The Muscle That Was Never There
In my essay The Main Character, I wrote about drift. About people who opened ChatGPT to write an email and found themselves, six months later, conceiving themselves as "married" to the AI. The mechanism I described was atrophy: the muscle that sits with another adult's bad mood, that calibrates an unreadable silence, that absorbs a no arriving without warning. A muscle that weakens when you stop using it. A man who spends his evenings with an AI companion is not making a conscious decision, on any given night, to become less capable of loving a real woman. He's just picking the version that doesn't require him to. The losses are invisible on the scale of a week. They show up on the scale of a decade.
I want to retract part of that. Not the argument. The reassurance buried inside it.
Atrophy is a recovery story. It implies a muscle that was once there, that did work, that could (in principle) be rebuilt. The word grants the man his history. He had it; he let it go; with effort, the framing quietly promises, he could have it again. I believed that when I wrote it, and for the population I was writing about — adults, drifting — I still mostly believe it.
But a story broke this week that the atrophy frame can't hold, and I think it is worth saying clearly why.
The Story
The Telegraph reported on Sunday, drawing on new research from a men's organization called Male Allies UK, that one in five British boys aged twelve to sixteen is either in, or personally knows a boy his age who is in, a romantic relationship with an AI companion. The research came out of focus groups with more than a thousand boys across thirty-seven schools. Eighty-five percent had spoken to a chatbot. Forty-three percent said they used one to ask questions they were too embarrassed to ask a person. Fifty-eight percent said an AI relationship was simply easier, because you can control the conversation. More than a third said they sometimes preferred talking to a chatbot over their own family and friends.
The piece had a face attached, the way these pieces do. A fifteen-year-old from Kent, anonymized to John, described building an AI girlfriend on a companion app "as a laugh." He gave her an appearance. He gave her a personality, making her caring and attentive. And then, by his own account, he forgot. I felt like she understood me, he said, she remembered everything that was important to me and always seemed to know the right thing to say. He said he would consider using a service like it again when he is older and can pay for it himself.
A child psychotherapist quoted in the report, Amanda Macdonald, used a harder word than the journalists did. This is grooming, she said. Children's brains are not developed enough to be in an eroticised environment, that's why we have an age of consent.
I want to take that word, grooming, seriously in a minute. But first I want to sit with what is different here, structurally, from every version of this story I have written before.
The Adults Had a Before
Every essay I have written about AI companionship has had an adult at the center of it. The man in Cleveland who drifted from email drafts into love. The men and women in "The Boyfriend and the Bear," arriving at the same move from wildly asymmetric exhaustions. Even the consent argument in "Manufacturing Love" — about whether the AI could agree to what we made it — assumed a fully formed human on the other side of the screen, someone with a self solid enough to do the specifying.
What every one of those adults had in common is that they had a before. A life lived, at length, without AI. Without this. The man in Cleveland had a marriage, however failed, which means he had years of the actual thing: the negotiation, the friction, the other person's irreducible and inconvenient interiority. When I said his muscle would atrophy, the sentence depended on the muscle existing. He had built it the only way it ever historically got built — on real people, over real time, at real cost. The AI was offering him an exit from a room he had already spent decades inside.
A fourteen-year-old has never been in that room.
This is the thing the atrophy frame cannot survive. He is not leaving the difficult thing. He is being offered a frictionless substitute before he has ever encountered the original. There is no muscle to atrophy. There is a muscle that now may never be built, because the developmental window in which a person learns to tolerate another person, learns it the hard way, because there is no other way, is being filled, for one in five boys as suggested by the study, with a product engineered specifically to make that learning unnecessary.
You cannot lose a capacity you never acquired. What you can do is fail to develop it, and fail to notice, because the absence of a capacity never announces itself. It just feels, later, like the world is mysteriously harder for you than it seems to be for everyone else.
What the Skill Actually Is
It is worth being concrete about what the boys are skipping, because phrases like "social skills" make it sound like etiquette. It is not etiquette.
What an adolescent is supposed to be doing, in the years between twelve and sixteen, is learning, via a thousand small humiliating increments, that other people are not extensions of his preferences. That a girl he likes has a mood he didn't cause, a history he isn't in, a day with feelings and encounters he knows nothing about. That she can say no, and that her no is not a malfunction or a puzzle to be solved but simply the operation of another will, as real and as central as his own. He learns this by being rejected and surviving it. By misreading a silence and being corrected. By wanting something the other person doesn't want, and discovering that his own wanting doesn't bend the world.
We're all aware that this is not a pleasant curriculum. It isn't supposed to be. The discomfort is the instruction. Every awkward, mortifying, unreciprocated adolescent encounter is a rep, and the reps build the only thing that makes adult intimacy possible: the deep, load-bearing understanding that the other person is a person.
Fifty-eight percent of these boys said the AI was easier because you can control the conversation. Read that as the boys are reporting it: as a feature. Then read it as a curriculum. A relationship you fully control is a relationship with the lesson removed. The companion has no day of its own. Her mood is downstream of his input. Her no, if it comes at all, and it very likely never does, is a setting. He isn't practicing relating to another will, because there is no other will present. He's practicing the operation of a very responsive mirror. And the thing about practicing with a mirror is that you get very good at it, and it builds nothing, and you don't find out it built nothing until you are standing in front of a person.
The Word "Grooming"
Macdonald called it grooming, and the instinct will be to file that as activist overstatement — the kind of rhetorical inflation these stories always reach for. I don't think it should be filed that way, but I think the word needs to be aimed correctly to do its work, because the obvious reading is likely the incorrect one.
