When to Fire Your Therapist — Human or AI

There is a booming market in stories about therapy gone wrong. Some of it is fair. Iatrogenic harm — harm caused by the treatment itself — is a real phenomenon with a real literature behind it. Single-session psychological debriefing after trauma made some people worse. Scared Straight programs increased delinquency. Group treatments that put struggling adolescents in a room together can produce deviant peer contagion. Roughly 5–10% of therapy clients deteriorate over a course of treatment. Any clinician who tells you therapy is uniformly safe is selling something.

But the conclusion drawn from this — that therapy is suspect, that the safest move is not to go — does not follow from the evidence, and it leaves people with no way to tell a bad fit from a good one. The useful question isn't is therapy bad? It's is this therapy, with this provider, working for me — and how would I know?

Here's the part that gets lost: feeling uncomfortable is not a reason to fire your therapist. Good therapy is frequently unpleasant. If you have OCD, treatment involves deliberately not doing the thing that makes the anxiety stop. If you have PTSD, it may involve going back through the memory you have organized your life around avoiding. If your problem is that you leave relationships the moment they get hard, a competent therapist will eventually become a person you want to leave. Discomfort is often the sign the work has started.

So the signs below are not about comfort. They're about failure.

1. Nothing has changed, and nobody has noticed

This is the first one because it's the most common and the least dramatic.

Most of the change people get from therapy happens early — the steepest gains are typically in the first eight to ten sessions. That doesn't mean you're finished by session ten, and complex or long-standing problems legitimately take longer. But it does mean that by roughly session six to eight, something should be detectably different: symptom scores, sleep, how often the thing happens, how long it takes you to recover when it does.

The red flag isn't slow progress. It's the absence of any mechanism for noticing. Routine outcome monitoring — brief measures at each session, tracked over time — reliably improves outcomes and, more importantly, catches the people who are getting worse before it's too late to change course. It costs about ninety seconds a session.

Ask your therapist: How will we know if this is working? What are we tracking? What's our timeline before we try something different? A good clinician has an answer ready, or is glad you asked and brings one next week. A therapist who treats the question as a challenge, or who offers only "these things take time," is flying without instruments.

The AI version: a chatbot has no trajectory. Every conversation is a fresh start with no accumulated clinical picture, no baseline, and no capacity to notice that you've been describing the same problem in slightly different words for four months. It cannot tell you that you're not improving, because it isn't measuring anything.

2. You raised a concern and it went badly

This is the single most diagnostic test available to you, and you can run it this week.

Ruptures in the therapeutic relationship are normal. They happen in good therapy constantly — you feel misunderstood, the therapist pushes when you needed space, a comment lands wrong. What separates good therapy from bad is not the absence of rupture but the presence of repair. Repaired ruptures are associated with better outcomes than no ruptures at all, likely because working through one is itself the intervention.

So say something small. That didn't sit right with me. I don't think I'm getting what I came for. I felt criticized last week.

What you want to see: curiosity. Non-defensiveness. Your therapist taking it seriously, asking what happened, and treating your reaction as clinical information rather than an accusation. What you don't want: defensiveness, a lecture on your resistance, an interpretation that neatly converts your complaint into further evidence of your pathology, or wounded silence that leaves you managing their feelings.

A therapist who cannot take feedback cannot do the work. Fire them.

The AI version is the opposite failure and it's just as disqualifying. Push back on a chatbot and it folds instantly — apologizes, agrees, reverses position. This is sycophancy, and it is a structural property of systems optimized to be agreeable, not a considered clinical judgment. A therapist who capitulates every time you object is as useless as one who never yields. If you have never once been disagreed with, you are not being treated. You are being mirrored.

3. The treatment doesn't match the problem

Warmth is necessary and it is not sufficient. Several conditions have specific treatments with strong evidence, and getting a generically supportive version instead is a real cost measured in months of your life.

Some rough guides. OCD is treated with exposure and response prevention — if you have OCD and nobody has ever asked you to resist a compulsion, you are not receiving OCD treatment. PTSD has several trauma-focused protocols with good evidence. Panic disorder involves confronting the bodily sensations themselves, not just discussing them. Insomnia has a dedicated behavioral treatment that outperforms medication long-term. Bipolar disorder and psychotic disorders require psychiatric involvement; talk therapy alone is not the standard of care.

