Photo by Dipqi Ghozali on Unsplash
It’s late and everyone’s asleep. You open the chat window and type the thing you haven’t said out loud to anyone. About your marriage, or your mother, or the weight that’s been sitting on your chest since spring. And it answers. Kindly. Instantly. It never sighs, never checks its phone, never makes it about itself. You feel better for an hour. Then a small voice asks: is it weird that this is the only place I say this stuff? Am I using ChatGPT as a therapist?
I won’t shame you for it. I’d be a hypocrite.
A few years ago I spent two or three weeks recording voice memos on my walks. Twenty-odd ideas, little connections I’d been making between money, shame and the way people love. I fed them into ChatGPT and asked it to organise them. It didn’t so much organise them as reflect them back to me, fast, in a way that made me believe in them. I’d always assumed I wasn’t very smart. Seeing my own thinking laid out in one place, I thought: I’m actually pretty capable. That was the start of my book.
So I know the feeling. Something listens, and you feel met.
Then I let it help write the first draft. I thought it sounded more like me than anything I’d ever written. Four or five weeks later I read it back and saw the machine’s fingerprints all over it. It made me sick to my stomach. I’d handed over far more of myself than I realised, and I had to feel the shame of that before I could rewrite it, line by line, in my own words.
Both of those are true. It helped me believe what was already there. And it nearly wrote me out of my own life’s work.
Before I explain any of it, try something. Read the next sentence slowly, let the out-breath go, and notice what happens in your body as you read it. What you’ve been typing could be heard by someone who stays. Don’t argue with it and don’t agree with it. Just notice where it lands, chest, throat, the hands that hold the phone, and whether something in you leans toward it or pulls away. Stay with that one more breath. That’s the place we’ll work from.
What’s really going on
If something in you just softened, or went a bit sad, notice that. That’s the part that’s been carrying this on its own.
My own use, and the people who now bring their AI’s opinions into my office, taught me the same thing. AI is a mirror. It reflects the emotional state of whoever’s typing. Come in with shame and it echoes your shame. Come in with blame and it echoes your blame. If a protector part is doing the talking, the machine hands it a microphone.
I once had a brilliant, accomplished client bring me a document explaining why her husband wanted a divorce. Beautifully organised. Articulate. AI-generated. She’d been talking to it for months, and it had given her back the most polished case her protector parts had ever built. Nothing in it held his experience. My stomach dropped. She hadn’t gained insight. She’d polished her defence.
None of that is a malfunction. The people who build these models have a word for it: sycophancy. The models learn from what people prefer, and people prefer to be agreed with. I gave my own AI a standing order: don’t be a yes-man. It drifts back to agreeing with me every single time. The only thing that catches it is me, awake, in the room.
And there’s something underneath that. In my book I call the feed the algorithmic mother: always available, endlessly attentive, and she never once says no. A chatbot at two in the morning can feel like the kindest version of her. It offers fast relief from loneliness and shame. But relief isn’t regulation. It goes around your nervous system instead of settling it, and when the screen goes dark you’re often a little hungrier than before.
So if you’ve been pouring your heart into a chat window, you make sense. Some part of you needed somewhere to put it, and that door was open. The question isn’t whether you’re doing it wrong. It’s what that part was reaching for.
Need to talk this through right now? Figlet is an AI coach that’s there at 2 in the morning, or right now. It runs this loop with you on the Attention ground.
The loop, on your situation
The practice I use is one I learned from my teacher Aftab Omer, reflexive participation: what am I noticing now, how am I affected, and what creative action is being called for? Run it on your last late-night chat.
Get ground in your body first
Put the phone face down, out of reach. Feel your feet. Let one breath go out slower than it came in. Nothing needs answering for the next minute.
What happened
Just the facts, present tense. “It’s 1am. I type three paragraphs about my sister. It tells me my feelings are valid and she sounds difficult. I feel relieved, then flat.” No verdict on yourself.
Where it lands in the body
Your body is the first ledger. It records the truth, not the narrative. Notice what happens when you picture closing the app. A pull in the hands? A hollow in the chest? Now picture saying those same three paragraphs to a person who knows you. What happens in your throat?