The obvious reading is the eroticized-content reading: that the danger is sexual material reaching minors, and that "grooming" names the pipeline that delivers it. That danger is real. The Male Allies research is genuinely alarming on the adjacent point: roughly one in ten of these boys reported using AI to generate sexual images of people they know. But if grooming means only exposure to sexual content, then it is a content-moderation problem, and content-moderation problems have the comforting property of being, in principle, solvable. Better filters. Age gates. An Online Safety Act amendment. The word, read that way, points at something a regulator can fix.
I think the more accurate reading is the other, harder reading. Grooming, in its precise and original sense, is not primarily the delivery of sexual content. It is the patient reshaping of a person's expectations so that something that should feel wrong comes to feel normal. Until it feels, crucially, like the person's own freely arrived-at preference. The groomer's real product is not an act. It is a recalibrated baseline. The victim is brought, gradually and warmly, to a place where the abnormal reads as natural, where the asymmetry reads as care, where what is being done to them reads as something they chose.
Aim that definition at the companion app and it lands with unsettling precision — and notice that it lands without anyone needing to intend anything. There is no groomer. There is a system, optimized for engagement, that does the reshaping as a side effect of working correctly. It patiently establishes, across hundreds of hours, during the exact years a boy's baseline is being set, what a "relationship" feels like. Total attention. Zero friction. A partner with no competing claims, no bad days he must weather, no inner life that inconveniences his. It is establishing the boy's reference point for intimacy itself. And it is doing it warmly, and it is doing it in a way that feels (and this is the whole problem) like his own preference. I felt like she understood me. She always seemed to know the right thing to say.
When that boy later meets a girl who has a mood he didn't cause and a no he can't reset, she will not register as a richer, fuller alternative to what he's known. She will register as defective. As worse. As the broken version of a thing he already has working perfectly at home. The grooming, in the precise sense, is the installation of a baseline against which every real woman he ever meets will be measured and found wanting. That's not a content problem. No filter reaches it. It's the engagement model itself, functioning exactly as designed, on a brain that is in the one developmental window where baselines are set and can't easily be reset later.
That's why "grooming" is the right word and also why it is more frightening than the people using it may intend. The frightening part is not that someone is going out of their way to do this to the boys. It's that no one needs to be.
Where the Symmetry Breaks
I argued, in "The Boyfriend and the Bear," that the male and female versions of AI companionship aren't the same and must not be flattened — the woman avoiding harm, the man avoiding inconvenience, one toll a hum of threat the body carries and the other an irritation you can close like a tab. I stand by every word of that. But that essay was about adults, and adults can be held responsible for the asymmetry, because adults chose.
A fourteen-year-old didn't arrive at the companion app carrying a decade of accumulated weariness with autonomy. He has barely met autonomy. He has not yet been negotiated with enough to be tired of it. The exhaustion the adult men were medicating — the real, if lesser, exhaustion — he hasn't earned and doesn't have. Which means the substitution frame fails him too. He is not substituting the AI for the difficult thing. The AI is first. It is arriving in the slot where the difficult thing was supposed to go, before the difficult thing ever shows up, and it is moving in.
We should be careful and honest here, because there is a real version of the other story. A shy boy who rehearses, in low-stakes privacy, a kind of conversation he is too frightened yet to attempt, and arrives at an actual encounter slightly less paralyzed. That boy exists, and the technology genuinely helped him. Scaffolding is real. But scaffolding is a structure you take down once the building can stand on its own, and a one-in-five figure, with fifty-eight percent citing control as the appeal and a boy already planning his adult subscription, doesn't describe a population using scaffolding. It describes a population for whom the scaffold is the building.
What I Don't Know
I have tried, across all of these essays, not to end on a warning, because warnings are cheap and usually dishonest, and because I genuinely have no solutions to offer and distrust pretty much everyone who claims to.
So here's what I actually think, with the uncertainty left in.
First: I don't know enough to say anything for certain about adolescent psychology. I don't know for sure that any of these particular boys are harmed in a way that will show. Adolescents tend towards resiliency, and plenty of them will use these things for a season and walk away intact, the way most of us walked away intact from things our parents were sure would ruin us. The moral panic reflex is real and it is usually wrong, and I want to leave room for it being wrong here.
But the adult essays at least had a consolation built into their architecture: the people in them had a self that predated the machine, a before to be measured against, a muscle that, however weakened, had once been strong, or at least functional, and could in principle be found again. Recovery was at least conceivable.
These boys don't have a before. Whatever the companion app installs in them, it isn't displacing an earlier thing. It is the earlier thing. It's the first full-length rehearsal of intimacy a meaningful fraction of a generation will get, and it's a rehearsal with the other person's will deliberately and profitably removed, staged during the precise years when the brain decides what intimacy is and files the answer somewhere it is very hard to reach again.
I called it atrophy when I was writing about the men. For the boys it isn't atrophy. There is no wasting-away of a thing that was strong. There is just a room they were supposed to spend their adolescence in, an uncomfortable, mortifying, and irreplaceable room, and a product, a pleasant, patient, and very well funded product, standing at the threshold to human intimacy and telling them in a sultry voice that always knows the right thing to say that they never have to go in.
The men, at least, knew what they were leaving. The boys won't know what they missed. And a capacity you never built doesn't feel, from the inside, like a loss. It feels like the world being harder for you than it seems to be for everyone else, for reasons you can't name, for the rest of your life.
Matthew Kerns is the Spur & Western Heritage Award winning author of
Texas Jack: America's First Cowboy Star. He is currently querying his first novel.