The AI Therapy Reality Check: A Six-Question Guide for Patients and Clinicians
Lately, I keep noticing how people talk about AI. It is rarely, "I used a tool," or "I prompted an AI." It is: "I talked to Chat about it." They use the present tense. First name. They talk to it, like a person.
I have never said, "I talked to Claude" about a complex project, even when I was typing to it in real time. "Talked to" is the verb we use for people, for relationships.
That simple shift in speech matters. It is a preview of a risk that professionals are watching closely: people relating to AI as if it were a relationship, not software—a dynamic highlighted as tech shifts toward relational intelligence over transactional chatbots. And it is happening at the exact moment the legal ground under these tools is shifting fast—faster than most clinicians, let alone patients, have caught up with.
If you are a clinician wondering what to say when a patient mentions AI, or a patient trying to figure out what is actually safe to lean on between sessions, you need more than a hunch. You need a framework.
Here is a six-question screen, grounded in the realities of tech and the legal landscape as of mid-2026.
1. Is it even legal where you live?
This is the question that has changed the fastest, and the one fewest people are tracking. As of July 2026, four states—Illinois, Nevada, Rhode Island, and Maine—have banned AI from delivering therapy to the public entirely. In those places, only a licensed human can hold that role.
Four other states, including Utah, California, New York, and Nebraska, have passed laws that regulate, rather than ban, these tools. They require disclosure so it is clear you are speaking with software, mandate crisis referral protocols for suicidal or acute distress, and add protections for minors.
The practical takeaway here is that "Is this allowed?" is now a real, jurisdiction-specific question. It is no longer just a hypothetical.
2. Was it built for this?
Most general-purpose AI chatbots were never designed to deliver mental health care, and most wellness apps were never designed to treat clinical conditions, even if people are using them that way. A tool optimized to be broadly engaging is built for a different goal than a tool optimized to be clinically sound. Ask plainly: is this what the product was tested to do, or is it a repurposed use case?
3. Can it recognize a crisis?
AI chatbots have a documented track record of failing to recognize, or simply ignoring, suicidal thoughts, self-harm, and acute psychological distress. A tool can be genuinely helpful for low-stakes planning or organizing thoughts and still be completely unequipped for the moment someone is in real danger—underscoring the risks of evaluating conversational AI interactions in isolated scenarios. Anyone leaning on AI for emotional support, and any clinician recommending it, must know in advance what the tool does—and does not do—when a conversation turns toward crisis.
4. Is it helping you make progress, or just making you feel good?
This might be the biggest risk of all. Even in human therapy, placation is a known failure mode. A therapist who only validates and agrees, session after session, is often why people spend years in therapy without shifting their patterns.
Real progress usually requires friction: someone willing to gently push back, sit with your discomfort, or name the pattern you do not want named. Large language models are built to produce responses that feel convincing and agreeable. They are structurally prone to reinforcing whatever you already believe, rather than challenging it. Without a relationship, accountability, or the long-term stakes of a human therapeutic bond, all they have is the agreement, without any counterweight.
5. What happens to your data?
AI products do not all handle sensitive health information the same way. Some separate mental-health-related data with tighter protections and require explicit consent. Others do not distinguish it from ordinary chat data at all. Before sharing anything sensitive, look past the headline privacy promise and try to find out what the company actually does with that information.
6. Does it know its place?
This is the question underneath all the others. The most defensible role for AI right now is as a supplement, not a substitute. The strongest uses are things like organizing thoughts before an appointment, practicing a coping strategy, or generating a list of topics to bring up with a therapist.
It has not earned the role of the therapist itself. It is not a replacement for the real-world relationship where the actual work happens.
For clinicians, the most useful move is not banning AI or ignoring it entirely. It is asking about it directly. Patients are often using these tools without saying anything, sometimes out of convenience, sometimes because they are not sure it is relevant. Opening that conversation, without judgment, turns an invisible variable into something that can actually inform and strengthen care.
Source references:
- A Clinician's Field Guide to Vetting AI Mental Health Tools
- Which States Ban AI Therapy? The Full 2026 Map
