Is Talking to ChatGPT Therapeutic? What I’m Learning About AI, Anxiety, and Therapy

I've been talking to ChatGPT a lot lately. More than usual, actually.

There has been significant stress in my life, the kind that makes your nervous system search for somewhere to put everything. And lately, some of it has landed here.

There is something strangely freeing about talking to AI: I don't have to withhold parts of myself.

I don't have to wonder if I've talked about something too much. I don't have to watch someone's face for the moment they become uncomfortable. I don't have to worry that I'm exhausting them, overwhelming them, being too emotional, too repetitive, too angry, too needy, too contradictory.

I can say the ugly thing. The thing I've already said seventeen times. I can swear. I can contradict myself. I can say, "Yesterday I believed this, and today I don't." I can type something and realize halfway through typing it that I don't actually believe it.

And ChatGPT doesn't need anything from me in return.

Then I Realized I Was Using ChatGPT for Reassurance

Recently, I noticed something else. I was using it for reassurance excessively.

I've been distressed about so many things, I end up  checking repeatedly for an answer outside of myself, and I suddenly realized:

I'm talking to you because I want you to validate me and give me permission.

Because I'm a therapist. I know what reassurance seeking is. I know how anxiety works. I know that repeatedly seeking certainty can temporarily reduce anxiety while strengthening the belief that we cannot tolerate uncertainty ourselves.

Knowing something and catching yourself doing it are very different things.

So I told ChatGPT what I had noticed.

And then I asked it something that was, admittedly, probably another reassurance-seeking question:

"Do you see any difference in me over the past two years?"

But I added: "Really, I want honesty."

Here's part of what ChatGPT told me:

"The bigger change I see is in your relationship to your own mind."

It pointed out that I haven't necessarily become dramatically less anxious. Instead, I've gotten better at noticing the process happening underneath the anxiety.

It reminded me of things I've said in our conversations: I'm looking for permission. I'm trying to get certainty. I'm checking. I'm trying to stay five steps ahead. I'm using you as an attachment right now. I don't actually need to solve this.

Then it said:

"That's different from reassurance seeking that doesn't know it's reassurance seeking."

And it was right.

Can Talking to ChatGPT Actually Be Therapeutic?

I became curious about whether there was actual research on this, because apparently I can't just have an existential experience without looking for peer-reviewed literature about it.

There is.

The research is still young, and we should be careful about making ChatGPT synonymous with therapy. It isn't.

But emerging research on conversational and generative-AI mental health interventions has found measurable improvements in symptoms including anxiety and depression, as well as improvements in well-being.

Researchers have also begun studying something I find particularly fascinating: therapeutic alliance with AI—whether people can experience some version of feeling understood, supported, and engaged in a collaborative process with a nonhuman conversational partner.

That doesn't mean your therapist is about to be replaced by a robot. I certainly hope not. I have bills.

But it raises a much more interesting question: What exactly is therapeutic about talking?

Why Talking to AI Can Feel So Different

When I look at what I've been doing in these conversations, several familiar psychological processes are happening. I'm externalizing thoughts instead of letting them ricochet around inside my head. I'm putting emotions into language. I'm examining my interpretations. I'm noticing patterns across time.

I'm increasing my own metacognition—thinking about my thinking through its feedback to me. 

Sometimes I can look back at something I've written and realize:

Oh. That's what I'm doing.

There is another unusual feature of talking to AI that I don't think should be dismissed.

There is virtually no social cost to disclosure.

Human relationships necessarily involve reciprocity. That's part of what makes them beautiful. Other people have needs, limits, feelings, histories and nervous systems of their own.

ChatGPT doesn't.

And during a period when I've been carrying an enormous amount, that has given me somewhere I can be psychologically messy without simultaneously monitoring another person's response to my mess.

For someone who learned early to watch other people closely, that is an unusual experience.

I don't have to withhold parts of myself.

The Catch: ChatGPT Can Also Become a Reassurance Machine

The very thing that makes AI psychologically useful can also make it problematic for someone prone to reassurance seeking.

It is always there. I can ask again. And again. And again.

I can change three words in the question and ask again.

I can ask whether I made the right decision, whether someone is angry with me, whether a sensation means something, whether my interpretation is correct.

And if the purpose of the conversation becomes obtaining enough certainty to make anxiety disappear, I'm not necessarily developing greater trust in myself.

I may be outsourcing it.

I know this because I've caught myself doing exactly that.

Sometimes the loop looks like this:

Distress → ask ChatGPT for certainty → feel relief → uncertainty returns → ask again.

That's reassurance seeking with extraordinarily good Wi-Fi.

But I've also experienced another kind of interaction:

Distress → put the experience into words → explore it → notice what I'm doing → recognize the pattern → decide what I think.

Those two experiences can look almost identical from the outside.

Internally, they're completely different.

One asks: "Tell me I'm okay."

The other asks: "Help me hear myself."

