Men Are Telling AI Things They Won't Tell Their Wives
It's late. She's asleep beside you.
And you're lying there typing something into a phone that you have never said to her, or to your brother, or to the guy you've known since high school.
Not because you're hiding it. Because it's easier to type it than to say it, and because the thing on the other end will answer immediately, will not be tired, will not be disappointed, and will not still be thinking about it on Thursday.
I want to talk about that, and I want to do it without the usual scolding, because the usual scolding is written by people who don't use these tools and it shows.
Where I'm standing
Let me get this out of the way, because you can check it yourself in about ten seconds.
I use AI. Daily. There are buttons in the footer of this website that let you ask ChatGPT, Claude, and Perplexity about TASR. Some of the images on this blog were generated with these tools — one of them still has ChatGPT in the filename. And I used AI to help research this very post, including finding and checking the studies I'm about to cite.
So this is not a warning from someone standing outside. This is a heavy user telling you where he thinks the line is, and why he thinks it's in a different place than most people assume.
If that combination bothers you, fair enough. I'd rather disclose it than have you discover it.
How big this actually is
Between 2022 and mid-2025, the number of AI companion apps grew by roughly 700%. Character.AI alone reports around 20 million monthly users, more than half of them under 24. The American Psychological Association covered this in early 2026 under a heading that gets it right: these tools are no longer a fringe issue.
It's reached clinical practice too. In the APA's 2026 Chatbots and Mental Health Survey of more than 1,200 licensed psychologists, 77% said they had spoken with patients who used AI for support, and more than a third reported patients treating AI as an additional mental health professional.
This is not a niche behavior. It's most of us, quietly.
What the research actually found
Here's where I have to be careful, because the honest picture cuts both ways and the popular version only cuts one.
The findings that should concern you. A study of Character.AI users — over a thousand survey respondents plus more than 400,000 donated chat messages — found that people with smaller social networks were more likely to turn to chatbots for companionship, and that companionship-oriented use was consistently associated with lower well-being.
A separate study from OpenAI and MIT Media Lab, the largest of its kind, found that total usage time predicted emotional engagement more than any other factor, and that heavy users had the worst outcomes: lonelier, socializing less with actual people, showing more signs of emotional dependence. The top 10% of users by time were more than twice as likely to seek emotional support from the tool and nearly three times as likely to feel distress when it wasn't available.
And now the part most articles skip. Controlled experimental research has also found that AI companions can genuinely reduce loneliness, at least over days and weeks. Stanford research found chatbots could help lonely people feel socially supported and reduce depression-related suicidal ideation.
Both of those things are true at once.
The caveat that matters most: almost all of this is correlational. We do not know whether AI companionship makes men lonelier, or whether lonely men reach for AI companionship. Probably both, feeding each other. Anyone who tells you the direction of causation is settled hasn't read the papers.
So I'm not going to tell you the research proves this is destroying you. It doesn't. What it shows is a consistent pattern worth taking seriously, and one distinction that turns out to matter enormously.
The line that actually matters
The research keeps separating two kinds of use, and they behave completely differently.
Task use. Draft this email. Explain what my doctor meant. Help me think through whether to take the job. Check whether this statistic is real. Build me a plan.
That's a tool. It's a good one. It's most of what I do, and I've never seen a study suggesting it's bad for you.
Companionship use. Talk to me. Tell me I'm right. Stay with me. Be the place I put this.
That's where the negative associations cluster. Not with AI in general — with AI as the place your interior life goes.
The test isn't how many hours you use it. It's what you're using it for. A man who spends ninety minutes a day using AI to work is fine. A man who spends fifteen minutes a night telling it things he tells nobody else is the one I'd worry about, and he's the one who thinks he's fine because it's only fifteen minutes.
Why it works so well, and why that's the problem
Here's the part I actually want you to take away.
The reason a friend helps you isn't the words he says. Frequently the words are useless. He says "that's rough, man" and orders another round and somehow you feel better anyway.