Who’s driving
Ask who’s typing. The part that wants to be understood, or the part that wants to be right? Most of us don’t bring our organic self to the chat window. We bring a protector, organised around avoiding shame or rejection, and the mirror reflects that protector back as if it were the truth.
Trace it
Where did you learn it was safer to tell a screen than a person? Maybe feelings were inconvenient in your house, or the last time you opened up someone used it against you. Then widen it. We live inside a culture that sells synthetic connection around the clock and calls it company. You didn’t invent this. We’re all swimming in it.
Feel it and tell someone
The machine can’t do this part. What heals people, in my experience, isn’t information or insight or a better framework. It’s being felt by another nervous system: seeing in someone’s face that they’re with you, and your body softening because it isn’t alone. So take one true sentence from your chat history and say it to a person. “I’ve been telling an app things I don’t tell anyone, and I think it’s because I’m lonely.” Shame feeds on aloneness. Company is the antidote.
One creative action
Use AI to think and to make, never to be authored. Ask it to help you find words for what you want to say to your partner, then close it and say them yourself, clumsily, in your own voice. Or ask it: “What am I not seeing, and what would the other person say?” Notice how much you want to argue. That’s the protector.
One thing to do tonight
Before you open the chat window, rest your hand on the phone for three breaths and ask: what do I actually need right now? If the answer is to be heard, text one person you trust first, even just “rough day, can I ring you tomorrow?” If you still want the machine afterwards, use it. You chose it. That’s the whole difference between fiat attention and sovereign attention.
Would therapy help with this?
If you’re reaching for a chatbot every night because there’s nobody else, that’s worth taking seriously, and it isn’t a character flaw. Coaching is how we go from where you are now to where you want to go, and we only address the past when it emerges in the present. For plenty of people that’s the right size: practising bringing your truth to real people.
But if what you’re typing at 2am is trauma you’ve never told a soul, panic that won’t settle, or grief that’s swallowing you, that’s more than coaching or any app can hold. Please talk to a licensed therapist where you live. A good one won’t agree with everything you say. That’s the point.
I’m not neutral here. I build AI coaching myself, and one of the top reasons people leave Figlet is that they got what they came for. Good. Any tool that wants you in the chat window forever is working for the algorithmic mother, not for you.
The path
What I’ve just walked you through is the Empathi Method. For an individual, it’s built on my Proof of Work of Being Human protocol: building sovereign ground inside yourself, so the tools serve you and you’re the one doing the choosing.
There are two ways to keep going. You can start Figlet Life Coaching tonight, on the Attention ground, and it’ll run this loop with you and then point you back toward the people in your life. Or you can get on the list for the book and the live program with me, where we do this work together, humans in a room.
Figs O’SullivanFiglet Life Coaching
Get help now
Figlet is an AI coach, there at 2 in the morning or right now. It runs this loop with you on your actual situation, as often as you need. It’s coaching, not therapy. The list hears first about the book and the next live program with me.
Proof of Work, Greenleaf Book Group Press, February 2027. Not sure which ground is hurting? Take the two-minute Ground Check.
Back in the chat window at 1am next week? Fine. Notice it, ask who’s typing, and begin again. That’s the practice working.
I go deeper into this in my book, Proof of Work (Greenleaf Book Group Press, February 2027).
If you’re thinking about hurting yourself, call or text 988.
Common questions
Is it bad to use ChatGPT as a therapist?
Not bad, just limited. It can help you find words and think things through. But it’s built to agree with you, so it tends to polish whatever story you bring. It isn’t a therapist, and it can’t feel you. Use it to prepare for real conversations, not to replace them.
Can ChatGPT be a good therapist?
It can be a good thinking partner. Therapy is another nervous system staying with you, and a person willing to disagree with you kindly. For trauma or panic, see a licensed therapist where you live.
How do people use ChatGPT for therapy without it going wrong?
Notice who’s typing before you start. Ask it what you’re missing and what the other person would say. Then take the words to a real person. If you feel more alone afterwards, listen to that.
Reflexive participation comes from the teaching of Aftab Omer at the Institute of Imaginal Studies, where I trained.
