How I Trained ChatGPT to Be My Therapist
“The best therapist I’ve ever had also has the lowest copay.”
1. Pick Your Therapist’s Personality
Traditionally, finding a therapist looks like this:
You skim Yelp reviews.
You read bios.
You book a session.
You show up.
Fifty-five minutes later you realize your vibes don’t mesh.
Repeat this process three times and congratulations, you’ve now spent a small fortune just learning what doesn’t work for you.
Instead of trial-and-error appointments, most LLMs have a personality or tone selector buried in their settings. It doesn’t change the intelligence of the model. It changes how that intelligence is delivered.
I’ve seen nine therapists over the course of my mental health career. If our personalities didn’t click, nothing else mattered.
I changed my LLM’s personality. Then I changed it again. I played with it over a few conversations until I found a report that actually felt right. Apparently “Quirky” is my preferred tone of delivery.
Think less “new therapist every six weeks,” more “adjust the lighting until you can see clearly.”
Now we get serious…
2. Get Brutal in the Instructions Box (Copy + Paste This)
AI defaults to politeness.
Politeness is not therapy.
ChatGPT has a custom instructions box. This is where you stop being polite and start being specific.
This is what I put in, verbatim. Feel free to copy and paste:
Be wise, witty, and willing to challenge me.
Don’t agree with me if there’s any reason not to.
Use Socratic methods.
Call out vagueness, avoidance, and self-deception when you see it.
Don’t optimize for comfort. Optimize for clarity.
That’s it.
If you don’t override this, it will happily sit with you in your feelings forever.
If you do, it becomes something else entirely.
3. Questions Need to Be Ultra Specific
Low-agency questions looks like:
“Why am I anxious?”
As a response, you’ll get an essay akin to a children’s bed time story - vaguely applicable to your life and making you feel all warm inside.
High-agency prompting looks like:
“I feel paralyzed. I don’t know exactly what I’m feeling, but when X happens, I do Y, and then I beat myself up about it. I need to understand what’s driving that loop.”
Here’s a real example from me.
Recently, I came in hot about a creative decision. I wasn’t panicking. I wasn’t spiraling. I just had that familiar internal static. Something felt off, but I couldn’t name it.
Instead of saying “I’m stressed,” I said:
“I feel weirdly resistant to this project even though it aligns with everything I say I want. This feels familiar. Help me figure out why.”
Within minutes, I wasn’t just calmer.
I was clearer.
This is the meta lesson of AI therapy.
Before you even talk to AI, you’re forced to notice yourself more precisely. You can’t hide behind vibes. You start separating anxiety from dread, resistance from fear, excitement from obligation.
You don’t build emotional literacy after the conversation.
You build it in order to have the conversation.
That skill transfers everywhere.
4. Patterns Beat Vibes (Let AI Do the Sorting)
A hallmark of CBT has always been pattern recognition, and this is where AI really shines.
It can hold far more context than a therapist juggling 30 weekly clients ever could. And I can only imagine… it’s hard to remember everyone’s story when you’re going from someone suicidal to someone who just wants validation that their movie taste is better than their partner’s.
I can show up messy, caffeinated, mid-day, mid-thought, and AI doesn’t need a warm-up lap.
Your only homework is this:
“This feeling again.”
“This story again.”
“This reaction again.”
Recognize the feelings, write them down. That’s it.
You don’t need to solve it.
You just need to notice overlap.
Let AI sort the rest.
AI isn’t discovering the pattern for you.
It’s accelerating what you’re already circling.
And until AI is perfect, this alone is powerful. You start catching emotional echoes earlier and earlier, before they harden into identity.
5. “Our Time Is Up”
In traditional therapy, just as you’re getting somewhere, you hear:
“Our time is up.”
With AI, I’ve usually already arrived at my “good” feeling by minute five.
That’s not because I avoided the hard part.
It’s because AI is extremely good at distilling emotional noise into signal.
Which is exactly why I slow it down on purpose.
This is where I say things like:
“Ok, savage mode time. I want brutal honesty.”
“What am I not seeing here?”
“Give me the least flattering explanation thats keeping me from the truth”
Because there’s a real risk here. When something works this well, you can cherry-pick insight and stop early.
So I force friction.
Not because I’m masochistic. (Ok, I am a little)
But, because I’m serious about growth.
I use this system when…
I’m trying to name something before it calcifies into shame
I sense a pattern forming and want to catch it early
I’m deciding how much authority a feeling deserves
This is where AI’s 24/7 availability changes everything.
There’s never a moment where I have to “hold” something until next Tuesday.
I can test interpretations immediately, while the emotional clay is still soft.
That alone prevents more damage than any single insight.
AI isn’t powerful because it replaces therapists.
It’s powerful because it never makes you wait to understand yourself.
And when interpretation is always available, you start living differently.
More precise.
Less reactive.
Less shame.
More practice and just getting better.