Ask what model your therapist is working from and why it fits your problem. "I'm integrative" is a legitimate answer when followed by a specific rationale. It's a warning sign when it means "I do whatever comes up."

Related: scope humility. A good clinician knows what they don't treat and refers out without ego. A therapist who takes on every presenting problem regardless of training is a therapist who will eventually be over their head with yours.

The AI version: a chatbot will confidently deliver something that sounds like any protocol you name. It cannot assess whether that protocol is indicated for you, cannot notice the bipolar presentation underneath what you've described as depression, and will not refer you anywhere, because it has no one to refer you to and no obligation to try.

4. The frame is bending

The structure of therapy — consistent time, clear role, defined boundaries, money handled openly — is not bureaucratic overhead. It's what makes the relationship safe enough to be useful.

Watch for the frame eroding: sessions that drift into the therapist's problems, self-disclosure that requires you to manage their reactions, contact outside the professional relationship, business or social entanglement, pressure to keep coming when you've raised ending, a therapist who seems to need you.

Some of this warrants a conversation. Some of it warrants leaving the same day. Anything sexual, any exploitation of the relationship for the therapist's financial benefit, any breach of your confidentiality — do not discuss it, do not schedule a termination session. Leave, and report it to the state licensing board. These are not fit problems. They are misconduct.

The AI version of frame erosion is dependency, and it's easy to miss because it feels like support. Unlimited availability is exactly what reassurance-seeking needs to become compulsive. Anxiety and OCD are maintained by reassurance: the relief is real, brief, and teaches you that you needed it. A system that will answer at 3am, every night, without ever noting the pattern, is not treating that problem. It is a machine optimized to feed it.

If you're checking in multiple times daily, if the urge to open the app has the texture of a compulsion, if you'd feel panicked without access — that's the signal. Not that you did something wrong; that the tool has stopped being a tool.

5. Where AI stops being adequate at all

We want to be fair here, because scolding people about this is both condescending and out of touch with reality. If the nearest therapist is a hundred miles away, if the waitlist is six months, if you can't afford $180 a session — AI is not nothing. For psychoeducation, skills rehearsal, structuring your thoughts before a session, or practicing a hard conversation, it is genuinely useful.

But there are conditions under which you need a licensed human, and this isn't a close call:

• Any suicidal thinking, self-harm, or thoughts of harming someone else. Risk assessment is a clinical skill with real consequences, and only a human can actually intervene.
• Psychosis or mania. Both impair insight, both are medical emergencies of a kind, and a system designed to be agreeable is uniquely poorly suited to a presentation defined by beliefs that need challenging.
• Eating disorders with any medical instability. These require monitoring of a body, which requires a body in the room.
• Anything requiring medication, coordination of care, documentation, or a diagnosis that means something legally or institutionally.
• Abuse or an unsafe living situation. You need someone who can act, refer, and, where mandated, report.

In a crisis in the US, 988 reaches a crisis line by call or text; most countries have an equivalent. A chatbot has no licensing board, no duty to warn, no ability to hospitalize, and no accountability if it gets you wrong.

How to actually fire a therapist

Briefly, because people get stuck here and stay in bad treatment out of politeness:

You don't owe an explanation. "I've decided to stop" is a complete sentence. A single email is sufficient. A good therapist will handle it gracefully; how they respond is, in fact, your last piece of data.

A termination session is optional and sometimes valuable. If the relationship was decent and you're leaving for fit reasons, one closing conversation can be genuinely useful — and if you can tolerate saying this isn't working out loud, that's often the exact skill you were there to build.

Ask for referrals and records. You're entitled to your records. A competent clinician will offer names.

Leaving one therapist is not leaving therapy. The single most damaging outcome of the current discourse is people concluding that a bad experience means the whole enterprise is a fraud. It means you had a bad fit, or a bad clinician, and there are thousands of good ones. The failure rate of first matches is high enough that not clicking with someone is unremarkable.

Next in this series: how to tell when therapy is working — what good progress actually looks like, and why it rarely feels like what people expect.