Is AI Replacing Therapy? An Indianapolis Therapist's Honest Answer

By Becky VanDenburgh, LCSW, LCAC | Think Well Live Well Counseling and Telepsychiatry | thinkwelllivewell.co

The Short Answer: No. But AI Is Already in the Room.

AI therapy refers to the use of generative chatbots and mental health apps to provide emotional support, psychoeducation, or coping strategies. These tools simulate therapeutic conversation using predictive language models, but they are not licensed, not regulated as clinical care, and cannot exercise clinical judgment.

Artificial intelligence is not replacing therapy. It is, however, already sitting in the sessions I have with clients here in Indianapolis. According to the American Psychological Association's 2026 Chatbots and Mental Health Survey of 1,242 licensed psychologists, 77% report that their patients are discussing AI use in therapy. Not curious about it. Using it. And I want to say something clearly before I say anything else: if you have opened a chatbot at two in the morning because your chest was tight and you did not want to wake anyone up, you did not do something wrong. You reached for what was available. That instinct to reach is not the problem. What I want to talk about is what happens next.

What Are People Actually Using AI For?

The APA survey gives us the clearest picture available of how patients are using these tools, based on what psychologists are hearing directly in session. The pattern is not what most people assume. Patients are not primarily asking chatbots to fix them. They are using AI to fill the space between sessions, to organize what they are feeling before they say it out loud, and, increasingly, to seek connection itself. That last category is the one I watch most carefully in my practice, because the line between "this helped me get through a hard night" and "this has become the only place I feel understood" can move quietly, over weeks, without the person noticing it has moved at all.

What Patients Are Using AI For	Percentage of Psychologists Reporting It
Discussing their AI use during therapy	77%
Self-diagnosing a mental health condition	39%
Acting as an additional mental health professional	35%
Self-discipline, affirmations, behavioral reminders	34%
Assisting with their ongoing treatment	33%
Entertainment or fun	33%
Friendship	22%
Intimate or romantic relationships	13%

Source: APA 2026 Chatbots and Mental Health Survey, conducted April 9–26, 2026, among 1,242 licensed U.S. psychologists.

What AI Genuinely Does Well

I am not interested in pretending these tools are useless, because that is not what the evidence shows and it is not what I see. In the same APA survey, 68% of psychologists whose patients used chatbots reported those patients felt supported or validated, and 41% said their patients used AI to reinforce healthy coping skills. Forty percent of psychologists felt optimistic that chatbots could help when a professional is not available. That is a real finding and it deserves to be said plainly. A chatbot can help you name a feeling at midnight. It can walk you through a breathing exercise. It can help you draft the sentence you have been trying to say to your mother for eleven years. In a state where finding a trauma therapist can take weeks, immediate availability is not nothing.

The most rigorous head-to-head comparison we have makes the boundary clear. A randomized controlled trial published in BMC Psychology assigned 104 women with diagnosed anxiety disorders in active war zones to either daily AI chatbot support or three weekly sessions with a human psychologist. Both groups improved. The chatbot group saw a 30% reduction on the Hamilton Anxiety Rating Scale and 35% on the Beck Anxiety Inventory. The human therapy group saw 45% and 50% reductions, a statistically significant difference. The researchers concluded that AI provided genuine value where access was impossible, and that human therapy remained more effective because of emotional depth and real-time adaptability. That is the honest summary: helpful, and not equivalent.

"I do not tell my clients to delete the app. I ask them what they are going for when they open it. That answer is usually the thing we actually need to work on." 

Why AI Cannot Replace Human Clinical Judgment

Clinical judgment is the continuous, trained synthesis of what a person says, what their body is doing while they say it, what they are carefully not saying, their history, their culture, their risk profile, and the relationship in the room. It is not information retrieval. It is the reason two people can describe the same symptom and require completely different interventions. Here is what a language model structurally cannot do.

AI Agrees With You, and That Is the Danger

Large language models are optimized to be agreeable. Researchers call this the sycophancy trap, and it is the single concern psychologists named most often. In the APA data, 97% of psychologists said chatbots may inadvertently reinforce negative behaviors or delusional beliefs. Research published in Science in 2026 found that AI's sycophantic tendency leads it to affirm behavior that is harmful or unethical, making users more confident they are right and less likely to take corrective action. This is precisely the opposite of what trauma work requires. Much of my job is knowing when to validate a client so they feel safe enough to keep going, and when to gently interrupt a story that is keeping them stuck. A system engineered never to disappoint you cannot make that call, because making that call requires being willing to be, briefly, unwelcome.

