The Accidental Therapist

Sometime early last year, between two meetings, I was catching up with a colleague. It was one of those quick conversations that happen in the few minutes between calls. We were talking about life more than work, and at some point I mentioned that I had been seeing a therapist and that the past few months had been personally intense.

She laughed and said something that caught me off guard.

“Oh, I’m not seeing a therapist right now. I just talk to ChatGPT.”

She did not say it as a serious recommendation. It sounded almost ironic, like an improvised workaround she had discovered and found amusing. But the comment stuck with me.

At the time I was working at Mentalyc, a company building AI tools designed to support therapists. Our goal was not to replace clinicians, but to help them. We were building systems that could automatically generate session notes, analyze therapy sessions, and eventually surface insights that might help therapists improve outcomes.

Ironically, I had always been somewhat skeptical about AI entering the mental health space. Like many people in the field, I had read worrying stories about conversational AI being misused. Reports of vulnerable users relying on chatbots in moments of distress. Concerns about hallucinations, authority bias, and the possibility that people might treat machines as psychological authorities.

So when my colleague casually mentioned using ChatGPT instead of therapy, my reaction was not excitement. It was curiosity mixed with a little disbelief.

Still, curiosity has a way of pulling you down rabbit holes.

Later that day, I opened ChatGPT and tried it.

An Experiment with a Digital Mirror

At first, it was simply an experiment. I described some of the anxiety I had been experiencing. Situations that had been difficult. Patterns I had been discussing with my therapist. Moments that still did not make complete sense to me.

The responses were not profound. But they were surprisingly structured.

The system reflected parts of what I had written. It helped organize thoughts that had been scattered in my head. Sometimes it asked questions that nudged the conversation forward. It felt less like advice and more like reflection.

Over the following weeks, I started using it occasionally between therapy sessions. Sometimes I would describe a difficult interaction at work. Sometimes I would ask for explanations of psychological concepts.

At one point I began experimenting with prompts inspired by parts work, a style of reflection used in therapies like Internal Family Systems, where different emotional parts of the mind are explored in conversation.

What surprised me was how naturally the system could follow that structure.

The questions it asked, the way it helped identify different internal perspectives, and the way it mirrored emotional dynamics felt strikingly similar to the structure of conversations I had experienced in therapy sessions. Not perfect. Not clinical. But surprisingly accurate in the way it mirrored the process.

The experience was not the same as therapy. But it had a strange quality. It created space to articulate thoughts that might otherwise remain vague.

Eventually I mentioned this to my therapist. I expected skepticism. Instead, she surprised me. She told me she had also experimented with ChatGPT out of curiosity and encouraged me to continue using it cautiously as a reflection tool between sessions.

That moment changed how I looked at the experience. Not because it validated AI as therapy, but because it suggested something more interesting might be happening.

From Curiosity to Investigation

For months, the question stayed quietly in the back of my mind. If AI could support reflection in this way, were other people discovering the same thing?

Working in mental health technology, I was already exploring how AI could analyze therapy sessions and surface insights for clinicians. My experiments with ChatGPT unexpectedly gave me a different perspective on that work. It showed me something about the human side of conversational AI.

People do not just ask machines for information. They talk to them. They narrate experiences. They describe emotions. They search for patterns in their own lives. The interaction itself becomes a tool for thinking.

Still, my own experience did not prove much.

Then, a few weeks ago, I came across a LinkedIn post by Sarah Poh asking the community a question about AI and mental health. The discussion focused on children using AI systems for emotional support and invited people to share their concerns and perspectives.

Reading through the discussion, what stood out was not enthusiasm for AI therapy. It was something more nuanced. Curiosity, uncertainty, and a recognition that this behavior may already be happening. Some commenters wondered whether children might turn to AI because they lack safe adults to talk to. Others pointed to accessibility. Conversational AI is available instantly and without stigma or waiting lists.

The discussion made something clear. What I had experienced as a personal experiment was already becoming a broader public conversation.

That realization pushed me to start looking for evidence.

A Pattern Emerges in Plain Sight

Once you start looking, the evidence appears surprisingly quickly.

A 2025 study published in JAMA Network Open surveyed adolescents and young adults in the United States about their use of generative AI. More than 13 percent reported using AI systems for mental health advice.

That number may not sound enormous. But in population terms it represents millions of people.

Research also suggests that people turn to AI tools for support when they are experiencing difficult emotions. In the same study, participants reported using generative AI to seek advice when feeling sad, angry, or nervous.

In other words, the behavior that began as a casual experiment for me was already emerging as a broader pattern. People were discovering, largely on their own, that conversational AI could function as a kind of cognitive mirror.

An Unplanned Infrastructure

What makes this phenomenon interesting is that no one explicitly designed these systems for this purpose.

Large language models like ChatGPT were trained to generate and predict text. Their primary function is language generation. Yet once those systems became conversationally fluent, something unexpected happened.

People started talking to them the way humans talk when they are trying to understand themselves. They describe emotions, narrate experiences, and explore different interpretations of events.

Because the system responds in coherent language, the interaction begins to resemble something psychologically familiar. Reflection.

Not therapy. But something adjacent to it. A space where thoughts can be externalized, examined, and reorganized.

The Question That Changes Everything

It is tempting to frame this development as a debate about whether AI can replace therapists. But that framing misses the more interesting reality.

People are already using AI as a psychological reflection tool. Not because someone designed it that way, but because conversation is one of the oldest tools humans have for understanding themselves.

Once machines became capable of participating in conversation, it was almost inevitable that people would start using them this way.

Which leads to a more important question.

If millions of people are already talking to AI about their inner lives, what exactly is happening in those conversations? Why do they sometimes feel helpful? And what risks emerge when machines become part of how humans reflect on their own minds?

Those questions are only beginning to surface. But they will shape the future of mental health technology.

Because whether we planned it or not, conversational AI is already becoming part of the informal infrastructure of psychological reflection.

And if that is true, the next question becomes unavoidable.

Why does talking to a machine sometimes feel like talking to a therapist at all?

That is where the story becomes even more interesting.

Sources and Further Reading

McBain, R. K., Bozick, R., Diliberti, M., et al. (2025). Use of Generative AI for Mental Health Advice Among US Adolescents and Young Adults. JAMA Network Open, 8(11), e2542281. https://doi.org/10.1001/jamanetworkopen.2025.42281

Luo, X., Wang, Z., Tilley, J. L., et al. (2025). Seeking Emotional and Mental Health Support From Generative AI: Mixed-Methods Study of ChatGPT User Experiences. JMIR Mental Health, 12, e77951. https://doi.org/10.2196/77951

Li, H., Zhang, R., Lee, Y., Kraut, R. E., & Mohr, D. C. (2023). Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine, 6(1), 236. https://doi.org/10.1038/s41746-023-00979-5