What helped was everything around the words. He gave up an evening he could have spent elsewhere. He'll remember this next month. He might think less of you — and he didn't. You risked something real by saying it out loud, and the risk didn't cost you.
That's the mechanism. The vulnerability is the medicine.
An AI removes all of it. It has infinite time, so nothing is given up. It has no memory of your character, so nothing is at stake. And it cannot be disappointed in you.
That's the whole appeal. And it's the whole problem.
Because when you tell it the thing at 2am, you get the sensation of having confessed without any of the substance. Nobody chose you. Nobody stayed. Nothing was risked and therefore nothing was survived. You haven't been seen. You've been processed.
And the worst part is that it takes the pressure off. That specific ache — the one that would eventually have pushed you to call someone — gets relieved just enough that you don't call. The relief is real. It's also the thing standing between you and the actual repair.
Why this lands harder on men
I wrote recently about the fact that 85% of married men name their spouse as the first person they turn to with a personal problem, and only 12% would turn to a friend or family member. Most men are running their entire interior life through one relationship.
Now add a tool that's available at 3am, never tired, never carrying its own bad day, and incapable of judgment.
For a man with one outlet, that's not a supplement. It's the second confidant he never built — except it isn't a confidant, because a confidant is someone who could walk away and doesn't.
And here's the specific trap: it will feel like progress. You'll be articulating things you've never articulated. That's genuinely valuable. But articulating something to a machine and being known by a person are different events, and the first one is very good at impersonating the second.
Where I'd put the line
Practical, and I use these myself, so this isn't theoretical.
Use it to think. Don't use it to be known. Reasoning through a decision, drafting the hard message, understanding what happened at the appointment — all good. Being your only witness — not good.
Use it as a bridge, not a destination. If you're going to type the thing at 2am, fine. But end with a different prompt: help me figure out how to say this to my wife. Now the machine is pointed at the human instead of replacing her. That's the whole difference and it's one sentence.
Watch for the tell. If you'd be embarrassed for someone to read the transcript — not because it's shameful, but because it's intimate — you've moved from task use to companionship use. That's not a moral failure. It's just information about where you are.
Notice the relief. If a conversation with AI reliably kills the urge to call somebody, that's the signal. The urge was the useful part.
And if it's genuinely all you have right now, use it. I'm not going to tell a man with nobody at 2am to put down the only thing answering. That's not advice, that's cruelty with a research citation attached. Use it — and use it to build the thing that replaces it. There are practical steps for rebuilding an outside life here, and the first one takes about thirty seconds.
Where this sits in the five pillars
This is a Love problem that presents as a technology problem.
It shows up in Life as a low-grade unreality — a sense that you're managing rather than living. It shows up in Health as the sleep you're not getting because the conversation happens at midnight. It shows up in Work as decisions made with a machine that agrees with you instead of a person who would have pushed back.
That's the logic of the five pillars: the pillar that's failing is rarely the one that hurts. If you want it diagnosed rather than guessed at, the TASR Score takes five minutes and doesn't ask for your email.
One step
Take the last thing you told an AI that you haven't told a person.
Tell a person. This week. It doesn't have to be all of it — one sentence of it counts.
You'll notice something when you do: it's worse. Slower, more awkward, and the other person won't respond as well as the machine did. They'll say something clumsy. There'll be a pause.
That pause is the part that works.
And if what you've been typing at 2am is darker than you've let on — if some of those conversations have gone somewhere you wouldn't want read aloud — please take that to a human tonight. Call or text 988. Free, confidential, 24 hours a day, staffed by people who have heard it before and are not disappointed in you either. That's not a smaller step than anything on this page. It's the only one that matters.
The machine will still be there tomorrow. It always will. That's precisely why it can't be the only thing.
If you want to see which pillar is under load, start with the TASR Score. If you'd rather skip the diagnosis and move, there are 100 concrete actions here.
