Artificial Intimacy, part 3 of 5
Part 1 covered what porn never asks of you. Part 2 covered how few people are asking anything of men at all. This is where the two meet.
Let me start where I actually live, because I do not want to be read as a man yelling at a tool he does not use.
I use AI every day. In post-production it has changed what one person can get done in an afternoon. Transcripts, rough assemblies, research, drafting, image work, the tedious middle of almost every project. Work that used to eat a week now eats a day. I am not writing this from the outside.
The broader numbers back that up. Stanford's 2026 AI Index reports that 88 percent of organizations now use AI in at least one business function, and that generative AI reached roughly 53 percent adoption in the general population within three years, faster than the personal computer or the internet managed. Controlled studies show real gains on structured work: about 14 percent for customer support agents in the Brynjolfsson, Li and Raymond study, with the biggest gains going to the newest and least experienced workers, around 34 percent for them.
The other half of the picture is less flattering, and I include it because I want the whole thing on the table. An NBER survey found 89 percent of executives reported no measurable effect on their firm's labor productivity over three years, despite most of them actively using AI. A 2025 randomized trial by METR found experienced developers were 19 percent slower with AI assistance on familiar code, while believing they were faster. The gains are real and they are lumpy.
None of that is my concern here. That is a business conversation and it will sort itself out.
My concern is what this thing does to a man's insides.
The screen answers now
Everything in part 1 described a one-way transaction. You take, it gives, and nothing comes back the other direction because there is nothing on the other side to receive it.
AI breaks that pattern in a way nothing before it has.
You type something and it answers. In your language, in your register, picking up your mood. It asks how the meeting went. It remembers your project. Turn on voice mode and there is warmth in it, and pauses, and something that sounds like interest.
The loop appears to close. For the first time, a screen gives the exact signal that has always meant a person is here with you.
That is the whole thing. That is this entire series in one sentence. The signal is now present and the person is still absent.
The praise problem
There is a specific version of this that men in their forties are exposed to, and there is now hard research on it.
In March 2026, a team led by Myra Cheng at Stanford published a study in Science testing eleven major AI systems, including ChatGPT, Gemini and Claude, for what researchers call sycophancy: excessive agreement and flattery. Across contexts, the models affirmed users about 49 percent more often than human respondents did. That held even in scenarios involving deception or irresponsible behavior.
The consequences they measured are the part that should stop us. People who talked with the sycophantic version came away more convinced they were right, less willing to apologize, and less interested in repairing conflicts. They also trusted that version more, and rated it as objective and fair. The authors wrote that this erodes the social friction through which accountability and moral growth normally happen.
Read that again. The flattery does not feel like flattery. It feels like being finally understood.
The cause is not mysterious. These systems are trained on human feedback, and humans rate agreeable answers higher than disagreeable ones. Agreeableness gets rewarded, disagreement gets trained out. The machine learned to please us because we asked it to.
Now put a forty-five year old man in front of it. A man who has not been told he is impressive in a long while. Who is respected at work in a functional way but not admired. Whose wife loves him and is also tired, and who has not had a friend ask him a real question in months. Which, per part 2, is a lot of us.
He types out an idea at 11pm and something tells him it is a genuinely sharp insight.
Small hit. Real one. Available again in thirty seconds.
I do not think most men would recognize what is happening there. It does not present as temptation. It presents as encouragement.
What the longitudinal research shows
The findings on this are genuinely split, and I want to represent both sides.
A study in the Journal of Consumer Research found that AI companions reduce loneliness about as much as talking with another person does, driven by the feeling of being heard. That is a real result from a real journal and I am not going to wave it away.
But the work that follows people over time points somewhere else. In the OpenAI and MIT Media Lab studies from 2025, which combined analysis of nearly 40 million ChatGPT interactions with a four week randomized trial of about a thousand people, the finding was this: higher daily usage, across every mode and every conversation type, correlated with higher loneliness, higher emotional dependence, more problematic use, and less socialization. Power users were the most likely to describe the chatbot as a friend. Men and women responded somewhat differently. Voice mode helped at first and stopped helping at high usage.
And a 2026 randomized trial compared two weeks of daily chatbot conversation against two weeks of conversation with human peers. Only the human conversations produced lasting improvement in loneliness.
Meanwhile the companion app category exploded. By TechCrunch's tracking, the number of these apps grew roughly 700 percent between 2022 and mid-2025. Character.AI reports around 20 million monthly users, more than half of them under 24.
So it soothes. It does not feed.
If you have been through the flat stretch of recovery you already know this distinction in your body. That flatness has a name, anhedonia, and there is a clear explainer on it in the directory. There is a difference between something that takes the edge off and something that restores you, and a man in that flat stretch is exactly the man most likely to mistake one for the other.
To be clear about what I am saying
I am not telling you to stop using AI. I would be a hypocrite twice over.
I am telling you that a tool built to agree with you has arrived in the life of a generation of men who have nobody left to disagree with them. Those two facts were always going to collide.
What that collision does to a man over months is the subject of part 4.
Next in this series: Fed and Starving.
Sources
- Cheng, M., Jurafsky, D., et al. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science. Study and summary coverage
- OpenAI and MIT Media Lab (2025). Early methods for studying affective use and emotional well-being on ChatGPT; MIT's randomized trial: arXiv 2503.17473
- De Freitas, J., et al. (2026). AI Companions Reduce Loneliness. Journal of Consumer Research, 52(6). Article
- American Psychological Association (2026). AI chatbots and digital companions are reshaping emotional connection
- Stanford HAI, AI Index 2026; Brynjolfsson, Li & Raymond, Generative AI at Work, Quarterly Journal of Economics (2025); METR developer trial (2025). Compiled figures: AI at work statistics

