Ten minutes into a first session, before the intake paperwork is even finished, most experienced couples therapists have already formed a private guess.
Everyone in that room has a presenting problem, and the ones about money and in-laws and whose mother is coming at Thanksgiving are more or less interchangeable. The guess forms somewhere else.
It forms in the texture of the first sixty seconds after somebody finally says the thing out loud. Whether that sentence names a problem or names a defect in a person. Whether a small joke gets made, and more importantly whether it gets taken. Whether one of them goes quiet in a way that looks like composure and isn’t.
Then there’s the story. Ask a couple how they met and one of them lights up while the other produces a set of facts. Ask about the worst year and you find out whether it happened to two people or to a unit.
Loud marriages survive for decades. Quiet ones end. Volume turns out to be one of the least informative things in the room. What the clinician is reading is closer to a set point: how much warmth is present relative to how much cold, and whether either person is still reachable once the temperature drops.
Almost all of the research below traces back to one lab, John Gottman’s at the University of Washington, and it has been challenged on exactly the grounds you’d expect. That argument is worth having, and it’s at the bottom of this piece. The signals themselves are still the ones therapists watch for.
1. The first ninety seconds of a complaint carry more information than the hour that follows
There is a version of “we need to talk about the credit card” that is a request, and a version that is an indictment. Same topic, same amount of frustration behind it, completely different opening.
Therapists call the second one a harsh startup. It usually arrives with a “you always” or a “you never” or a sigh with some sharp edges on it.
Sybil Carrère and Gottman coded the conflict conversations of 124 newlywed couples in three-minute blocks and followed them for six years. Seventeen of those couples divorced, and every one of them had opened their discussion with more negative emotion and less positive emotion than the couples who stayed together. The first three minutes were enough to predict the six-year outcome.
For the wives in that sample, the opening three minutes were as predictive as the full fifteen. Nothing said afterward improved the picture.
Which is a strange and slightly unfair thing to know. Most of us think of an argument as something that gets decided by the end. It appears to be mostly decided by the beginning.
2. Criticism is survivable, but contempt lands as a verdict on the person
Criticism says you did something careless. Contempt says you are the kind of person who does careless things, and adds a little flourish to make sure it registers.
The delivery system is usually nonverbal. An eye roll. A short laugh in the wrong place. The word “obviously” with a half-second of lag in front of it.
Contempt is one of the four behaviors Gottman labeled the Four Horsemen, alongside criticism, defensiveness and stonewalling, and his lab reported that this cluster reliably predicted divorce.
It’s worth being precise about what the data actually showed, though. In Gottman and Levenson’s 1992 coding of 73 couples, the disgust-and-contempt code by itself did not separate the two groups significantly. What did separate them were defensiveness, stubbornness and listener withdrawal.
Contempt earns its reputation mostly by being the easiest to see. It belongs to a family of behaviors that all accomplish the same thing, which is making the other person unreachable for the remainder of the conversation.
More Bolde Stories
3. A small joke that lands mid-argument is doing more work than the apology afterward
Somewhere around minute four, one of them says something faintly ridiculous. A voice. A callback to an old family joke. A flat “okay, I’m being insane about the dishwasher.”
What the therapist is watching is not the joke. It’s the other person’s face.
Gottman, Janice Driver and Amber Tabares built a coding system specifically for these moments and found that the most effective repairs were pre-emptive ones made in the first three minutes, before the conflict had fully escalated. The repairs that worked were affective: humor, affection, self-disclosure, taking responsibility for a piece of it, some version of “we’re okay.”
Appeals to reason did much less. Being right is not a repair attempt.
And a repair only counts if it’s accepted. A couple where one person keeps reaching and the other keeps declining looks, on paper, like a couple with one difficult member. In the room it reads as something further along than that.
4. One of them can actually be moved off a position, and it happens in front of you
Moved out loud, mid-conversation, in a way that costs something. The concession has to land somewhere the other person can see it.
This was the finding that surprised Gottman’s own team in 1998, when they followed 130 newlyweds for six years and discovered that active listening, the paraphrase-what-you-heard technique that had been standard practice for years, barely predicted anything.
What predicted a great deal was whether the husband let his wife’s position change his. “We found that only those newlywed men who are accepting of influence from their wives are winding up in happy, stable marriages,” Gottman said in the University of Washington’s announcement of the study.
That result has not held up everywhere. Hyoun Kim, Deborah Capaldi and Lynn Crosby tested the same models on 85 married or cohabiting couples from at-risk backgrounds in 2007, and men’s rejection of influence did not predict whether the relationship survived.
The broader version of the signal is less gendered and probably sturdier anyway. Somebody in the room has to be movable. If both people arrive with a fixed position and leave with the same one, the fight was never a negotiation.
5. The room needs roughly five warm moments for every cold one, and the imbalance is visible from across a table
This is the number therapists actually carry around in their heads.
