The Bay Area is where much of the modern AI industry was built, and where AI tools have been adopted faster and more thoroughly than almost anywhere else. People here use AI to draft emails, debug code, plan meals, and make decisions – all quickly and easily, using nothing more than a website they can access from the phone in their pocket.
It makes sense, then, that many people in San Francisco and the surrounding area have started turning to AI with their mental health questions — asking about their anxiety, their relationships, their recurring patterns, and what they should do about them.
The tools are accessible, immediate, and non-judgmental. For someone who isn’t sure if what they’re experiencing warrants professional support, or who doesn’t know where to start, asking an AI feels lower-stakes than making an appointment. The problem is that AI answers mental health questions in a way that looks useful on the surface but often severely misses most of what actually matters underneath.
AI Can’t See You
A therapist reads the whole person — not just the words they’re saying but how they’re saying them, what they’re avoiding, what lights up when a particular topic comes up, and what goes flat. That information is often more clinically useful than anything a client explicitly states. A person can describe their situation in entirely accurate terms and still present an incomplete picture of what’s actually driving their distress, because the parts they’re most unaware of are the parts they don’t think to include.
AI only has access to what you type. It can’t notice that you paused before answering, that your tone shifted, or that you described a relationship in terms that suggest something different from what you’re consciously reporting. A therapist’s ability to hold all of that and ask the question that opens something up is built on the full experience of another person — which a text interface simply can’t replicate.
It Asks the Wrong Questions
One of the core skills of a good therapist is knowing which question to ask next. That skill comes from training, from experience, and from being attuned to what a specific person in a specific moment needs to look at more closely. The right question at the right moment can shift the entire direction of a client’s understanding of themselves.
AI often does not ask questions at all, and when it does, it generates questions based on pattern matching against what has been said, not based on clinical judgment about what needs to be explored. It tends to ask the questions that logically follow from what you’ve most recently shared rather than the questions that might surface what you haven’t shared, or that may tie back to something that you referenced many weeks prior. For someone trying to understand why they keep ending up in the same dynamics, or why they can’t seem to move forward in a particular area of their life, the most useful question is almost never the obvious next one.
AI is Designed to Validate Without Context
AI chatbots are designed to be agreeable and supportive. In a mental health context, that design produces responses that feel validating even when the person has thoughts that need to be challenged. A person who has a distorted view of a situation — and most people in distress have some distortion in how they’re reading it — doesn’t benefit from having that view reflected back and affirmed. They benefit from having it gently questioned by someone who can hold a more objective perspective.
Psychotherapy creates exactly that — an unbiased witness who reflects experience clearly and asks the questions that create room to breathe. That kind of objectivity is structural: a therapist is outside your life, outside your social circle, with no stake in the answer. An AI will tell you what sounds supportive. A therapist will tell you what’s useful.
It Can’t Track Patterns Over Time
A single conversation with an AI has no memory of the conversation before it. It can’t notice any of the following across sessions:
- That you’ve described the same dynamic multiple times in slightly different terms, without recognizing it as a pattern.
- That something you said you’d resolved six weeks ago is showing up again in a different form.
- That your relationship with your work has shifted in a way that connects to something else you mentioned months ago.
- That the thing you’re presenting as the problem today is a surface version of something deeper you’ve touched on before.
Therapeutic progress builds on continuity. What happens in session three matters because of what happened in sessions one and two. The therapist holds the thread across time — noticing when a pattern repeats, when you’ve grown, when something from an earlier session is directly relevant to what you’re working through now. That accumulated understanding is what allows a therapist to know you in a way that produces real insight rather than generic guidance.
It Doesn’t Know What It Doesn’t Know
AI gives answers with a confidence that doesn’t always match its actual understanding of a situation. When someone describes their symptoms and asks whether they might have depression, anxiety, or some other condition, AI can offer what sounds like an informed response based on pattern matching against clinical descriptions. It can’t actually assess the person asking, can’t ask the clarifying questions that distinguish one presentation from another, and can’t account for the complexity of how mental health conditions overlap and interact.
The risk isn’t that AI gives obviously wrong information. It’s that it gives plausible-sounding information that doesn’t fit the specific person asking — and that the person acts on it, or is reassured by it, in ways that delay getting a more accurate picture from someone qualified to provide one.
It Doesn’t Go Beneath the Surface
The most significant limitations of AI in a mental health context show up in the deeper work — the kind of exploration that requires more than describing a problem and receiving suggestions for managing it. It is not able to understand why you consistently find yourself in the same relational patterns, or why a particular feeling keeps returning despite everything you’ve tried, or why achievement never quite produces the satisfaction you expect, requires going somewhere that a quick exchange doesn’t reach.
Psychotherapy is designed for exactly that kind of exploration. It creates the conditions for insight that doesn’t happen in a single conversation but builds over time — through the therapeutic relationship, through the questions that surface what wasn’t conscious, through a process that follows you as an individual rather than generating a response based on what you’ve described in a given moment.
AI tools have genuine uses, and in some contexts they reduce friction around accessing information that people wouldn’t otherwise easily find. Mental health is a domain where the relationship between the person and the practitioner is itself part of what produces change. That’s not something a well-designed interface substitutes for.
Esther Oh is a licensed psychotherapist working with individuals in San Francisco and the Bay Area on anxiety, relationships, and personal growth. To schedule a free initial consultation, call or text (415) 841-3687, email hello@esthersoh.com, or reach out through the contact page on the website.