Most people asking this question have already tried pasting a problem into a general chatbot and found it strangely unsatisfying. It was articulate. It was encouraging. It gave you five numbered tips. And somehow you closed the tab feeling exactly as stuck as when you opened it.
That experience is worth taking seriously, because it points at the real question. Not whether an AI can coach, but what coaching actually is, and whether the thing in front of you is doing it.
What coaching is, and is not
Coaching is not advice. A good coach spends most of the session listening and asking, and comparatively little of it telling. The reason is not modesty. It is that the answers people actually act on are the ones they arrived at themselves. An answer handed to you is something you evaluate. An answer you reached is something you own.
This is why the five numbered tips felt hollow. They were advice, delivered instantly, about a situation the machine had known about for nine seconds. No human coach worth paying would do that.
Where AI genuinely beats a human coach
There are three places where the machine is not merely cheaper but actually better, and they are all about availability rather than intelligence.
- It is there at 3am. Coaching moments do not arrive in the hour you booked. They arrive when you cannot sleep, when the email lands, when you are sitting in the car outside the house not going in yet. A human coach is, at best, available for one of those hours a fortnight.
- It has no ego and no schedule. You are never managing its feelings, never worrying you are boring it, never conscious that the clock is running at a few pounds a minute. People say things to it they would take three sessions to say out loud to a person.
- It remembers everything, exactly. Not the gist. The actual thing you said about your mother in week two, ready in week nine.
Where it does not
A skilled human coach reads the pause before you answer. They notice the thing your face does when you say the word fine. They can sit in silence for forty seconds because they know the silence is doing the work. None of that survives a text box.
A human also carries accountability of a kind software cannot: the specific discomfort of telling a person you respect that you did not do the thing. That discomfort is a feature, and it is genuinely load-bearing for some people.
How to test one in five minutes
Forget the marketing. Open it and tell it something real but small, with a detail you are slightly embarrassed by. Then watch what comes back. You are looking for four things.
- Does it reflect before it directs? The first response to anything heavy should show you it understood, not hand you a plan.
- Does it ask one question, or stack three? Three questions means it is filling space, and you will answer only the easiest one.
- Does it push back? Say something self-critical and absolute, like everyone else finds this easy. A coach questions that sentence. A chatbot agrees with it warmly and moves on.
- Does it land on something you could do today? Not a framework. One action, small enough that you would actually do it.
Most tools fail the third test, and it is the one that matters most. Agreeable is the default failure mode of language models, and agreeable is the opposite of coaching. If it never once makes you slightly uncomfortable, it is not coaching you. It is keeping you company.
The question underneath the question
Often what people mean by is it any good is really will it work for me, and that turns on something simpler than technology: will you use it. A brilliant coach you see twice and abandon does nothing. A decent coach you talk to for four minutes most evenings changes your year.
So when you are judging one, judge the thing you will actually do. Can you open it without preparing? Does it ask so little of you that a bad day does not break the habit? Does it close the conversation and let you sleep, or does it need you more than you need it?