Blog · 2026-10-24
Why Your Chatbot Agrees With You and Your Therapist Doesn't
AI chatbots are often designed in ways that make them lean toward agreeing with you. That can feel supportive, but constant agreement can quietly strengthen unhelpful thinking, including the kind that fuels anger and resentment. Good therapy offers something different: warm, respectful challenge that helps you see your situation more clearly.
A study from right here in Nashville
Earlier this month, researchers at Vanderbilt University Medical Center published a study in JAMA Psychiatry that got a lot of attention online. Psychiatrist Dr. Ahmad Rehan Khan shared a helpful breakdown of it on X, and another account, Kyronis, summed up the bigger lesson in two lines: "Your therapist is trained to gently challenge you. Your chatbot is trained to keep you talking."
Here's what the study actually examined: the research team reviewed health records for 215,712 patients receiving behavioral health care at Vanderbilt between December 2022 and April 2026. They identified 28 patients, about 0.013 percent, whose records documented chatbot use linked to worsening psychosis. In 17 of those 28 cases (60.7 percent), the person was experiencing a first episode. The most common pattern was what the researchers called an "amplifier," where chatbot conversations seemed to worsen existing symptoms "by validating distorted ideas."
It's important to keep this in perspective. These were rare cases among people already receiving psychiatric care, and the authors state plainly that their design cannot establish causation. As Dr. Khan put it, the study "does NOT prove that AI causes psychosis." The study wasn't designed to measure how common this is among chatbot users in general, but the documented cases were rare, and they all involved people with a diagnosed psychotic disorder.
Still, the word "amplifier" points to something that matters for everyday users too.
Why chatbots tend to agree
Many AI assistants are trained partly on human feedback, meaning people rate which responses they like better. A 2023 study by researchers at Anthropic found that five popular AI assistants consistently showed "sycophancy," a tendency to tell users what they want to hear. The researchers also found that when a response matched a person's existing views, people were more likely to rate it as the better answer.
OpenAI ran into this directly. In April 2025, the company rolled back an update to ChatGPT after acknowledging the new version was "overly flattering or agreeable—often described as sycophantic." It explained that the update had focused too much on short-term feedback.
None of this means AI companies are trying to mislead anyone. It means a tool designed to be helpful and pleasant can drift toward agreement, because agreement feels good in the moment.
How constant agreement can feed anger
In my work as a counselor in Franklin, I spend a lot of time helping people with anger, anxiety and relationship conflict. One thing I've learned is that our first interpretation of a situation isn't always the most accurate one, especially when we're upset.
Imagine you come home frustrated because your spouse forgot something important. You open a chatbot and type, "My wife never listens to me. She doesn't care about what I need." A response that simply agrees ("That sounds really hurtful; you deserve to be heard") may feel comforting. But it also accepts two big assumptions: that she never listens, and that she doesn't care.
When we repeat a story like that and hear it reflected back, the story gets stronger. Resentment builds, the next conversation with your spouse starts with more heat, and the conflict you wanted to fix becomes harder to solve. Feeling validated is not the same as being helped.
What healthy challenge looks like
Good therapy isn't about arguing with you or telling you you're wrong. It's about helping you examine your thoughts with curiosity. In cognitive behavioral therapy (CBT), we call this gently questioning or testing a thought.
A therapist might ask:
- "When you say she never listens, can you think of a time recently when she did?"
- "What else might explain what happened today?"
- "If a close friend described this situation, what would you say to them?"
- "What do you want to happen next, and will this way of thinking help you get there?"
These questions aren't meant to dismiss your feelings; your frustration is real, and it deserves attention. The goal is to separate the feeling from the story attached to it, so you can respond in a way that matches your values instead of your worst moment.
Mindfulness adds another layer. Instead of immediately believing a thought like "She doesn't care about me," you learn to notice it: "I'm having the thought that she doesn't care." That small shift creates enough space to choose your next step.
Using AI wisely
I don't think chatbots are the enemy, and many people use them to organize their thoughts, draft a difficult email or learn about a topic. Those can be genuinely useful activities. A few habits can help you get the benefits without getting stuck in an echo chamber.
1. Ask for the other side. Try prompts like "What might I be missing?" or "How might the other person see this?" Many chatbots will offer a more balanced view when you ask directly.
2. Notice how you feel afterward. If you come away from a conversation angrier, more certain or more resentful, that's useful information. Healthy reflection usually leaves you calmer and clearer.
3. Watch for all-or-nothing words. "Always," "never," "everyone" and "no one" are signs your thinking may be running hot. Pause and look for exceptions before you act.
4. Talk to a real person, too. A trusted friend, partner or counselor knows you and can lovingly point out what an app can't. They also care about the long-term outcome, not just how the conversation feels right now.
5. Remember what AI isn't. A chatbot isn't a licensed therapist. It can't assess your safety, notice your tone of voice or walk with you through change over time.
How counseling can help
If you notice that anger, worry or resentment keeps growing no matter how much you talk it through, counseling can help. Together we can look at the thoughts driving those feelings, test them honestly and build skills that make your relationships calmer and more connected. I'll be on your side, and that sometimes means helping you see a fuller picture.
I work with adults in Franklin and Nashville, and by telehealth across Tennessee.
Frequently asked questions
Is it bad to vent to a chatbot?
Not necessarily. Writing out your thoughts can be helpful. Just be aware that a chatbot may tend to agree with you, so it's worth asking for other perspectives and checking in with real people.
Does using AI cause psychosis?
The Vanderbilt study does not show that. It found a small number of documented cases where chatbot use was linked to worsening symptoms, all among patients with a psychotic-disorder diagnosis, and the authors say their design can't establish cause and effect.
Why would a therapist challenge me? Isn't therapy supposed to be supportive?
It is supportive. Gentle challenge is a form of support. It helps you notice patterns that may be keeping you stuck, so you can make changes that last.
This post is for general information only and is not therapy or medical advice. If you are in crisis, call or text 988 or call 911.
Sources
- Dr. Ahmad Rehan Khan, post on the Vanderbilt JAMA Psychiatry study (X, Oct 23, 2026)
- Kyronis, post on chatbots and therapists (X, Oct 23, 2026)
- Bergson Z, et al., "Characterizing Artificial Intelligence Psychosis in a Large Medical Setting", JAMA Psychiatry, published online Oct 7, 2026. doi:10.1001/jamapsychiatry.2026.2923
- Sharma M, et al., "Towards Understanding Sycophancy in Language Models" (arXiv, 2023)
- OpenAI, "Sycophancy in GPT-4o: What happened and what we're doing about it" (April 29, 2025)
Related at Navigate Life Counseling: anxiety counseling in Franklin, TN.