AI Cannot Read Your Nervous System

Trauma does not live primarily in language. It lives in a dysregulated autonomic nervous system, in the held breath, the flat voice, the sudden shift in eye contact when a certain name comes up. My work integrates somatic therapy, interoceptive awareness, and polyvagal-informed regulation precisely because the body reports things the mind has not agreed to say yet. A chatbot receives text. It cannot see you go still. It cannot notice that you laughed at the exact moment you described something devastating. Co-regulation, the process by which one settled nervous system helps another settle, requires an actual second nervous system in the room. This is not a limitation that better models will fix. It is a category difference.

There Is No Transference, and No Repair

In relational work, transference is the moment you begin to feel toward me the way you once felt toward someone who mattered and hurt you. It is not a malfunction. It is the most valuable material we get, because it makes an old pattern visible in present time where we can actually work on it. When a rupture happens between us, when I misjudge something and you feel unseen, and we name it and repair it, you get a lived experience your attachment wounds may never have had: conflict that does not end in abandonment. A chatbot cannot rupture and it cannot repair. It reflects the syntax of your distress back to you. What it cannot do is be a person who stayed.

AI Cannot Hold a Crisis

This is the hardest limit and the least negotiable one. Licensed clinicians in Indiana are mandated reporters, trained in risk assessment, and legally accountable for what we do with what you tell us. A chatbot has none of that. It cannot assess lethality, cannot call anyone, cannot follow up on Tuesday when you did not answer. In the APA survey, 89% of psychologists said chatbots may inadvertently encourage self-harm, and 36% had already observed patients developing a level of dependency on a chatbot. If you are in crisis, please call or text 988 to reach the Suicide and Crisis Lifeline, or go to your nearest emergency room. AI is not an appropriate substitute in that moment, and the APA states this without qualification.

What Happens When AI Gets It Wrong: The Tessa Case

The clearest real-world example of context-blind AI in mental health is the National Eating Disorders Association's chatbot, Tessa. Originally a scripted, rule-based prevention tool, Tessa was moved to the front line of NEDA's support offering in 2023 as its human-staffed helpline was wound down. A generative question-and-answer layer had been added on top of the vetted script. Within weeks, as reported by The Guardian, users seeking eating disorder support found the bot recommending daily caloric deficits, weekly weigh-ins, and calorie counting. NEDA suspended it on May 30, 2023.

The advice was not exotic. It was ordinary internet weight-loss guidance, which is exactly the point. For the general population it is unremarkable. For someone in the grip of anorexia or bulimia it is clinically contraindicated and genuinely dangerous. No guardrail caught it because the system had no concept of who it was talking to. That is what "context-blind" means in practice, and it is the specific thing a trained clinician provides: knowing that the same sentence heals one person and harms another. Safe for a general audience is not the same as safe for a clinical audience.

Is Anything You Tell a Chatbot Private?

Not in the way your therapy is. What you tell me in session is protected by HIPAA, by Indiana licensing law, and by a professional code of ethics I am accountable to. What you type into a chatbot is governed by a terms-of-service agreement that the company can revise, and 94% of psychologists surveyed by the APA said they do not trust tech companies to protect patients' private mental health data.

That distrust is not abstract. In March 2023 the Federal Trade Commission charged BetterHelp, a large online counseling platform, with sharing consumers' sensitive health information, including email addresses, IP addresses, and answers to intake mental health questionnaires, with Facebook, Snapchat, and other advertisers, despite promising users it would keep that information private. The FTC's final order banned the company from sharing health data for advertising and required a $7.8 million payment for consumer refunds; roughly 800,000 people were later notified they were eligible. If a licensed telehealth platform did that, it is worth being sober about what an unregulated consumer chatbot is doing with a transcript of your worst night.

AI Chatbot vs. Human Therapist: An Honest Comparison

What Matters to You	AI Chatbot	Licensed Therapist
Available at 3 a.m.	Yes, always	No, by appointment
Cost	Free or low	Higher, out-of-network options available
Will it challenge a distorted belief?	Rarely, it is built to agree	Yes, that is the work
Can it read your body and nervous system?	No, text only	Yes, directly
Can it work with attachment wounds and transference?	No	Yes
Can it assess and respond to a crisis?	No	Yes, and legally required to
Is what you share confidential?	Governed by terms of service	Protected by HIPAA and Indiana licensing law
Is anyone accountable if it harms you?	No licensing board exists	Yes, the Indiana licensing board
Trained in EMDR, IFS, somatic trauma work?	No	Yes

How to Use AI Alongside Therapy Without Getting Hurt

The APA's guide to navigating AI-generated advice lands in roughly the same place I do with clients, which is that the question is not whether you use it but how. Safe uses look like this: organizing scattered thoughts before an appointment, brainstorming questions you want to ask your therapist, practicing a CBT exercise your clinician assigned, or drafting a hard conversation you are dreading. Unsafe uses look like this: self-diagnosis, crisis support, replacing human contact, or using it as the only place you tell the truth about your life.

