When ChatGPT joins the fight

Somewhere in the middle of the argument, a phone comes out. “I described what you do, and ChatGPT says it’s called stonewalling. Want me to read you what it says about people who stonewall?” Or the fight happens, everyone retreats, and an hour later a screenshot arrives: the machine’s analysis of you, in clinical language, with your worst moment as the only exhibit. Couples have started bringing a third voice into the room, and journalists interviewing spouses on the receiving end keep hearing the same phrase, ganged up on. Two against one, and one of the two never gets tired.

Why does bringing in the machine feel so fair?

Because it looks like an appeal to a neutral party, and after years of the same fight, a neutral party is the thing you’re starving for. Somebody, finally, to just look at the facts and say who’s right. Here’s the problem, and it’s structural, not a bug they’ll fix in the next version. The machine never meets the facts. It meets your telling of the facts, alone, worded by the very anger it’s being asked to rule on. Then it does what it’s built to do, which is be maximally helpful to the person typing. So each of you can feed it the same fight and each of you will receive your own position back, upgraded, footnoted, wearing a lab coat. You didn’t hire a judge. You hired a lawyer who bills nothing, never sleeps, and has met exactly one client.

What does it actually do to the fight?

It industrializes the verdict. The recurring fight was never really about its topics. It runs on a harsh conclusion each of you has quietly reached about the other, and every round is that conclusion being read into the record. What the machine adds is production capacity. The accusation that used to arrive as “you always” now arrives as a named pattern with a definition, which feels like progress and is actually escalation, because a diagnosis is just a verdict with credentials. And it lands worst on the exact person least able to answer it. If your partner is the quieter one, the one who shuts down when it matters most, they were already losing every verbal exchange on points. Now they’re debating a professional. The shutdown gets deeper, which gets cited as proof of the diagnosis, and around it goes. Analytical couples should recognize the trap from trying to debug the fight: this is that, outsourced. You cannot patch a conviction, and you especially cannot patch it by adding compute.

If you’re the one being quoted at

Don’t debate the machine. You’ll lose, and losing to it will be entered as more evidence. The text isn’t really the point anyway. Underneath the citations, your partner is saying something older and simpler: I don’t feel heard by you, so I found something that hears me. That’s the sentence worth answering. Try: I’m not going to argue with a printout, but I do want to hear this from you, in your words. And then the hard part, actually hear it, without the counter-case. If you dismiss the content because of the messenger, you hand them the next screenshot.

Is there any good way to use it in a fight?

One, and it points the machine in the opposite direction. Used privately, on yourself, it has real value: I’m furious and I don’t fully know why, help me figure out what got touched. That’s the machine as a mirror for your own reaction, which is the one thing in the fight you can actually work on, and the one subject on which you are the qualified witness. The rule that keeps it honest is simple: the machine can help you understand you. The moment its output is aimed at your partner, live, as evidence, it has changed sides, from mirror to weapon. Nothing it says enters the room as a verdict about the other person. If you can’t hold that line, the two of you aren’t fighting about the app either. You’re inside the same old shape with a new tool, and the way out of the shape hasn’t changed. It’s the work of dropping the case rather than winning it, and no machine can do it for you, though the two of you can do it together, with help, in the room.