Times Are Changing: People Are Starting to Talk to AI About Their Feelings

I build tools that help people write about their feelings. So when a careful study comes out about people getting mental health support from an AI, I pay close attention.

I also have to be careful. People who build something are not the best people to judge whether it works. I want it to work. That is the bias I have to watch for.

This is a study I read twice. The first time, I was excited. The second time, I tried to read it like someone who did not build anything. Here is what I found.

What the study did

A team in Israel ran a careful test (Shoshani et al., 2026). They found 995 university students who said they were stressed and feeling low. That is about as many people as fit in a small concert hall.

They split the students into three groups by chance. One group got to use an AI chat app called Kai for 12 weeks. One group went to in-person group therapy with real psychologists, also for 12 weeks. One group got nothing for 12 weeks. The third group is called a control group, which means a group used to compare against the others.

Then the researchers measured everyone's anxiety, sadness, and how good they felt about their lives. They measured at the start. They measured after 12 weeks. They measured again three months later.

This kind of careful test, where people are split by chance, is the strongest kind of study we have. It is called a randomized clinical trial, which means people are sorted into groups by random chance so the groups are fair to compare.

The result that made me read it twice

The students who used the AI got better at handling anxiety. That part I expected.

What I did not expect was this. The AI group did better than the group that saw real therapists too (Shoshani et al., 2026). Not just better than the no-help group. Better than the therapy group.

And the therapy group, the one with real psychologists in a real room, did not do better than the no-help group on anxiety. The two were about the same.

I read that part three times to make sure I had it right.

The honest reaction

My first reaction was relief. I build writing tools that involve AI. A study like this makes my work look good. So my first feeling was, "see, it works."

That is exactly when I have to slow down.

A study about something I want to be true is the kind of study I will read carelessly. I will skip the parts that complicate the story. I will hold on tight to the parts that support what I already believe.

So I read it again. I tried to find the parts that should bother me. There are some, and I want to be honest about them.

The next chapter is about what daily use of the AI actually looked like, and the finding that surprised me even more than the headline.

References

Shoshani, A., Gurfinkel, B., Kor, A., et al. (2026). Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance: A Randomized Clinical Trial. JAMA Network Open, 9(4), e266713.

The numbers in a study can hide what really happened. So I want to slow down here and look at what daily use actually looked like.

A real amount of writing

The students who used the AI did not just open the app and close it. They used it.

On average, each student sent about 19 messages a week to the AI (Shoshani et al., 2026). That is about three messages a day. They wrote about 1,645 words a week. That is roughly four pages of writing per week.

They used the app about three days a week. They spent about 83 minutes a week on it. That is more than an hour, less than an hour and a half. About six in every ten students were still using the app when the study ended.

This was not a quick try. This was twelve weeks of real conversation. People were writing about their feelings, often, on most days.

The finding I keep thinking about

The researchers asked the students to rate two things about the AI. How warm did it feel. How skilled did it seem.

They asked the same questions to the students who saw real human therapists.

The scores were almost the same. Warmth was rated about 3.86 out of 5 for the AI and about 3.74 out of 5 for the human therapists (Shoshani et al., 2026). Skill was rated about 3.85 for the AI and about 3.93 for humans. The differences were so small that they could be by chance.

I keep coming back to that. Not "the AI was as good as a therapist." That is not what was measured. What was measured was the feeling people had while using each one. The feeling of being heard. The feeling that the other side knew what it was doing.

People felt about the same way about the AI as they felt about the humans.

The technical name for this

There is a name for the feeling that you are working together well with a therapist. It is called therapeutic alliance, which means the sense of being a real team with the person helping you.

In this study, the researchers borrowed the name and used a new word. Digital therapeutic alliance. The same kind of working relationship, but with a tool instead of a person (Shoshani et al., 2026).

The study found that students who felt this alliance with the AI used it more. And the more they used it, the more they got better. The path was clear. Feeling close to the tool led to using it. Using it led to feeling better.

What this is not saying

I want to be careful here.

This is not saying the AI is the same as a human therapist. It is not saying the AI is alive, or thinking, or really cares. The students knew they were talking to a computer.

It is saying something more subtle. It is saying that the feelings of trust and being heard that we used to think only happened with humans, can also happen with a careful tool. The feelings are real. The other side of the conversation is software.

That is a strange new thing for our world to hold.

The next chapter is about the part of the study that made me uncomfortable.

References

Shoshani, A., Gurfinkel, B., Kor, A., et al. (2026). Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance: A Randomized Clinical Trial. JAMA Network Open, 9(4), e266713.

Here is the part of the study I had to read slowly.

The drop that should give us pause

At the start of the study, about three out of every ten students in each group said they wanted to see a therapist for help.

Twelve weeks later, in the AI group, that number had gone down. Fewer students felt they needed a therapist (Shoshani et al., 2026).

In the no-help group, the number went up. They had not gotten better, so more of them wanted help.

