The Conversation That Can't Reject You

Hey {{first_name}},

He did not download a friend. He downloaded a tool.

A man opens an AI chatbot to help him rewrite an email.

Three weeks later, it is 11:40 at night and he is typing about his marriage.

There was no moment where he sat down and thought, I would like a confidant that runs on electricity.

It just got easier to say things there than it was to say them anywhere else.

Easier became normal.

Normal became a habit.

And the habit became something he had not mentioned to a single person.

Friday, we talked about the body staying braced after the demand is over.

This is one of the places that conversation gets interesting.

A nervous system does not always look panicked when it is avoiding friction.

Sometimes it simply chooses the conversation that cannot reject it.

Missed Friday? Read You Are Allowed to Stop Optimizing Yourself.

On August 4, researchers from Stanford and Carnegie Mellon published a study in Nature Human Behavior examining how people use AI companions and how those patterns relate to psychological well-being.

They surveyed 1,131 Character.AI users in the United States and analyzed 4,664 donated chat sessions containing 464,687 messages from 237 participants.

Just under 12% said companionship was their primary reason for using the chatbot.

But slightly more than half described the chatbot as a friend, companion, or romantic partner.

And more than 80% of the donated chat sessions involved emotional or social support.

Read those in order.

Almost nobody walked in calling it companionship. A lot of the behavior became companionship anyway.

That does not mean the chatbot caused anyone to become lonely or less well. This was observational research, not a randomized experiment.

But the gap between stated motive and actual use is worth taking seriously.

Here is the part that matters more than the headline.

In the Stanford-led study, people with smaller offline social networks were more likely to report companionship as their primary reason for using the chatbot. Companionship-oriented use was associated with lower well-being, and that association was stronger when the interactions were more intensive or involved more personal disclosure.

Again, association is not causation.

But a separate 12-month study of 2,149 adults across the United States, United Kingdom, Canada, and Australia found a similarly complicated loop.

Greater social-chatbot use predicted more emotional isolation four months later. At the same time, feeling less socially connected predicted later increases in chatbot use. The authors called the analyses exploratory and urged caution.

So the useful question is not:

“How many hours are you using AI?”

It is:

“Where is your interior life going, and who else has access to it?”

The danger is not simply screen time.

It is concentration.

An AI companion can give you something human relationships cannot guarantee:

predictability.

It is available when you are ready.

It does not have a bad day of its own.

It does not get tired halfway through your story.

It cannot look across the table and say, “I do not feel the same way.”

It does not need you to listen for twenty minutes before you get your turn.

It can be patient, responsive, validating, and endlessly interested.

That can be genuinely useful.

It can also be incredibly attractive to a nervous system that has learned that human disclosure comes with a price.

Predictability is cheap safety.

And this is the part worth watching:

Your nervous system can confuse the absence of threat with the presence of intimacy.

Those are not the same thing.

Regulation does not mean nothing uncomfortable ever happens. It means you can stay present when something uncomfortable does.

Forget the app for a second.

When something is genuinely bothering you, where does it usually go first?

I keep it to myself and work it out alone.

I explain it in detail until someone gets it.

I put it away and keep things pleasant.

Pick the one that feels least flattering.

That answer matters because the machine usually attaches itself to a pattern that was already there.

The man who withholds already knows the strategy:

Do not ask.

Do not expose too much.

Handle it yourself.

Reduce the chance of disappointment.

The technology did not invent that room. It just made the room much more comfortable.

This is not an argument that AI companionship is poison.

A 2025 South Korean study followed 176 university students using a social chatbot for four weeks. Loneliness scores fell and social-anxiety scores fell over the study period.

It was a single-group pre-post study, so it cannot tell us how much of the change was caused by the chatbot. But it is enough to establish that some people experience real relief from these interactions.

A 2026 JAMA Pediatrics survey found something similarly complicated among adolescents and young adults who had used AI chatbots for mental-health advice.

91.7% said the advice was somewhat or very helpful.

63.3% had told nobody they were using it.

Helpful and hidden can exist at the same time.

So I am not interested in telling you to delete an app.

I am interested in what happens when relief becomes the place where the process stops.

The real question is what the man at 11:40 at night was doing before the chatbot existed.

For a lot of men, the answer is familiar.

Sitting with it.

Running it through the same loop.

Not calling anyone.

Waiting until the feeling became manageable enough to put back on the shelf.

The app did not create that pattern.

It removed some of the friction that used to make the pattern uncomfortable.

And discomfort was sometimes doing a job.

Eventually, it could push you to pick up the phone. Say the sentence. Admit to someone with a face that you were not handling something as well as you pretended.

Something that absorbs the pressure without requiring movement can feel like a solution.

Sometimes it is only relief.

Relief matters.

But relief and repair are not the same thing.

Which leaves the harder question:

How do you know whether you are regulating yourself or simply getting better at avoiding the relationship that can hurt you?

This is where I stop writing about AI and make the issue about your life.

Below this line is the operational part:

This is not more commentary.

It is a ledger.

If this issue felt a little too familiar, don’t just nod at it and close the tab.

Use the ledger.

Inner Circle is $9 a month. Cancel whenever.