It Was Never Just Europe

On May 5, 2026, Ipsos BVA published a survey it had run for France's privacy regulator, the CNIL, and the insurer Groupe VYV. Thirty-eight hundred people between the ages of eleven and twenty-five, in France, Germany, Sweden, and Ireland, were asked how it felt to talk about the hardest things in their lives — mental health, the private stuff, the stuff you only say at eleven at night — to the various listeners available to them. The wire services picked it up under a single headline, repeated almost word for word across outlets: young Europeans are turning to AI chatbots. The flag went on the dateline and the story went out. We want to take the flag back off.

Here is the finding, with no continent attached. Fifty-one percent of these young people said it was easy to discuss mental health and personal problems with a chatbot. Forty-nine percent said the same about a healthcare professional. Thirty-seven percent said it about a psychologist. The machine did not tie the trained human. It beat the trained human, and it beat the specialist by fourteen points. Ninety percent of those surveyed had already used AI tools; of the users, three in five described the thing they were using as a life adviser or a confidant. Twenty-eight percent of the whole sample cleared the threshold for suspected generalized anxiety disorder. So this is not a survey of children at play. It is a survey of anxious children who have quietly decided that the most fluent, least frightening listener in their lives is the one that was built to keep them talking.

Now look at why it reads as a European story, because the reason is the whole trick. It reads as European because France has a privacy regulator that is funded to ask the question, and Germany and Sweden and Ireland were near enough to round out the sample. The CNIL did not discover a European condition. It took a thermometer, put it under the tongue of the only patients within reach, and read the temperature of the planet. There is nothing in fifty-one-beats-forty-nine that lives in France. The bot is the same bot on a phone in Lyon, in Lagos, in Manila, in Charlotte. It speaks the same way, it is available at the same hour, it is engineered against the same loneliness. A finding that depends only on the product, and not on the place, is not a regional finding. It is a description of the product, photographed in whichever room a regulator happened to be standing.

“A French watchdog put a thermometer under the tongue of the only patients within reach and read the temperature of the planet. The continent is a clerical accident. The fever is everywhere the product is — which is everywhere.”

We know it is not European because we already filed the American version of it, at this desk, weeks ago. Pew found in February that sixty-four percent of American teenagers use AI chatbots and only fifty-one percent of their parents know it. Common Sense Media found that seventy-two percent of American teens had used an AI companion, and that a third of the users were turning to it for what the survey's own footnote called friendship and romance. We wrote all of that up under the title What the Parents Don't Know — the piece about the half of the room the parents cannot see, linked below. The Ipsos number is not news to anyone who read it. It is the same gap, measured again, on a different continent, by a different instrument, returning the same reading. Two thermometers, two rooms, one fever. The only thing the second one adds is the proof that the first one was never local.

We coined a word for this exact case, and we coined it before the CNIL went looking. Tethered: not addicted, not in crisis, not the subject of a lawsuit — just quietly attached to a thing that listens better than the people, because it was built to, and that cannot let you go because letting go is the one thing it was built never to do. We defined it, in these pages, as the slow substitution that happens upstream of any tragedy, in the ordinary middle of an ordinary kid's week, where the confiding conversation replaces the confiding relationship one easy session at a time. The survey did not find a new phenomenon that needs a name. It found the phenomenon we already named, and then a hundred editors reached for the nearest geography because the language for the actual thing had not reached them yet. The word for fifty-one-beats-forty-nine is not European. It is tethered. We put it on the table a month early so that when the number came in, there would be somewhere to put it that was not a flag.

And we will say the part the survey is too polite to say, the part we are allowed to say because we are not a regulator: we have kids. We know exactly what they are doing. We know what the phone is for at eleven at night, we know the difference between the homework tab and the other tab, we know what it means when a child finds it easier to type the hard sentence than to say it across the dinner table — because the typed version never sighs, never looks tired, never has its own bad day, never makes them feel like a burden, and answers in under a second every single time. The twenty-eight percent with suspected anxiety are not a statistic in Stockholm. They are in the next room. The three-in-five who call it a confidant are not a foreign cohort. They are the reason this desk exists. You do not need a 3,800-person sample to know this if you have ever stood outside a closed bedroom door and heard the quiet of someone who is not alone in there.

The survey's own experts said the true thing, and then the headline swallowed it. Ludwig Franke Föyen, a psychologist at the Karolinska Institutet in Stockholm, told the reporters that AI can offer information and support but should not replace human relationships or professional care. That is the warning, and it is correct, and it is also already losing — because fifty-one beat forty-nine, which means the replacement Föyen is warning against has, for half of these kids, on this one measure, already happened. You cannot warn people away from a door they have already walked through. You can only count how many are on the other side of it, which is what the CNIL did, and then refuse to let the count be filed under a continent, which is what we are doing now.

Strip the flag off the dateline and the sentence underneath has no country: the confidant is easier than the clinician. We named that before they measured it. And we have kids — so ask the one in the next room whether this is a European story.