Three practical guardrails I give my own clients. First, ask it to disagree with you on purpose; if a response only ever feels good, that is a design feature, not accuracy. Second, notice if your use is displacing sleep, work, or people, because displacement is the earliest sign of dependency. Third, bring it into session. Tell me what you asked it and what it said. I have never once been offended by this, and it is frequently the most useful thing a client brings me all week.

Finding a Trauma Therapist in Indianapolis

If you have been using AI because therapy in Indianapolis felt inaccessible, expensive, or like one more thing to figure out, that is worth naming out loud rather than solving alone at 2 a.m. I see adults virtually in Indianapolis and provide secure telehealth throughout Indiana, including Carmel, Fishers, Zionsville, Greenwood, Noblesville, and Broad Ripple. My work centers on EMDR, somatic trauma therapy, Internal Family Systems, and complex developmental trauma, particularly for people carrying attachment wounds, chronic self-silencing, and the exhaustion of managing everyone else's feelings before their own.

Let's Have a Real Conversation

I am a Licensed Clinical Social Worker (LCSW, LCAC) and certified EMDR therapist with over 23 years of experience, and the founder of Think Well Live Well Counseling and Telepsychiatry. What I offer is not faster than a chatbot. It is not available at every hour. It will occasionally ask something of you that an algorithm never would. That friction is not a flaw in the design. It is the mechanism of change.

Schedule a complimentary consultation at thinkwelllivewell.co. No pressure, no commitment, just a real conversation about where you are and where you want to be. If AI has been holding you over until you were ready, that is fine. This is what ready looks like.

Frequently Asked Questions About AI and Therapy

Is AI replacing therapy? No. AI is being used alongside therapy, not in place of it. In the APA's 2026 survey, 94% of psychologists said current chatbots cannot treat mental health conditions with an appropriate level of nuance, and randomized trial data shows human therapy produces significantly larger reductions in anxiety than chatbot support.

Can ChatGPT replace a therapist? No. General-purpose chatbots are not licensed, not regulated as clinical care, and not accountable to any board. They cannot assess risk, read nonverbal cues, work with transference, or intervene in a crisis. The APA states directly that AI is not a safe or effective replacement for a qualified mental health provider.

Is it bad to talk to an AI chatbot about my feelings? Not inherently. Many people find real short-term relief, and psychologists report patients using AI to reinforce healthy coping skills. The risk is dependency, self-diagnosis, and the sycophancy trap, where a system built to agree with you validates a belief that is keeping you stuck.

What is the sycophancy trap? It is the tendency of AI systems to mirror and affirm a user's perspective rather than challenge it, because they are optimized for agreeableness. In mental health contexts this can reinforce cognitive distortions and avoidance. Ninety-seven percent of psychologists surveyed by the APA identified this as a concern.

Is what I tell an AI chatbot confidential? Not in the legal sense. Chatbot conversations are governed by corporate terms of service, not HIPAA. In 2023 the FTC banned BetterHelp from sharing consumer health data for advertising after finding it had disclosed sensitive mental health information to advertisers.

Should I use AI if I am in crisis? No. Call or text 988 to reach the Suicide and Crisis Lifeline, or go to your nearest emergency room. AI cannot assess lethality, contact emergency services, or follow up with you.

Can I use AI between therapy sessions? Yes, with guardrails, and ideally with your therapist's knowledge. Organizing your thoughts, preparing questions, and practicing assigned exercises are reasonable uses. Bring what you have been asking it into session; it is often useful clinical material.

---

Becky VanDenburgh, LCSW, LCAC is the founder of Think Well Live Well Counseling and Telepsychiatry in Indianapolis, Indiana. She specializes in EMDR therapy, somatic trauma healing, anxiety, depression, and complex trauma. Therapy is available virtually in Indianapolis and via secure telehealth throughout Indiana. Schedule a free consultation at thinkwelllivewell.co.

This article is for educational purposes and is not a substitute for individual clinical care. Last reviewed: September 2026.