Gottman and Levenson coded fifteen-minute conflict conversations from 73 couples and sorted them by whether positive behaviors consistently outran negative ones. The couples they called regulated ran a positive-to-negative ratio of about five to one. The nonregulated group ran closer to one to one, and in some cases below it.
Four years later, 71% of the nonregulated couples had considered ending the marriage, against 33% of the regulated ones.
The important part is what counted as positive. All of it was mundane. Assent. A neutral description of the problem. A laugh. Agreeing that one small piece of the complaint is fair. Which is why a couple can be living inside a five-to-one marriage and have no idea.
The authors were candid that some of this analysis was post hoc and needed cross-validation on new samples.
More Bolde Stories
6. One of them going quiet is not calm, and the body usually gives it away
Stonewalling looks like the most reasonable behavior in the room. One person is escalating and the other has simply stopped participating.
Inside the quiet one, the reading usually runs hot. Withdrawal tends to arrive downstream of arousal, and the giveaway is physical: shallow breathing, a jaw that isn’t moving, a hand very still on the arm of the chair.
In that same 1992 data set, listener withdrawal was one of the sharpest behavioral differences between the two groups, running roughly twice as high in the nonregulated couples.
The physiological picture was thinner than expected. Most of the group differences in heart rate and skin conductance failed to reach significance, and the clearest effect was that wives in nonregulated marriages had faster heart rates and more vasoconstriction than wives in regulated ones.
The stronger version of that finding came earlier, from Levenson and Gottman’s 1985 study of 30 couples, where higher physiological arousal during a lab conversation predicted how much marital satisfaction dropped over the following three years. Small sample, and only 19 couples were relocated for the follow-up, so hold it loosely.
Still, a clinician can see the difference between a person who is listening and a person who has left. So can a spouse, usually.
7. Small remarks get answered, including the ones that don’t deserve an answer
Half of what gets said in a therapist’s waiting room is content-free. Look at this. Did you see the time. That’s the same car we rented.
These are bids, and their only real function is to ask whether the other person is still available.
In the apartment-lab newlywed research, couples still married six years later had turned toward each other’s bids about 86% of the time. The couples who had divorced were at roughly 33%.
What makes this observable in ten minutes is that couples don’t perform it. Nobody arrives at counseling having rehearsed how they’ll respond to a throwaway comment about the parking.
8. The story of how they met still gets told like a good story
Every couples therapist asks some version of this. Rapport is the smaller reason.
Kim Buehlman, Gottman and Lynn Katz interviewed 52 married couples about their history: how they met, how they decided, the good years, the hard ones. Then they coded the interviews for warmth, negativity, and how much each person seemed to be enjoying the telling.
Three years later, those interview codes classified divorced and intact couples with about 94% accuracy, while self-reported marital satisfaction at the same time point predicted nothing at all.
The single strongest predictor in the whole set was disappointment. Anger came nowhere close. What carried the weight was the flat, slightly embarrassed tone of someone describing a marriage that didn’t turn out to be what they’d pictured.
Its counterpart on the good side was fondness: whether a spouse could still produce a compliment, an ounce of pride, a specific detail about the first meeting that clearly still pleased them.
A couple in trouble can usually recite the facts of how they met. The charm is the part nobody can perform on request.
More Bolde Stories
9. The worst year of the marriage gets described as something they went through together
Ask about the hardest stretch and you get one of two shapes.
One is a list of things that happened. The layoff, the miscarriage, the year with his mother in the spare room. Delivered as weather.
The other has a “we” running through it, and a note of something close to pride. The Buehlman study coded this as glorifying the struggle, and couples who did it were significantly less likely to have separated three years on. Couples who described their lives as chaotic and beyond their control, without that thread, were more likely to have gone.
Same events either way. The hardship does no work on its own.
What the therapist is listening for is whether two people built a shared account of the bad years, or whether each of them privately survived the same period alone in the same house.
Ten minutes is a real signal and a bad verdict
Now the argument this whole field has been having for twenty-five years.
The famous accuracy figures, the 91% and the 93.6% and the 94%, come from statistical models fitted to the same small samples they were then tested on. Richard Heyman and Amy Smith Slep ran the numbers in 2001 and showed that when you cross-validate those equations on fresh data, and account for how uncommon divorce is in any given short window, accuracy and predictive value drop sharply. Their conclusion was that impressive first results should be read with extreme caution.
That is a real hit, and it lands squarely on the numbers. The behaviors come through it intact. Harsh startups, contempt, refused repairs and withdrawal show up as risk markers across many labs. What doesn’t survive scrutiny is the claim that any of them yields a reliable forecast for one particular couple.
Which is roughly what good clinicians say when you push them. The ten-minute read isn’t a prediction. It’s a description of the pattern currently running in the room, and patterns are the one thing in a marriage that can be changed on purpose.
Nobody’s marriage ends because of an eye roll. It ends because the eye roll went unremarked for nine years, and then somebody stopped making the joke, and then somebody stopped answering the comment about the parking.
All nine of these are visible from the outside in under an hour. Most of them are visible from the inside on any ordinary Tuesday, to anyone who decides to look.
