Talking to AI Instead of People? When Chatbots Become Anxiety Avoidance

It’s Thursday night. A friend texted two days ago asking if you want to grab dinner this weekend. You haven’t answered. Instead, you’ve opened a chatbot and asked it how to reply, then rewritten its suggestion, then asked it whether the new version sounds too eager. Somewhere around draft five, you close the app, still feeling a little better, and the text stays unanswered.

Most of the public conversation about AI and mental health asks whether a chatbot can replace a therapist. It’s a fair question, and we’ve answered it directly in Is ChatGPT a Good Therapist? But for people living with anxiety, there’s a sharper question underneath it: is AI quietly replacing the human moments that would make you less anxious?

At an anxiety clinic, we don’t think of a chatbot as merely a weaker substitute for connection. For a socially anxious mind, it can be something more specific and more sticky: an almost perfect avoidance tool. Understanding why — and learning to use AI as a step toward people rather than a place to hide from them — is what this article is about.

🛟 Why a Chatbot Is the Perfect Safety Behavior

In CBT, a safety behavior is anything you do to get through a feared situation that also prevents you from learning the situation was survivable. Staying quiet in a group. Over-rehearsing what you’ll say. Leaving early. Having a drink first. They work in the moment, which is exactly why they keep anxiety alive. We cover the full list on our social anxiety therapy page.

Now look at what a chatbot offers, feature by feature:

• No face. Nobody is watching you blush, stumble, or go blank.
• No pause. There’s never an awkward silence you have to fill.
• No judgment you can see. It doesn’t raise an eyebrow, check its phone, or seem bored.
• Infinite edits. You can delete, rephrase, and regenerate until it feels safe.
• No consequences. It won’t be hurt, disappointed, or remember that you said something odd.
• Always available. It’s there at 11 p.m. when texting a friend feels like an imposition.

Every one of those features removes something social anxiety fears. That’s what makes the conversation feel so easy. It’s also what makes it so limited: the things a chatbot strips out — being seen, tolerating a pause, risking a reaction — are the exact experiences an anxious nervous system needs in order to recalibrate.

🔍 What the Research Is Starting to Show

The evidence here is early, and it’s worth reading carefully rather than dramatically.

In 2025, researchers at the MIT Media Lab, working with OpenAI, ran a four-week randomized controlled study with nearly 1,000 participants who used a chatbot every day. Across chat styles and conversation types, people who used it more each day tended to report more loneliness, more emotional dependence, and less socializing with real people. Participants who already had stronger attachment tendencies, or who trusted the chatbot more, were more likely to end the study lonelier or more dependent.

Two cautions matter. First, this is correlation. It may be that heavy use pulls people away from others — or that people who are already isolated reach for the chatbot more. Probably some of both. Second, it’s one study of one product over one month. It isn’t a verdict.

But from a clinical standpoint, the pattern is familiar. It looks a lot like what happens with any avoidance strategy: the people most drawn to it are often the people for whom it carries the most cost.

The American Psychological Association reached a related conclusion in its November 2025 health advisory on generative AI chatbots and wellness apps. It advised against using these tools as a substitute for care from a qualified professional, and it encouraged clinicians to ask clients directly about their chatbot use. That second recommendation is one we take seriously — and it’s part of why we’re writing this.

🔁 The AI Avoidance Loop

When chatbot use starts serving anxiety, it tends to follow a predictable sequence:

Notice what’s missing from this loop: any evidence. The anxious prediction (“they’ll think I’m weird”) is never checked against what a real person actually does. That’s the gap exposure-based CBT is built to close, and it’s the gap a chatbot, however supportive, tends to widen.

This is a different loop from the one we described in our ChatGPT article, where people ask an AI for reassurance about health, relationships, or intrusive thoughts. Reassurance-seeking is about calming a fear. The avoidance loop is about dodging the situation the fear is attached to. Many people do both.

🧩 Five Ways AI Quietly Becomes Avoidance

None of these is a problem in itself. Each one becomes a problem when it starts happening instead of the human interaction rather than before it.

