. It is genuinely good at generating the questions. It is in no position to answer them, because it has never met that family.
Why she can’t see what you would see instantly
Try this sometime. Ask AI a question about something you know cold, whatever you’d be comfortable being deposed about. The funding requirements you handle every week. A Medicaid rule in your state.
You’ll find the holes in about ninety seconds. Something oversimplified, something that’s true in most states but not yours, something that reads authoritative and is quietly two years out of date.
Now ask it about something you don’t know well, and read that answer.
It looks identical. Same fluency, same structure, same calm confidence. Nothing in the writing signals which answer is solid and which one isn’t.
That’s the mechanic underneath most of what’s frustrating about this. The confidence does not vary with the accuracy. What varies is your ability to catch the difference, and that tracks almost perfectly with how much you already know about the subject.
So your client, reading about trust funding for the first time in her life, has no way to see what you’d see immediately. Not because she’s careless or credulous. The tell isn’t in the text. There is nothing there to catch.
Which is worth sitting with for a second, because it also means the same thing is true of you the moment the subject changes to something outside your practice.
We’re on the other side of this too
An attorney runs their firm’s website through ChatGPT, asks it to evaluate the SEO, and sends us the output as a support ticket. Confident list of what’s wrong. No context about what we’ve already tested, what the analytics actually say, what changed on that site four months ago, or what we’re deliberately not doing yet.
Sometimes the ideas are worth having. AI will raise something we hadn’t gotten to yet. More often, it surfaces something we floated six months ago that got set aside for perfectly good reasons at the time. The budget wasn’t there, the site couldn’t support it, two other things were ahead of it in line. And now it looks like exactly the right next move. That’s a real conversation and I’d like to have more of them.
What the tool can’t do is know any of that history. It doesn’t know it’s re-proposing something already on the table. It doesn’t know we tried a version of it in March. It doesn’t know that recommendation was written for e-commerce and doesn’t survive contact with a legal site, where the accuracy bar is different and the stakes are somebody’s estate. It doesn’t know three firms in that market are already doing it, which is precisely why we aren’t.
So the ideas arrive with no history attached, and they arrive as instructions rather than as questions. That second part is what turns a good conversation into a standoff.
Four lines that change the meeting
We can’t stop it from happening. We can decide what happens in the next sixty seconds. Four lines that work, in roughly this order:
“Good, you did the homework. Let me tell you what it got right.” Start by agreeing with something. If you open by correcting, she defends, and now she’s arguing instead of listening. Find the accurate part first. There almost always is one.
“Here’s what it didn’t know about your situation.” Not what it got wrong. What it couldn’t have known. Your state, your family, your funding history, the thing you told me in 2019. That puts the gap in the tool’s missing context rather than in her judgment, and it makes you the person with the information instead of the person with the objection.
“Ask me the question you asked it.” This is the one that changes the meeting. It converts a verdict back into a question and puts you where you belong.
“Next time you look something up, bring it to me exactly like this.” Invite the behavior. A client who researches and brings it to you is more engaged, more likely to implement, and far less likely to quietly handle it herself with a form off the internet.
You are not competing with a chatbot for authority. You’re teaching someone how to use a tool she’s going to use anyway, in a way that routes her back to you instead of around you.
We Call This the Pre-Advised Prospect
The prospect who calls your firm today has usually already asked her questions somewhere else. She’s read a summary of how a trust works. Often she’s already priced a do-it-yourself option. She arrives later in her own decision process, better informed, more skeptical, and asking sharper questions.
Here’s how that tends to show up on your end: the close rate slips, and the reasonable conclusion is that the leads got worse. Tire-kickers. Price shoppers. People who were never serious.
It’s almost always the wrong read, and it’s an expensive one, because it points the firm at the wrong fix. Decide you have a lead quality problem and you’ll buy more leads or fire your marketing, spending more money to feed the process that’s actually the bottleneck. It’s the same person as before, arriving in a posture your consultation was never built to meet.
That shift is the subject of our industry report, The Estate Planning Reset: what has changed in how clients find, choose, and hire an attorney, why the usual diagnosis is wrong, and what to do about it in what order.
You can download it here: https://imsrocks.com/estate-planning-reset/
