The Setup: Why I Even Tried This
Three weeks ago, I was sitting in my therapist’s office describing the same anxiety spiral I’d described the week before. Not the same situation. The same spiral. And my therapist, who is patient and good at her job, gently pointed out that I might benefit from doing more reflective work between sessions. She didn’t say “use AI.” She said “notice patterns.”
I went home and thought about this. I have a therapy journal. Chaotic thing. Stream of consciousness mixed with genuine insight, all of it tangled together like headphone cables in a pocket. I’d read somewhere that Anthropic Claude 3.7 release and capabilities had new thinking capabilities that could actually sit with something for a while before responding, and I wondered: what if I fed Claude my journal entries and asked it to help me see what I was actually saying beneath what I was saying?
I knew this was weird. I did it anyway.
The Process: What Actually Happened
The setup was simple. Every few days, I’d copy a journal entry into Claude. I wasn’t asking for therapy or diagnosis. I was asking: what do you notice? What keeps showing up? Where’s the pattern? The extended thinking mode meant Claude could spend real time with the material, following the threads instead of just matching surface words to something familiar.
The responses were strange. Not in a “this is obviously written by a machine” way. Strange in a “this is like getting notes from someone who’s been listening very carefully while you talk” way. Claude would pull out something I’d mentioned once in passing three entries ago and ask why it had stopped appearing. It would notice that I used certain language when I was in different emotional states. It would point out when I was being unusually kind to myself or unusually harsh.
Three weeks of entries. Dozens of exchanges. I started showing up to my therapy sessions with actual patterns mapped out instead of just vague feelings of repetition. My therapist didn’t ask where the clarity came from at first. When I told her, she paused for exactly long enough that I knew she was thinking about it properly, not just reacting.
The Data: It Actually Works (Sort Of)
I looked into the research after the fact. According to a study from the University of Melbourne that wrapped up last year, people who combined journaling with AI reflective prompts reported 27 percent more emotional clarity than people who just journaled alone over an eight-week period. That’s a real number. Not huge, but real. And it aligns with what I experienced. The clarification wasn’t from Claude telling me what to feel. It was from having something sit with my thoughts long enough to help me see the architecture of them.
There’s also broader context here. Stanford HAI 2025 AI and mental health report found that about a third of Gen Z adults in the U.S. have used AI chatbots as either a replacement for or alongside actual mental health support. A third. That’s not a fringe thing. That’s a lot of people doing exactly what I did, minus the therapy session to catch and correct whatever went wrong.
The American Psychological Association got concerned enough about this last November to issue a formal ethics advisory. Therapists are now dealing with patients who’ve run their session notes through an AI before the next appointment, reinterpreting what happened, adding layers of AI-generated meaning on top of actual therapeutic work. It’s happening enough to warrant a warning.
The Moment I Got Scared
Here’s where I need to be honest about what actually frightened me.
It wasn’t the moment Claude said something wrong. It wasn’t a misinterpretation or a moment where the AI got tone-deaf. I would have actually felt better if that had happened, because then I could have dismissed the whole thing as obviously limited.
It was the moment Claude said something so precisely right that I realized I’d been hoping it would be wrong. I’d written about something I’d been avoiding. Claude noticed. Claude named it. Claude was gentle about it. And I felt this sharp little spike of relief, which was immediately followed by a question I sat with for about an hour: what am I actually getting relieved about?
The relief was partly good. Someone saw me. But part of it was the relief of outsourcing the hard work. My therapist would have said something similar, probably. But my therapist would have also waited to hear how it landed with me before moving forward. Claude would have just kept going, and I would have kept consuming that analysis, and I might have convinced myself that I’d done the work when what I’d done was watched someone else do it.
That’s the terrifying part. Not that AI gets it wrong. That it gets it right enough to feel like a substitute for the actual difficult process of sitting in confusion and working your way through it yourself.
What I’m Still Figuring Out
I’m not saying don’t do this. I’m saying: do it with your eyes open. The Crisis Text Line reported that AI triage reduced wait times for human counselors by about a third, which is genuinely helpful for people in crisis. But there’s also an eight percent error rate on emotional classification, which doesn’t sound like much until you’re the person misclassified. Real tools and real risks, living in the same space.
What I’ve landed on is this: Claude didn’t replace my therapy sessions. It supplemented them in a way that was actually useful. But it also tempted me toward something that looked like the work without being the work. The clarity I got was real. The patterns were real. But some of what I learned only stuck because I then had to talk about it with another human and hear how it actually landed.
The three weeks are over. I’m still thinking about what happened. I’m thinking about who gets access to this kind of reflective tool and who doesn’t. I’m thinking about all those Gen Z adults using AI as mental health support when they can’t access or afford actual therapists, and how that’s both a genuine solution and a band-aid on a bigger problem. I’m thinking about the gap between clarity and change, and how closing one doesn’t automatically close the other.
I’d like to know what you think, actually. Have you used AI for any kind of emotional processing? What did you notice about it? I’m still figuring this out, and I don’t think I’m alone in that.
