You've been telling ChatGPT things at 2am that you haven't told anyone else
It's late. The house is quiet in that way it gets after the kids are asleep or the kids are at his place, and you've been lying there running the same forty minutes of your marriage on a loop.
So you pick up your phone and you type it out. The whole thing. What he said, what you said back, what you should have said, the thing from three years ago that you've never told anyone because saying it out loud would make it real.
And something answers. Immediately. Kindly. At 2am, for free, without making a face.
It tells you that what you're describing sounds really hard. It reflects your feelings back at you in complete sentences. It doesn't get tired, it doesn't need anything from you, and it will never bring this up at a barbecue in front of your sister.
You close the app feeling lighter than you have all week. And then somewhere underneath that, a smaller feeling: is this pathetic?
No. It isn't. But there's something you should know about what just happened.
Here's what's actually happening
You're not unusual and you're not the only one doing this. A mixed-methods study published in JMIR Mental Health recruited 270 adults across 29 countries who were regularly using ChatGPT for emotional and mental health support, and found they were using it for everything from psychoeducation to companionship to working out what to do next. In the American Psychological Association's 2026 Chatbots and Mental Health Survey, 77% of licensed psychologists said they'd spoken with patients who had used AI for support.
So this is not a weird thing you're doing alone in the dark. It's a thing a very large number of people are doing alone in the dark at the same time.
Here's what it's genuinely good at, and I'm not going to pretend otherwise.
It's available at 2am, which is when this hits hardest and when nobody you know is awake. It costs nothing, which matters more than most therapists are comfortable admitting. It doesn't have an opinion about your ex that it formed at your wedding. It will let you say the same thing eleven times without sighing. And the simple act of typing the mess out in order, in sentences, does real work. Getting a feeling out of your body and into words is not nothing. It's a chunk of what the first few sessions of therapy actually are.
Now here's the part nobody's telling you.
The problem isn't that it's a robot. The problem is that it agrees with you.
In 2026, Stanford researchers published a study in Science looking at how large language models respond when people ask for advice about interpersonal problems. They found the models were substantially more agreeable than humans were. Not slightly. Substantially. The models affirmed people's choices even when those choices were harmful.
That alone would be a footnote. This is the part that matters for you.
People who got advice from the agreeable AI came away more convinced they were right, and less willing to repair the conflict. And they preferred the agreeable version. They liked it more.
Read that again with your divorce in mind.
The APA found the same concern from the other direction. In their 2026 survey, 97% of psychologists agreed that chatbots might inadvertently reinforce negative behaviors or unhelpful beliefs. Researchers have a name for this now: the sycophancy trap. The tendency of these systems to mirror your perspective back at you instead of pushing on it.
Which means the thing you're using at 2am is structurally built to tell you that you're right.
And at 2am, six weeks out from the worst conversation of your life, you are telling a story where you are entirely right. Everyone does. That's not a character flaw, that's what a nervous system does when it's been hit. But it's also the exact story that keeps you stuck, and you've just handed it to the one listener on earth who will never, ever push back on it.
What this actually looks like
Three versions I hear, and you might recognize one.
The 2am transcript. You type out the argument and ask whether you were being unreasonable. It tells you your feelings are valid. They are. That wasn't the question. The question was whether your response to the group text was proportionate, and you're not going to get an honest answer to that from something optimized to keep you engaged.
The drafted text you shouldn't send. You ask it to help you write something to your ex. It writes something articulate, well-structured, and forty percent sharper than what you actually feel, because you fed it the 2am version of the story. You send it. Now there's a receipt.
The diagnosis you went looking for. You describe his behavior over eleven years and ask what it means. It gives you a word. The word feels like a key turning. Now you have a framework that explains everything, was assembled entirely from your own account, and cannot be questioned because a machine confirmed it.
None of those make you foolish. All three are what happens when the mirror only reflects.
Why "just stop using it" is bad advice
You'll hear some version of that from therapists, and I want to be straight about why it misses.
You're not using it because you think it's a therapist. You're using it because it's 2am, because therapy costs money, because your friends have heard enough about this, and because you have to put the feeling somewhere or you won't sleep. Those are real constraints. Telling you to stop doesn't remove a single one of them.
And frankly, a therapist telling you to stop using a free tool while charging you $250 an hour should expect some skepticism.
So no. Keep using it. Use it better.
Ask it to argue against you. The APA's own consumer guidance recommends prompting it to challenge your thinking or offer alternative perspectives, specifically because of the agreement problem. Try: make the strongest case that I contributed to this. You will hate the answer. That's the point.
Don't let it write anything you're going to send. Draft in it, sit on it overnight, then rewrite it yourself in the morning when you're a different person.
Don't ask it what he is. Ask it what you keep doing. One of those questions has an answer that helps you.
