Is ChatGPT really useful for emotional support?

Somewhere between 2am and never quite falling back asleep, a lot of people now open ChatGPT instead of calling a friend, writing in a journal, or waiting for Monday. This is not a fringe habit anymore. RAND recently found that nearly one in five U.S. adolescents and young adults now use AI chatbots for mental health advice, sharply up from the year before. A Brown University study of teens found much the same. And when people are asked why they chose an AI over a person, the biggest reason is not cost or a waitlist. It is fear of judgment: the quiet calculation that a machine will not remember this later, will not bring it up at a family dinner, will not look at you differently next week.

That calculation is understandable, and it is not wrong, exactly. It is worth noticing what it trades away. The same things that make a chatbot feel safe (it will not remember, it will not judge, it starts fresh every time, and besides, it is not a real person) are the same things that limit what it can ever become to you. So the question deserves an honest answer rather than a defensive one: is ChatGPT actually useful for this?

Yes. And also: useful for what, exactly?

The same question, two answers

Here is a small experiment. Ask ChatGPT and Wyser the same real question, and read the two replies next to each other.

Read both again and the shape of it is clear. One answers the question you typed. The other seems to suspect there is a real situation sitting behind it, and stays there.

How it lands

Go back to that one sentence: the person who’s willing to be clear and accountable ends up feeling like the only adult in the room.

If you have ever been that person, the one who apologises first, who owns the mistake before anyone asks, who does the emotional cleanup after an argument that was not only yours, that sentence is not new information. You already knew it. What you probably did not have was the exact words for it. That is the whole trick, and it is not really a trick: naming the specific, slightly bitter version of a feeling does something a complete and accurate list of six reasons cannot, which is to make you feel caught out and understood in the same second. One answer teaches you something true about people in general. The other risks being wrong about you personally, in exchange for the chance of being exactly right.

That is also why it can afford to say less. Six reasons is thorough. One sentence that lands is worth more than six that glance off, and a system built to be exhaustive rarely stops at one, because stopping at one is a bet on which single line will hit hardest for the specific person reading it.

Here is a second example, and a better one, because the excavation is visible in it. It is my own, which is the only kind I can show you: conversations on Wyser are accessible to you alone, and we do not read them. So this one is mine. Testing Wyser one evening, I asked whether the whole of human suffering traces back to a single need for love and attention, and how you would even justify a need like that. The reply did not begin with the philosophy of need. It began by noticing the question:

Yeah, that need for love and attention is real. And "justify" is already a slightly harsh frame, like you’re asking for permission to be human.

I had not asked about the word "justify." I had not noticed choosing it. But it was the most revealing thing in what I wrote, because it quietly assumed the need required a defence, and anyone who believes that about their own longing is saying something more important than the philosophical question on the surface. Answering what I actually typed would have been correct, and would have missed me entirely. Wyser answered the question underneath it instead, and only then went on to the substance.

None of that happens by luck, and it is not a clever prompt either. Getting a sentence to carry a mood, knowing when to stop explaining and simply say the true thing plainly, reaching for the word a person would actually use rather than the clinical one, hearing the ask underneath the ask: this is closer to editorial and psychological craft than to computation. What sits behind Wyser’s voice is years spent in philosophy, particularly the traditions concerned with how people actually come to know themselves, paired with a working grounding in the psychology research on how people process emotion, what makes a reflection land instead of glance off, and when less does more. That combination, of what to say and just as much what to leave unsaid, is not a setting anyone can switch on over a general-purpose model. It is the actual product.

What gets lost when nothing is remembered

There is a cost to the safety of a machine that forgets, and it is easy to miss, because it does not show up in any single conversation. It shows up in the tenth one.

Telling your story to something with no memory of the ninth time is its own quiet kind of exhausting, even when it does not feel like it in the moment. Every conversation restarts the account from nothing: here is who I am, here is what is going on, here is why this keeps happening. A friend who had heard it nine times would eventually say the thing that only makes sense with all nine in view. A system with no memory cannot, structurally, ever get there. It is not holding back. It has nothing to hold.

That matters because most of what actually needs untangling does not live inside one conversation. It lives in the gap between them: the anxiety that arrives every Sunday before a heavy week, the apology offered the same way each time, the version of "I’m fine" reserved for one particular kind of question. None of that is visible from inside a single exchange, however well answered. It becomes visible only to something that was there for the others too.

What else is different

Ask ChatGPT the same question next month and you will get a new, equally well-organised version of the same list, because it has no way of knowing this is the third time you have asked something in this shape.

Wyser is built on the opposite bet: that the same worry surfacing again matters more than any single answer to it. In practice that looks like this. You mention, again, that you are anxious about work. Instead of fresh generic advice, the reply notices what is repeating: *"I’ve noticed this worry about being judged has come up in a few of our conversations now. What could that worry be trying to protect you from?"* Nothing about those words is clever. What matters is that they could only be said to you, because they rest on everything you have already said.

It also means growth does not have to look like a straight line to count. A pattern that fades for a month and quietly returns is not a failure to hide from something that is actually watching for it. It is what real change usually looks like. A tool with no memory cannot tell the difference between backsliding and a bad day, because it cannot see either. There is a decent analogy between how Wyser works day to day and how Duolingo works: nobody doing their daily lesson is thinking about spaced repetition, they simply notice the exercises seem to know what they are bad at. You are not meant to see any of it. You are meant to feel, eventually, that something has been paying attention.

To be fair to ChatGPT

None of this makes ChatGPT bad at what it does. For research, drafting a hard email, understanding a diagnosis you have just been given, or getting unstuck on a decision with no clean answer, it is genuinely one of the best general tools ever built, and nothing here argues otherwise. Even the clinicians CNBC spoke to on this exact question do not say avoid it. They say use it for what it is good at, and notice where it runs out.

And it does run out, as anything would when it is built to be excellent at a fresh conversation rather than to carry the last one forward: the fourth time you raise the same fight, the pattern you cannot see because you are standing inside it, the "I’m fine" you only give to one particular kind of question. That is not a flaw awaiting a future update. It is simply not the job the tool was built to do.

Side by side

Try it yourself

Take whatever is actually on your mind this week. Ask ChatGPT, then ask Wyser. Not to see which is smarter, because they are not really competing on that. Notice instead which one you would want to return to in a month, once it has had the chance to notice something about you that you have not said out loud yet.

The question you type is rarely the whole of what you came to say.