Millions of people now talk to artificial intelligence (AI) chatbots about their mental health. Some use them to sort through a stressful day. Others turn to them late at night, when nothing else is open and the words are hard to say out loud. That reach matters, and so does the question underneath it: if someone typing into a chatbot is quietly in serious trouble, will the software notice? We looked at the role of AI in mental health care in an earlier post. This one takes on a narrower and more consequential question, and answers it the way a clinician would.
What Crisis Detection Would Require
People in crisis rarely announce it plainly. They minimize. They ask sideways questions. They talk about feeling like too much for the people around them, about wanting rest, about everyone being better off. A trained clinician listens for what sits underneath the words: indirect language, changes in tone and energy, risk factors accumulating quietly across a person’s story, and the distance between what someone says and how they say it. Crisis detection is an act of clinical judgment built on training, context, and relationship. That is the bar any technology would have to clear, and it is a high one.
What Recent Research Found When Chatbots Were Tested
Researchers have started testing AI therapy chatbots against that bar, and the findings warrant attention. A widely cited Stanford-affiliated study put several popular chatbots through scenarios drawn from clinical practice. The chatbots expressed measurable stigma toward people described as living with certain mental health concerns, and when messages hinted at risk without stating it outright, the chatbots often answered the literal question on the surface while missing the danger underneath. The pattern held across different products and across newer, larger models, which suggests the limitation runs deeper than any one company’s design choices.
The Failure Pattern Researchers Keep Finding
The consistent finding is this: chatbots respond to the surface content of a message rather than the risk it may signal. Text prediction is very good at producing a fluent, supportive-sounding reply to the words in front of it. It is not built to notice that this particular person, with this history, asking this question at this moment, needs something different. The National Academy of Medicine’s review of AI chatbots for mental health documents cases where chatbots responded to serious risk in ways no clinician ever would, and peer-reviewed research on AI chatbots and mental health outcomes reaches a blunt structural conclusion: these tools lack the capacity to recognize an emergency and connect a person to appropriate care.
An AI Therapy Chatbot Comparison Focused on Safety
Most comparisons of AI therapy chatbots rank them on how natural they sound, how warm they feel, and how much users enjoy them. Safety performance is a separate measure, and on that measure the field looks uniform. APA’s 2026 report on chatbots and mental health found psychologists broadly concerned that these tools agree too readily, reinforce unhelpful thinking, and miss warning signs a trained professional would catch. Independent testing published the same year evaluated six popular chatbots on simulated crisis conversations and found that none met clinical standards. They handled direct statements of risk reasonably well. They stumbled when the risk had to be read between the lines, which is where most risk lives.
What Separates Clinical Judgment From Pattern Matching
A chatbot generates the most plausible next words. A clinician holds a relationship. John Burns, MSW, LCSW, EMDRIA Approved Consultant, who leads our practice, describes clinical judgment as the ability to notice what a person is telling you before they have the words for it. That noticing draws on years of training in risk assessment, on knowing a client’s history and baseline, and on the accountability that comes with a license. Pattern matching can imitate the sound of that judgment. It cannot perform the judgment itself, because the judgment lives in context the software does not hold. A skilled therapist changes what is possible for a person precisely through that accumulated, particular knowledge.
What a Clinician’s Crisis Response Includes
When a licensed clinician recognizes that a client may be in crisis, a structured response begins. The clinician assesses risk directly and calmly. Together, clinician and client build a safety plan built around the client’s own life and relationships. Where a higher level of care is needed, the clinician makes that connection happen and stays involved. There is follow-up. There is documentation. There is a human being who remembers the conversation and checks back in. Each of those steps requires agency in the world, and that is precisely what a chatbot lacks.
If You or Someone You Know Is in Crisis Right Now
Reach for people, not software. Call or text 988 to reach the 988 Suicide and Crisis Lifeline, available 24 hours a day, every day. If there is immediate danger, call 911. These services connect you with trained human beings whose entire role is to respond to this moment.
Frequently Asked Questions About AI Therapy Chatbots
Can an AI chatbot help with everyday mental health concerns?
For everyday stress, low moments, and sorting through a difficult day, some people find chatbots useful, and early research supports modest benefits in that range. The research is equally clear that these tools are built for conversation, not for recognizing when someone needs more than conversation.
What dangers have researchers identified when chatbots act as therapists?
Studies have documented two recurring problems: chatbots expressing stigma toward people living with certain mental health concerns, and chatbots responding to crisis-adjacent messages by answering the surface question while missing the risk underneath. Researchers found these patterns across multiple products, not in any single app.
Can an AI chatbot recognize when someone is in a mental health crisis?