1. Rehearsal without delivery

You use AI to draft a text, an email to a coworker, or what you’ll say to your landlord. Drafting once is planning. Drafting seven times and never sending is over-preparation — one of the classic social anxiety safety behaviors, now with an infinitely patient assistant. The tell: the drafting session ends with relief, but the message is still sitting there.

2. Telling the chatbot instead of the person

Something your partner said hurt. You explain the whole situation to a chatbot, get a thoughtful response, and feel calmer. The conversation with your partner never happens. Over time, the chatbot knows more about what’s bothering you than the person it’s about. This pattern shows up often in relationship anxiety, where the fear of raising a concern can be stronger than the concern itself.

3. Company instead of contact

The evening chat with the AI is pleasant and low-effort. Texting a friend means wondering if you’re bothering them, waiting for a reply, and maybe not getting one. Gradually, the chatbot fills the time that used to go — or could go — to people. This is the pattern the MIT findings point toward, and it tends to be invisible from the inside because each individual evening feels fine.

4. Outsourcing the mind-reading

“What did she mean by ‘ok.’?” “Was my comment in the meeting weird?” “Is he mad at me?” Social anxiety is full of mind-reading — guessing at what others think, usually negatively. Asking a chatbot to interpret other people’s behavior feels like getting a second opinion. In practice, it keeps you in the guessing business, and the only person who actually knows what she meant is the one you’re not asking.

5. Pre-approval before every interaction

“Is this okay to send?” “Does this sound rude?” “Should I even go?” When nothing social happens without the chatbot signing off first, you’re borrowing confidence rather than building it. Every interaction you complete without that check is evidence you can do it alone; every one routed through the AI is a missed chance to collect that evidence.

If this pattern sounds familiar but shows up more at your desk than in your social life — leaning on AI so you never have to sit with an imperfect draft or a hard call at work — our article on AI, imposter syndrome, and work anxiety covers that version in depth.

⚖️ Practice Tool or Hiding Place?

The same behavior can be healthy or anxiety-driven depending on where it leads. Here’s a quick way to tell them apart.

If you want a single question to carry around, try this one: Does this conversation end with me moving toward a person, or away from one?

🪜 Turning AI Into a Stepping Stone, Not a Substitute

You don’t have to delete anything. In exposure therapy, we often build a ladder of situations from least to most anxiety-provoking and work upward, dropping safety behaviors along the way. AI can sit on the bottom rungs of that ladder — as long as you keep climbing.

Here’s an example of how someone might use a chatbot as scaffolding and then deliberately fade it out:

Two rules make this work. First, each rung removes a bit of AI, not adds more. Second, track what actually happens versus what your anxiety predicted. Most people find the gap between the two is the most convincing thing they’ve seen in years. That gap — not the chatbot’s reassurance — is what reduces anxiety over time.

This kind of ladder is easier with someone helping you set the pace, choose the rungs, and notice when an old safety behavior sneaks back in under a new name.

🧠 What a Therapist Offers That a Chatbot Can’t — For Anxiety Specifically

The usual answer to “why not just use AI?” is that a therapist offers a real relationship. That’s true. For anxiety, though, there’s a more practical point: therapy itself is an exposure.

• You’re seen by a person. Talking to a human who can see your face, hear your voice catch, and sit through a silence with you is, for many socially anxious clients, the first rung of the ladder.
• Safety behaviors show up in the room. A therapist can notice when you read from prepared notes, avoid eye contact, or apologize before every sentence — and help you drop those habits with someone right there.
• Exposures get assigned and followed up. A chatbot can suggest you text a friend. A therapist asks next week whether you did, what happened, and what you predicted beforehand.
• Someone will disagree with you. A good CBT therapist won’t simply validate the story that everyone at the party thought you were awkward. They’ll ask for the evidence.
• Progress is tracked over months, not messages. Your therapist remembers where you started and can show you how far you’ve come.

For some people, a video session feels more manageable than walking into an office at first. That’s a reasonable starting point. Research suggests telehealth can be as effective as in-person care for many anxiety concerns, and virtual therapy can be its own early rung — with in-person sessions as a later one if that’s part of your goals.