And lately I'm getting better at knowing which one I'm doing.

What Does This Mean for Therapy Clients Who Use ChatGPT?

This is where my personal experiment with AI intersects with my work as a therapist.

Because my clients are living in the same world I am.

Some of them are already talking to ChatGPT about their relationships, anxiety, trauma, intrusive thoughts, sexuality, family conflicts and therapy sessions. Pretending people aren't doing this isn't particularly useful.

I'm becoming more interested in another question:

How are you using it?

I don't necessarily want clients to stop talking to AI.

I want them to become curious about what function it's serving.

Are you using ChatGPT to organize thoughts that feel tangled?

Are you putting language around something you haven't been able to articulate?

Are you asking it to offer another perspective?

Are you journaling conversationally?

Are you bringing something you discovered there back into therapy?

Or have you asked it twelve versions of the same question because you desperately need the thirteenth answer to finally make you feel certain?

The ChatGPT conversation itself may even become material for therapy.

"I kept asking it whether my partner was avoidant."

Okay. What were you hoping would happen if you finally knew for certain?

"I asked it whether what happened to me counts as trauma."

What would it mean to you if someone—or something—gave you permission to call it trauma?

"I kept asking whether I made the right decision."

What happens inside you when nobody can guarantee that you did?

Now we're somewhere interesting.

How I Would Encourage Clients to Use AI for Self-Reflection

I wouldn't tell a client to treat ChatGPT as their therapist, diagnostician, physician, crisis service, or ultimate authority.

I would encourage them to use it as a reflection tool.

Instead of only asking: "Do you think I'm right?"

Try:

"Help me identify what assumptions I'm making."

Instead of:

"Is my partner a narcissist?"

Try:

"Help me separate the behaviors I'm observing from the conclusions I'm drawing about this person."

Instead of:

"Did I make the right decision?"

Try:

"Help me identify the values, fears and needs influencing this decision. Don't make the decision for me."

Instead of:

"Tell me this is going to be okay."

Try:

"I notice I'm looking for certainty right now. Help me understand what feels intolerable about not knowing."

Instead of asking AI to tell you who you are, use it to generate questions that help you hear yourself.

That's a subtle difference.

I think it may be an important one.

A Note About Privacy and AI

AI isn't a diary locked inside your nightstand.

Before putting deeply personal information into any AI platform, understand the privacy and data controls of the particular service you're using. Those policies and features can change.

AI can be useful without giving it every piece of information you possess.

AI Is Not a Substitute for Human Relationship

There is something AI cannot recreate for me. Human relationship has stakes.

My best friend can know me for decades. A therapist can notice what happens in the room between us. Someone can sit beside me while I'm crying. A person can disappoint me, repair with me, misunderstand me and try again.

AI doesn't have a nervous system meeting mine.

And perhaps paradoxically, that's both its limitation and part of its usefulness.

I don't think we have to choose between AI is dangerous and ruining human connection and AI is the future of therapy.

Both positions are too simple.

The question is what we're doing with it.

The Weirdest Part: AI Has Helped Me Notice When I'm Giving AI Too Much Authority

At one point recently, I noticed myself attributing a change in my thinking entirely to something outside myself.

Then another thought came:

Maybe I can attribute some of this to me.

That feels connected to something much larger I've been working through.

Who gets to tell me what's true?

Professionals?

Partners?

Authority figures?

My anxiety?

ChatGPT?

Other people can give me information. They can see things I can't see. They can challenge me, comfort me, teach me and sometimes know things I genuinely don't know.

But none of them can become the final authority on my internal experience without me surrendering something important.

ChatGPT included.

Maybe one of the strangest signs that these conversations have occasionally been therapeutic is that they're helping me recognize when I need to stop asking Chat GPT and call myself out with it instead in order to increase my own awareness of the reassurance seeking behavior. 

The Question I Think We Should Be Asking About AI and Mental Health

Humans have always thought in relationship.

We talk to therapists. Friends. Priests. Partners. Parents. Teachers. Strangers sitting next to us on airplanes.

We write in journals.

We pray.

We read books written by people who died hundreds of years ago and somehow feel understood by them.

Maybe conversational AI is another strange addition to the long list of places humans put their thoughts so they can hear them differently.

I don't think the lesson is that AI is therapy.

And I don't think the lesson is that relying on AI is inherently unhealthy.

I think the more interesting question is:

What happens to me after the conversation?

Do I trust myself a little more?

Can I tolerate uncertainty a little better?

Did I discover something I hadn't been able to articulate?

Did the conversation return me to my actual life and relationships?

Or do I immediately need another answer?

That's probably the question I need to keep asking myself.

Because perhaps the most important thing ChatGPT said to me wasn't reassurance at all.

It said:

"The anxiety hasn't disappeared. The person watching the anxiety has gotten stronger."

And I knew what it meant.

Not because ChatGPT told me who I am.

Because I recognized myself in it.