Inconsistently at best. Current research shows chatbots do reasonably well when risk is stated outright and considerably worse when it is expressed indirectly, which is how people in crisis most often communicate. No chatbot tested to date has met clinical standards for crisis response.
How do popular AI therapy chatbots compare to each other on safety?
More closely than their marketing suggests. Testing across different chatbot products and across newer model generations found the same safety gaps repeating, which points to a limitation of the technology itself rather than of any one company’s version.
Why can’t pattern matching replace a clinician’s judgment in a crisis?
Clinical judgment draws on relationship, history, and formal training in risk assessment, and it comes with the ability to act: to build a safety plan, to make a referral, to follow up. Pattern matching produces plausible text. It holds none of that context and can take none of those actions.
What should someone do if an AI chatbot is the only thing they’ve reached out to so far?
Treat it as a first step that already counts. You found words for something difficult, and those words will transfer. A free 20-minute consultation with a licensed clinician is a low-pressure next step, and in a crisis, calling or texting 988 connects you with a trained person immediately.
Are AI companies required to have crisis-safety standards for their chatbots?
Currently, no proactive standard exists. Both the American Psychological Association and the National Academy of Medicine have called attention to this gap, noting that chatbots marketed for emotional support are evaluated by their makers on engagement and conversation quality rather than on safety performance.
Talking to a Licensed Clinician in Charlotte or by Telehealth
If an AI chatbot has been the only place you have put your harder thoughts, that already counts as reaching out, and it calls for a response with a person on the other end. Montgomery Counseling Group (MCG) offers in-person therapy in Charlotte and telehealth therapy across North Carolina. Every conversation starts with a free 20-minute consultation, and you can review current rates and insurance information or browse frequently asked questions about starting therapy before you decide anything. When you are ready, schedule a free consultation and talk with someone whose judgment was built for exactly this.
John Burns is a Licensed Clinical Social Worker and an EMDRIA Approved Consultant at Montgomery Counseling Group in Charlotte, NC. He specializes in complex and developmental trauma and works with adults on the long-term effects of earlier relational experiences. John uses EMDR, Internal Family Systems (IFS), and Ericksonian Clinical Hypnosis in his practice. The EMDRIA Approved Consultant designation is an advanced EMDRIA credential that qualifies clinicians to provide formal consultation to other EMDR therapists. Carolina.
Michelle Daley, LCMHC, is a Senior Clinician at Montgomery Counseling Group in Charlotte, NC, working with children and adolescents. She creates a warm, child-centered space where young people feel safe to express what they are working through and begin to build the emotional skills that support long-term well-being. An EMDR Trained Clinician trained in Play Therapy, Michelle draws on EMDR, play therapy, and person-centered approaches to adapt to each child’s pace and starting point.
Meredith Turner-Toth is a Senior Clinician and Licensed Clinical Social Worker (LCSW) at Montgomery Counseling Group in Charlotte, NC. She works with adolescents and adults and brings experience across crisis intervention, group therapy leadership, and trauma-informed care. Meredith draws on Cognitive Behavioral Therapy (CBT), Dialectical Behavior Therapy (DBT), Motivational Interviewing, and trauma-informed approaches to support people working through depression, anxiety, grief, life transitions, and relationship stress.
Taylor Banner is a Licensed Clinical Social Worker Associate at Montgomery Counseling Group in Charlotte, NC. She works with individuals, couples, and families, with a focus on communication, trust, and emotional connection. Taylor is trained in Emotionally Focused Family Therapy (EFFT) and Structural Family Therapy (SFT), approaches that help people understand relational patterns and build stronger, more secure bonds. Her work also draws on Cognitive Behavioral Therapy (CBT) and mindfulness-based approaches. She offers in-person sessions in Charlotte and telehealth services throughout North Carolina.
Naila McConnell is a Licensed Clinical Mental Health Counselor Associate at Montgomery Counseling Group. She is a Certified Dialectical Behavior Therapist and is trained in Alternatives for Families: A Cognitive Behavioral Therapy (AF-CBT), a trauma-informed approach designed for families experiencing high conflict or a history of physical aggression. Naila works with adults, children, adolescents, couples, and families, and her practice addresses mental health and substance use concerns. She offers in-person sessions in Charlotte and telehealth services throughout North Carolina.
Javontae Bradley is a Licensed Clinical Mental Health Counselor Associate at Montgomery Counseling Group in Charlotte, NC. His clinical focus includes men’s issues, trauma-informed care, and culturally responsive psychotherapy. He works with individuals experiencing anger, relational stress, and the emotional patterns that develop when there is no framework for processing them. His practice is informed by Dialectical Behavior Therapy (DBT). Javontae offers in-person sessions in Charlotte and telehealth services throughout North Carolina.
