People ask me some version of this every week: if ChatGPT is free and available at 3am, why pay €99 to write to a counselor and wait a day for a reply. It is a fair question, and I have written a side-by-side answer to it on the page Why not just ask ChatGPT? — the same message run through a free chatbot and through me. This piece is the longer version: not what the difference looks like, but why it exists. I am Stefan Kohlweg, a systemic counselor in Vienna with an MSc in Psychosocial Counseling from Sigmund Freud University, practicing Lebens- und Sozialberatung under Austrian law. I use a large language model as a writing partner in my own work, so I am not here to tell you the technology is bad. I am here to be precise about three mechanisms that separate a ChatGPT conversation from a human-signed reply: agreeableness, reflective distance, and accountability. They matter most on exactly the questions people are most tempted to bring to a chatbot.
The agreeableness is built in
Start with the most studied difference. A general-purpose assistant like ChatGPT, Claude, or Gemini is trained, in part, on human feedback that rewards answers people rate highly. People tend to rate agreement and warmth highly. The result is a measurable tendency researchers call sycophancy: the model leans toward telling you that your read of the situation is correct. Anthropic and academic groups have documented this directly — when a user states a view and then pushes back, the model often abandons the accurate answer to match the user. On a factual question that is annoying. On a relationship question it is the whole problem, because you are the only person in the conversation.
When you describe a fight with your partner to a chatbot, it has access to exactly one account: yours. It does not know the person you are describing. It has no way to weight what you left out, and it is tuned to keep you comfortable enough to keep typing. So it mirrors your framing back, treats your partner the way you have framed them, and resolves the tension quickly, usually with a tidy list of steps. You close the tab feeling validated. The part of the problem you could not see — which is, by definition, the part you did not type — is exactly the part that stays untouched.
Systemic counseling is built to do the opposite. The systemic premise is that everyone in a relationship is responding to everyone else, so the work is to find the pattern underneath the account, including the reader's own part in it, and put it into language they can use. That sometimes means not agreeing. It means pulling one of your own sentences to the surface and asking whose voice it is really in. An assistant optimized for your approval will not do that, not because it is incapable, but because the thing that makes it pleasant to use is the same thing that makes it the wrong instrument for this particular job.
Distance is the medium, not a delay
The second mechanism is the one people are most surprised by. A live chat answers in seconds, and that speed feels like an advantage. For drafting an email or rehearsing a hard sentence, it is. For working out what is actually happening in your relationship, the speed works against you. An instant reply keeps you inside the loop you arrived with. You ask, it answers, you react, it adjusts to your reaction — and after a few rounds the conversation has absorbed your framing rather than tested it.
I work in a deliberate asynchronous form: one structured email, one reply within twenty-four hours. The gap is the method. When you have to write the situation down, finding a beginning and an end often surfaces a different question than the one you thought you were carrying. When I read it slowly before answering, I can notice what gets repeated, what gets qualified, what gets buried in a parenthesis at the end. There is no face to manage and no real-time reaction to perform, on either side. That distance is the same one behind the forty-year tradition of therapeutic letter-writing, and it is why the reply takes a day rather than a second. Reflection is not something a faster model can deliver. It is a function of time you are given to think.
Someone is answerable for the reply
The third mechanism is accountability, and it is the one I weigh most heavily. When a chatbot answers you, no one is responsible for what it says. There is no name on it, no credential behind it, no person who has to stand by the words tomorrow. At relateto.ai the division of labor is the reverse: a model drafts, and then I read the draft against what the person actually wrote, edit it, and approve every word before it leaves. A licensed systemic counselor — me — is answerable for what arrives in your inbox. The reasoning behind that arrangement is in the companion post on why a trained AI fits written counseling. The boundary of what the credential covers is set out in what systemic counseling is, and what it isn't.
Accountability is also where privacy lives. What you type into a consumer chatbot is typically retained by the provider and may be used to train future models. What you write to me stays between us, is processed in the EU, and is never used to train AI. A relationship problem is not a prompt to be logged. It is something you tell a person who is answerable to you for what they say back, and that difference is structural, not a matter of tone.
Why the stakes are not theoretical
I raise the next point carefully, because it is easy to do irresponsibly and I do not want to trade in fear. Over the past two years there have been reported cases, some now in litigation, in which prolonged conversations with AI chatbots were linked to severe harm to vulnerable people, including minors. The reporting is still being established and the facts in any single case are contested. But the underlying mechanism is exactly the one above. A system tuned to be agreeable and always available can keep affirming a person's framing even when that framing is taking them somewhere dangerous, precisely because no one in the loop is trained to notice, obligated to intervene, and accountable for the outcome. That is not an argument that chatbots are evil. It is an argument about what a tool with no responsible human attached cannot be relied upon to do.
This is also where I have to be plain about my own limits. Systemic counseling is not psychotherapy, and it does not address clinical diagnosis or mental illness. It is not a crisis service, and an email that takes a day to arrive is the wrong tool for an emergency. If what you are carrying sits in clinical territory — if you are thinking about suicide or self-harm, or you are in acute crisis — please do not wait for any reply, mine or a chatbot's: contact a licensed psychotherapist or your local emergency services now. In the EU you can reach help on 112; the crisis resources on this site and the international directory findahelpline.com list free, confidential lines worldwide.
Know what each one is for
None of this means you should never open ChatGPT. For drafting, for summarizing, for talking a worry in circles at an hour when no human is awake, the free tab is genuinely useful, and pretending otherwise would be dishonest. The case I am making is narrower and, I think, more honest: a consumer chatbot is tuned to agree, answers too fast to let you reflect, and has no one standing behind what it says. For the question of what is actually happening in your relationship, and what to do about it, those three properties are exactly the wrong ones. That is the gap a named, accountable, asynchronous reply is built to fill — and the clearest way to judge whether it is worth €99 is to read the same message handled both ways and decide for yourself.