🚩 Signs Your AI Use Is Doing Anxiety’s Job

This isn’t a diagnostic checklist, and one or two items don’t mean much on their own. But if several of these are true and getting more true, it’s worth paying attention:

• You regularly draft messages with AI that you then don’t send
• You’ve stopped initiating plans, calls, or texts you used to make
• The chatbot knows about conflicts or worries that the people involved don’t
• You ask AI to interpret other people’s messages or behavior before you respond
• You feel unable to send ordinary messages without checking them first
• Social events feel harder than they did six months ago, not easier
• You’d feel uncomfortable if a friend saw how often you use it
• You feel less lonely during the chat and more lonely after it

If you notice these, there’s no need for shame. Avoidance is how anxious brains protect themselves, and a tool built to be agreeable and always available is going to be very good at helping with that. Noticing the pattern is the first step out of it.

And if you’re ever in crisis — thinking about harming yourself, or feeling unsafe — please don’t rely on a chatbot. Call or text 988 to reach the Suicide & Crisis Lifeline, or call 911 in an emergency.

🌱 The Goal Isn’t to Quit AI

We’re not asking anyone to give up a useful tool. AI can help you organize your thoughts, plan a hard conversation, or find words when you’re stuck. Used that way, it can be a decent bottom rung on the ladder.

The goal is for the ladder to lead somewhere: back to the friend who texted, the partner you haven’t told, the dinner you keep not making plans for. Anxiety gets smaller when you collect evidence that people are safer than it predicts — and the only place to collect that evidence is with people.

Calm Anxiety CBT Therapy Clinic helps adults work through social anxiety, relationship anxiety, and avoidance patterns using CBT and exposure-based therapy. We see clients in person in Lakeview and Andersonville, and by telehealth throughout Illinois.

❓ Frequently Asked Questions

Is it bad to talk to AI when I have social anxiety?

Not on its own. Using AI to plan a conversation or organize your thoughts can be helpful. It becomes a problem when it replaces interactions with people rather than preparing you for them. The question to ask is whether your AI use is moving you toward people or away from them.

Can AI chatbots make loneliness worse?

Early research suggests a link. A 2025 MIT Media Lab and OpenAI study found that heavier daily chatbot use was associated with more loneliness and emotional dependence and less socializing with real people. The findings are correlational, so it isn’t clear whether the chatbot causes loneliness or lonely people use it more, but the pattern is worth noticing.

What is a safety behavior in social anxiety?

A safety behavior is something you do to get through a feared situation that also prevents you from learning the situation was manageable, such as over-rehearsing, staying quiet, or leaving early. Using a chatbot to script, approve, or replace social interactions can work the same way.

Can I practice social skills with a chatbot?

You can use a chatbot to rehearse, but rehearsal only helps if it leads to the real interaction. The parts of social situations that anxiety fears most, like pauses, facial reactions, and unpredictable responses, are exactly what a chatbot can’t provide. Treat AI practice as an early rung, and then move on to real conversations.

How does exposure therapy help if I’ve been avoiding people?

Exposure therapy builds a gradual ladder of social situations, starting with easier ones and working upward while dropping safety behaviors. Each step gives your brain real evidence that interactions go better than anxiety predicts, which reduces fear over time in a way that avoidance and reassurance can’t.

Should I tell my therapist I use ChatGPT or other AI tools?

Yes. The American Psychological Association has encouraged clinicians to ask clients about chatbot use. Knowing how you use AI helps your therapist spot avoidance or reassurance patterns and, when it’s helpful, build AI use into your treatment plan as a stepping stone.

Is virtual therapy a good first step for social anxiety?

For many people it is. Starting with video sessions can feel more manageable, and research suggests telehealth can be as effective as in-person care for many anxiety concerns. Some clients later add in-person sessions as part of their exposure goals.

Is AI therapy legal in Illinois?

Illinois law prohibits AI systems from independently providing therapy or making therapeutic decisions. Licensed professionals may use AI for administrative and supplementary support. Apps that present themselves as an “AI therapist” to Illinois residents deserve caution.