The miracle question is a specific intervention developed by Steve de Shazer and Insoo Kim Berg at the Brief Family Therapy Center in Milwaukee in the 1980s. In its original form, the solution-focused practitioner asks: Suppose tonight, while you sleep, a miracle happens. The problem that brought you here is gone. You wake up tomorrow — what is the first small thing you notice that tells you something has shifted? The question is not rhetorical. It is a structured invitation to describe a desired future in concrete, observable terms, without first having to understand or explain the problem. I am a systemic counselor in Vienna, licensed under Austrian Lebens- und Sozialberatung, trained in part within the European solution-focused tradition that grew out of de Shazer and Berg’s Milwaukee work. I use variants of this question regularly. And over the past year, I have started using it with people who write to me uncertain about their own AI use — people who are not sure whether what they are doing is a problem, and who do not yet want to name it one.
The problem with “am I addicted?”
Most people who contact me about AI overuse arrive at the question sideways. They are not searching for a diagnosis. They have noticed something — a pull toward the chat window when they feel unsettled, a preference for the AI’s responses over a friend’s, a slight shame about how much time they have spent in a conversation that no one else can see. They want to know whether this is a problem. They often frame it as a question about addiction, because that is the vocabulary available to them.
The difficulty with the addiction frame is that it is deficit-first. To ask “am I addicted?” is to begin with what might be wrong, and the self-assessment that follows tends to oscillate between minimising (“it’s not that bad”) and catastrophising (“I have completely lost control”). Neither position produces much useful information about what the person actually wants their relationship with AI to look like. The miracle question approaches the same territory from the other direction.
How to use it
The version I would suggest for self-reflection, adapted from de Shazer’s original Milwaukee formulation, goes roughly like this:
Suppose you go to sleep tonight, and while you sleep, something shifts in how you relate to AI — not that you stop using it, but that the quality of the relating becomes exactly right for you. You wake up tomorrow. What is the first small thing you notice that tells you something has changed?
The answer almost never comes back as a number of hours. It tends to come back as a texture. “I would open it when I wanted to, not because I needed to.” “I would close it without feeling like I had missed something.” “I would tell my partner I had been talking to it, without that feeling complicated.” “I would use it for the task, finish, and go back to what I was doing before.” These descriptions are diagnostic in a way that a usage log is not: they tell you what the person already knows a more functional relationship would look like, which means they already have a picture of the destination.
What the gap tells you
Once you have a description of the miracle morning, the next useful move is not to measure how far you are from it, but to ask: on a scale of one to ten, where ten is the miracle morning, where are you right now? And then: what is already working — what is already at that number and not lower? The solution-focused tradition, from de Shazer’s work to the German-language practice of Gunter Schmidt and the Austrian practitioners who trained me, puts as much weight on what is already present as on what is missing. The exceptions — the moments when the problem does not fully show up — are often more informative than the problem itself.
If you imagine a morning where you opened the AI tab with clear intention rather than reflexive pull, and then you sit with when that has already been true — even partly, even once this week — you have the beginning of a practical answer. Not a diagnosis. A direction.
Where the question runs out
The miracle question is a self-reflection tool, not a clinical instrument. It works well when the person using it is in reasonable contact with their own experience and has enough reflective space to answer honestly. It is less useful when the use pattern is severe enough that the person cannot reliably observe it, or when what is underneath the AI use is something the tool itself cannot hold — acute grief, relational trauma, the kind of sustained disconnection from other people that needs in-room support rather than a writing-based practice.
I am describing this here because I think there is a real gap between “I am fine, this is nothing” and “I need clinical help” — and most people who write to me about AI overuse are somewhere in that gap, in ordinary ambivalence about something that is neither fine nor catastrophic. For that territory, this kind of structured self-reflection is appropriate, and it is where the work I do is most useful.
If what you are describing is closer to the severe end — if the AI has genuinely replaced all human contact, if the pull to it feels physically compulsive rather than habitual, if you are in a mental health crisis and looking for support — this is not the right resource. Emergency services are reachable at 112 in the EU; findahelpline.com has international referral directories. What I offer is not psychotherapy and does not address diagnosis or mental illness.
Trying it in writing
Writing your answer to the miracle question tends to produce a more useful response than thinking it. The act of committing words to a page introduces a small reflective distance — you can read back what you wrote and notice whether it is actually what you meant. This is part of why the asynchronous email format I use at Relating to AI works for this kind of question: the writing itself is doing some of the thinking.
If you want to try the question in a structured way, the free Wunderfrage on this site is a short series of questions designed for exactly this territory — not a diagnosis, not a screening tool, but a starting point for getting your situation into words. A licensed human signs every reply, not a chatbot. You do not need to have decided whether you have a problem to use it.