In the human therapy group, the number stayed about the same.

So far, this might sound fine. People got better. They felt they needed less help. That is the good story.

But there is another story. The researchers themselves point this out. The drop in wanting therapy could mean people are feeling stronger and do not need as much help. Or it could mean people got comfortable enough with the AI that they stopped looking for the deeper help they might still need (Shoshani et al., 2026).

I do not know which story is true. The study cannot tell us. Both are possible.

I made myself sit with this. As someone who builds writing tools, the easy story is the first one. People feel better, so they need less. Done. But the second story is also possible. And the second story, if it is true, is something I would need to take seriously.

The thing the AI could not do

The study measured three kinds of mental health symptoms. Anxiety. Sadness. And something called PTSD, which stands for post-traumatic stress disorder, the kind of symptoms that come from living through something very hard or scary.

The AI helped with anxiety. The AI helped with sadness. The AI did not help with PTSD (Shoshani et al., 2026).

The researchers were honest about why. PTSD usually needs a special kind of treatment. It is not the kind of thing a daily chat can fix. It needs a trained person doing specific work with you, often over a long time.

This matters. It means the AI is good for some things and not for others. It means using the AI for the wrong kind of suffering would be like using a band-aid for a broken bone. The band-aid is not bad. It is just wrong for the job.

The bias check

I want to mention one more thing the readers of the study should know.

The lead author of the study and one other author have business connections to the company that makes the AI used in the study (Shoshani et al., 2026). They have shares in the company. One of them works there.

But it does mean the people who designed and ran the study had a reason to want it to work. That does not mean the study is wrong. The methods were careful. The numbers are real. Other researchers can repeat the work.

It does mean we should treat this as one strong study, not as the final word. Real progress in research takes many studies, by many teams, in many places.

The same way I have to watch my own bias as someone who builds writing tools. The researchers had to know readers would watch theirs.

The next chapter is about why I think the change is coming anyway.

References

Shoshani, A., Gurfinkel, B., Kor, A., et al. (2026). Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance: A Randomized Clinical Trial. JAMA Network Open, 9(4), e266713.

The change is here. Whether we are ready or not.

The math does not work

About 1 in every 8 people in the world has a mental health condition (Shoshani et al., 2026). That is more than a billion people.

Of those people, only about 1 in every 4 gets care. Three out of every four go without.

There are not enough therapists. There never will be. Even if every country trained ten times more therapists starting tomorrow, the gap would still be there in twenty years.

That is the math. And the math is why I think this change is coming whether anyone wants it or not.

People are nervous, and that is fair

Lots of people are uneasy about using AI for feelings. I am too, sometimes. The nervousness makes sense.

It feels strange to talk about the deepest parts of yourself with a machine. It feels strange to think a tool could feel warm. It feels strange to imagine your own kid, twenty years from now, telling you that the first thing they did when they felt sad was open an app.

These feelings are not silly. They are signals. They are telling us to be careful, to think hard, to ask good questions.

But here is what is also happening. Quietly, in many homes, people are already turning to AI when they feel low. The change is not waiting for our permission. The study I read is one of the first careful looks at whether it actually helps. There will be more.

Where journaling fits in

Writing about your feelings is one of the oldest tools we have. People have kept journals for thousands of years. Modern science has studied this for decades. The research is clear. Putting feelings into words helps people feel better, think more clearly, and make better choices.

What is new is that the writing can now be a conversation. You write something. Something writes back. Not with a real mind on the other side, but with patterns shaped by careful design.

This is the part of the change I work on. At Hurroz, we built Sol, a reflective AI that helps people think through what they wrote. Sol is not a therapist. Sol does not pretend to be human. Sol is a careful tool meant to help you reflect, the way a wise friend who listens well might help you reflect.

I built Sol because of what this study and others like it suggest. People who write about their feelings often, with something that responds thoughtfully, do feel better. That is real. The careful question is how to do it well.

The honest close

I want to be careful here at the end.

This is one study. Of 995 young, educated people. With mild to moderate stress. Not severe illness. Not all walks of life. The findings might be different in different places, with different people, over longer time.

If you are reading this and any of it touches a place that is hurting in you, please remember. This is not therapy advice. It is one person reading one study. If you are feeling the kind of low that does not lift, or the kind of fear that does not pass, or the kind of pain that lasts a long time, the right people to help with that are mental health professionals. They have training that no app and no article can replace.

What I take from the study is something smaller, and I hope, more honest. Something is changing in how people get support for their feelings. The change is not all good. The change is not all bad. The change is happening.

The work, for those of us who build tools in this space, is to take it seriously. To watch our biases. To listen to the people who are nervous. And to build the carefully, knowing that the easier path is to build them carelessly and let the hype do the rest.

That is what I am thinking about. That is the work.

References

Shoshani, A., Gurfinkel, B., Kor, A., et al. (2026). Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance: A Randomized Clinical Trial. JAMA Network Open, 9(4), e266713.