What ChatGPT Cannot Know About Your Divorce

Nearly every client now arrives with an AI tool already open.

They run my reports through it. They ask it what their strategy should be. They ask it whether a settlement offer is fair, whether their attorney is handling something correctly, whether the number I gave them is the right number. Then they come back to me with what it said.

I want to say something clearly, because it matters more than anything else in this piece.

Challenge me.

Ask me why. Ask me to show my work. Ask me what happens if we are wrong. Fifteen years inside family law firms taught me that the client who asks why walks out with a better outcome than the client who nods. I have never once been offended by a hard question, and I am not offended now.

The challenge is not the problem.

The problem is where the challenge is coming from.

Who is actually at the table?

A divorce that ends up well is worked by a team. An attorney. A financial strategist. Sometimes a business valuator, a mediator, a therapist. And the client, who is the only one at the table whose life this actually is.

Everyone in those chairs brings something they earned somewhere.

When an attorney challenges my analysis, they are drawing on cases they have tried. When a business valuator challenges my conclusions, they are drawing on defenses they have had to make in front of a judge who did not accept the first version. When a client challenges me, they are drawing on the one thing nobody else in the room has, which is knowledge of their own life.

An AI tool is not drawing on anything of the kind. It is a gigantic source of information, and no volume of information, however large, produces the thing that actually decides cases.

It does not get a chair. Not because it is unwelcome, but because it has nothing to bring to that particular table.

Here is what it cannot know, and cannot be taught by feeding it more:

• How attorneys actually operate, which is different from how the process is described.
• How a judge views a particular behavior, and how much that view will cost the person who displayed it.
• How a specific mediator moves a room when the negotiation stalls.
• How real people are affected by the realities of a divorce, and what they will and will not be able to live with afterward.
• What is genuinely worth fighting for, and what will cost more to win than to lose.
• When the right approach to a decision is to wait, or to concede, or to say nothing at all.

That is not information. That is strategy, and strategy is built inside rooms, over years, watching what happens after the decision rather than reading how the decision is supposed to work.

An AI tool is not a person. It is a gigantic source of information. Information is not strategy.

I tested it on purpose, and it was confidently wrong

A change in the law landed recently in a state where I work closely with business owners going through divorce. It touched an issue that was live in a case I was already handling, so I knew it well, and I knew it was recent.

So I opened an AI tool and challenged it deliberately.

I asked the way a client would ask. Then I asked the way an attorney would ask, with the correct terminology and the correct framing. Then I pushed, and told it plainly what I knew.

It held the outdated answer every time. It explained its reasoning. It laid out what it had found. It told me, with real confidence, that I had it backwards.

I did not have it backwards.

It never hedged. It did not tell me the area was unsettled, or that treatment varies by state, or that I should confirm the current rule locally. It was certain, and it was certain about something that had already changed.

I caught it because I knew the change existed before I typed the question. That is the only reason I caught it. A spouse asking that same question three weeks into the worst year of their life has no way to know the answer is out of date, and neither does the tool.

Asking a better question does not fix this

The usual response is that people simply need to prompt better.

That is not the issue, and I want to be direct about why.

A better prompt cannot manufacture experience the tool does not have. If a system has never sat in a mediation at hour seven, no amount of careful phrasing will produce what happens at hour seven.

And there is a second problem, which is worse.

A business owner I follow ran an experiment on her own company. She fed her real financials into an AI tool, exactly the way owners keep getting told to. The first read was fair and reasonable. Then she rewrote her prompt to sound optimistic, and the analysis turned optimistic. She rewrote it once more to sound anxious, and the same numbers told a heavier, more worried story.

The data never changed. Only her framing did.

Her conclusion is the part I keep sitting with. A prompt is never a neutral question. It carries your mood, your hope, and your fear, and the tool builds its answer around all of it. On the days you are least equipped to see clearly, that is precisely when it will agree with you the most.

So the prompt is not the solution. The prompt is a second point of failure.

Why divorce is the worst possible place for a tool that agrees with you

Nobody in a divorce is neutral.

The people I work with are frightened, angry, guilty, exhausted, and hopeful, sometimes inside the same hour. Every one of those states writes a different question.

A frightened spouse asks whether they are going to lose everything, and the answer gets built around loss. A spouse carrying guilt asks whether they are being fair enough, and the answer gets built around generosity. A spouse who has already decided to keep the house asks whether keeping the house is reasonable, and the tool finds the reasoning.

None of that is analysis. That is reflection with a confident voice.

This is the Confusion Tax in its purest form. The Confusion Tax is the unnecessary financial cost of making complex decisions under emotional stress. Nobody bills you for it. It arrives later, in the asset you kept and could not afford to carry, in the consequence nobody modeled, in the term that made sense in the room and stopped making sense the following April.

Am I against AI?

No.

I use these tools, and I use them well. They are genuinely useful, and they will keep getting more useful. Refusing to touch them would not make me a better strategist. It would make me a less informed one.

Use one to organize your documents and build your timeline. Use one to learn what the terms mean before your next meeting, so you are not paying by the hour to have vocabulary defined. Use one to draft the list of questions you want answered.

What I would not do is let one decide your strategy during one of the most significant and most challenging times of your life.

Divorce is a high-stakes financial restructuring disguised as "just" a legal event, constantly hijacked by emotion. And the hijacking does not stop at the courtroom door. The same fear, guilt, and exhaustion that drive bad settlement decisions are the very emotions typing questions into an AI tool at two in the morning. That is why the answers feel so right on the hardest nights, and why they deserve the least trust exactly then.

The restructuring is where wealth is protected or lost, and restructuring does not respond to confident language. It responds to strategy, applied to your actual facts, by someone who has watched what happens on the other side of the decision.

So bring me the challenge. Bring me the printout. I will show you my work, and I will tell you where the tool is right, because sometimes it is.

Then I will tell you the part it could not possibly know, and we will decide what to do about it with the people who have actually been in the room.

Strategy is the part you cannot automate.

This is why AI cannot sit at the table.

Divorce does not have to destroy your wealth. With the right strategy, it can protect it.