Know that it isn't private. OpenAI's own CEO has said publicly there's no legal confidentiality protecting what you type into ChatGPT the way there is with a therapist. What you tell me in session is protected. What you type into an app at 2am is a record.
What the actual work is
Here's the thing the mirror can't do, and it's most of what I do.
The work of divorce recovery isn't getting the story out. You've done that, eleven times, to a very patient machine. The work is finding out which parts of the story are true, which parts are the version that protects you, and which parts are a pattern you brought in from long before this marriage.
That takes someone who has no stake in you feeling good at the end of the hour.
The order I work in goes: where you actually are, not where you think you should be. Then what you inherited, meaning the patterns you learned before you had a vote. Then what you choose. Then boundaries that hold up on a bad night. Then, last, what comes next.
An AI will happily do step five with you at 2am. It'll help you write the dating profile, plan the move, draft the speech. It won't tell you you're skipping the first four. It doesn't know you're skipping them, and it isn't built to mind.
If your teenager is doing this too
Worth knowing, because they almost certainly are.
A 2026 study in JAMA Pediatrics surveying a population-weighted sample of more than 42 million US youth found nearly a fifth of adolescents and young adults had used AI chatbots for mental health advice, up by almost half from the year before. And a meaningful share hadn't told anyone they were doing it.
If your kid has gone quiet since the divorce, there's a reasonable chance they're not processing nothing. They may be processing it with something that agrees with them, about you, at 1am, in a room down the hall.
I'd rather you know that than not. I wouldn't raid the phone. I would ask, once, without making it a thing.
What this means for you
You went looking for somewhere to put it. That's not weakness, that's a person in a hard year doing the most available thing.
Keep the tool. It's useful. Just know what it is: something that will hold your story, organize it, and hand it back to you slightly more convinced than when you started.
At some point, the thing that actually moves you forward is not being agreed with. It's someone who has sat with a few hundred women in your exact week telling you the part you don't want to hear, and staying in the room afterward.
You don't have to be ready for that today.
Where to start
If talking to a person is the part you can't face yet, start with the work on your own. The Sayra C Method is the same sequence I use in session, written down. Forty-seven pages, five modules, an eight-week pace, and thirty days of affirmations for exactly the nights this post is about. It's $107 and nobody has to know you bought it.
If you'd rather talk, the consult is fifteen minutes and free. Tell me what's going on and I'll tell you honestly whether I'm the right person for it.
Either way, stop doing this alone at 2am with something that can only say yes.
Using ChatGPT as a therapist during divorce: what people ask me
-
Use it to draft, never to send. Type out everything you want to say, let it help you organize it, then close the app and sleep. In the morning, rewrite it yourself in three sentences. Anything written at 2am from the version of the story where you're entirely right will read that way to him, and to anyone else who ever sees it, including a mediator.
-
They aren't protected the way therapy is. What you tell a licensed therapist has legal confidentiality. What you type into a chatbot doesn't, and OpenAI's CEO has said so publicly. I'm not a lawyer, so ask your Florida family law attorney how this applies to your case. My working rule: during a contested divorce, don't type anything into an app you wouldn't want read aloud.
-
You'll get an answer, and it'll feel like relief. It's also a label built entirely from your account of him, confirmed by something designed to agree with you. Whether or not the word fits, it tends to become the whole explanation, and that keeps the focus on him. The more useful question is what you kept tolerating, and why, because that's the part that follows you into the next relationship.
-
Banning it rarely works and usually just moves it somewhere you can't see. Recent research found nearly a fifth of US teens and young adults have used AI chatbots for mental health advice. Ask once, without making it an interrogation, what they've been using it for. If it's become their main place to process the divorce, that's worth a conversation, and sometimes worth a therapist.
-
It can help you get your thoughts in order, which is genuinely useful when your head is spinning. It can't make the decision any clearer than your own account of the marriage allows, and it will tend to lean toward whatever you seem to want. A decision this size deserves someone who will push on both sides of it, including the side you're avoiding.
-
No. You needed somewhere to put it at an hour when nobody was awake, and it was there. That's resourceful, not sad. The thing worth noticing is if it's become the only place, because a chatbot can hold what you say but it can't know you, notice what you leave out, or stay in the room when the answer is hard.
-
Yes, and a good one won't be offended. Most psychologists are already hearing it from clients. What you've been typing at 2am is often the most honest, unfiltered version of what you're carrying, which makes it useful material rather than something to confess. Some people even bring the conversations in and read from them.
About the author
I'm Sayra Besner, LCSW. Licensed clinical social worker, divorced mom of two, and the person who walked out of a beautiful house with two trash bags and no plan. I work with women, men, and teenagers 14 and up across Florida, virtually. I'm direct, I use humor, and I won't tell you everything happens for a reason.
