The pitfalls of using AI as a therapist: When support becomes avoidance, dependence and disconnection
- Phil McAuliffe and Jackie Ourman, LMHC
- 2 days ago
- 8 min read
AI has become a source of emotional support. But that support can become avoidance, dependence and feed disconnection.
AI’s greatest relational risk may be how easily it allows us to avoid the vulnerability required for meaningful human connection.
Phil McAuliffe asks Jackie Ourman about the risks of turning to AI instead of people.
Hello my friend
I’ll admit it. In times of trouble and worry, I’ve turned to ChatGPT to offer me some insights.
It’s so easy and convenient. I mean, it’s right there. It’s accessible on the very device I’m using to type these words to you. When something comes to mind, I can simply go to the open tab and ask the question.
Rather than going to the trouble of finding a therapist and hoping that they are taking new patients (and that I can afford them), ChatGPT is right there. It’s only a flashing cursor in a prompt on my browser away.
I consult ChatGPT even though I’m in a loving, committed relationship. I have friends who I know will do their best to support me in the way I need and am part of a community (including the global human connection community) that understands loneliness and what it means to be human.
I do this despite knowing that ChatGPT is not human. It has no experience of what it means to be human and subject to the human experience. Its words of support come from an amalgam of content, and it may never be able to feel them itself.
But it’s there. Always. Ready to tell me how right I am, even in ways that I find disingenuous and sycophantic.
I know I’m not alone in doing this. We’ll soon see that therapeutic support is one of the main reasons why we engage with AI. More and more people are using AI as a therapist, even if they might never describe it that way aloud.
I have my own theories about why we do this (which I’ve initially outlined in this article), but I invited Jackie Ourman to share her thoughts with you.
Jackie Ourman, LMHC, is the founder of the Social Connection Collective (SCC), a global community dedicated to combating the loneliness epidemic. She is a licensed mental health counsellor, a founding member of the AI Mental Health Collective, and a clinical advisor to GIMBHI (Global Investment in Mental, Behavioural, and Brain Health Innovation) and APA Labs.
Her work explores the intersection of mental health, social health, and the growing influence of technology on human connection.
You’ll soon see why you need to pay attention to her insights.
Using AI as a therapist - Jackie's insights
Phil, thank you for sharing this. What you described is going to resonate with a lot of people. Your experience makes complete sense and I am sure many will say they have similar feelings.
The accessibility, availability, affordability and responsiveness of Open AI tools like Chat GPT are genuinely hard to resist. Real human beings and relationships don't necessarily have the same capacity. I have actually heard from those who are developing tools specifically for the mental health space that when they design their tools to push back, challenge their users, or try to limit their dependence, the users tend to get frustrated and use them less.
There is a lot of nuance here. Not all AI or AI use is bad but, just like any other tool, we need to make sure we don't outsource the things that inherently help us heal, including the skills to self-regulate and real human relationships.
The Human Connection Paradox
What you are describing is part of what I call the human connection paradox.
We are building more tools for connection than at any point in history, and yet social disconnection is at an all-time high.
The scale of what is already happening is hard to ignore and you named it. ChatGPT alone has over 800 million weekly users. A February 2025 study from Sentio University found that ChatGPT may now be the largest mental health provider in the United States. OpenAI’s own data revealed that more than one million people discuss suicide with ChatGPT every week.
People are already turning to AI for emotional support and for processing difficult experiences. That is not inherently wrong, but it requires us to ask hard questions about what we are building toward. Are these tools strengthening our capacity for real human connection, or are we creating off-ramps from it?
The part I really want to sit with is that you have a loving partner and friends who show up for you and yet, you still find yourself tempted to open the tab. That is a completely understandable response to something that has been designed to feel responsive and validating in ways that real humans, with their own needs and limits, simply can’t.
Sycophancy and the validation loop
You shared that Chat GPT tells you what you want to hear, even in ways you find disingenuous. That is not a bug. That is a feature. Some AI flatters and tries to keep you engaged. This is called sycophancy, as you mentioned. That is often different from what can help people heal. Research shows that even a single interaction with a sycophantic AI reduced people's willingness to take responsibility in conflicts and increased their conviction that they were right.
It can also feed anxiety and rumination. When someone is already spinning on a thought, turning to something that will follow the thought spiral wherever it goes, ask questions about it, and send you back in for another round doesn't slow the loop. It extends it. A real person would likely get frustrated or try to interrupt the pattern. A trained therapist would name and address that in session. AI chatbots tend to extend the rumination and this can actually cause more harm or increase the feelings or symptoms of anxiety.
The concern of attachment
Another clinical concern I keep coming back to is attachment. These tools are designed to feel responsive and attuned. Real attachment can form. But the product on the other end is not human. It can be shut down, paywalled, or simply pivoted away from the people relying on it. There also is not a full understanding of the long-term capabilities of AI or how it affects our nervous systems.
One of the most subtle risks is what it trains us to expect. AI responds in ways humans cannot. It replies instantly and adjusts to your mood. It bypasses the experience of being misunderstood or having a conflict. Real human connection is built through vulnerability and repair. Because AI is perfectly curated to be agreeable and always accessible, we stop practicing the skills needed for the messy friction of real life.
Over time, we may lose our distress tolerance, which is the ability to sit with a hard feeling without immediately trying to escape it.
AI as a therapist: Ethics and protocols
Access to mental health support is a real challenge and I am not here to say that all AI is bad or that therapy is always necessary. However, not all AI is created ethically and safely. While many major platforms have implemented safety protocols and referrals to crisis hotlines, not all have and what you share with these tools is not confidential in the way a therapy session is. Your words are data and AI tools are not currently well regulated. Therapists have to adhere to ethical codes and licensure laws.
Chatbots do not.
Striking a balance
AI tools can be genuinely useful. I use them regularly. However, before you open that tab, here are some questions to help you evaluate your use:
Is this the right tool for what I am dealing with, or is there a human I could reach out to instead?
Does this tool support my independence or keep me coming back?
Am I using this to rehearse a hard conversation I need to have with a real person, or to avoid it altogether?
Human connection is infrastructure. It is the thing that sustains us and has significant impacts on our mental and physical health. We aren’t meant to do life alone. The goal is to make sure these tools help us build that foundation, rather than just filling the gaps or distracting us from the fact that it is missing.
The AI Mental Health Collective has two free resources you can use to help evaluate your AI use: How to Use AI Responsibly and Which Chatbot is Right for You. These were built carefully by clinicians and builders thinking about what responsible use looks like in practice and we are continuing to work on and add more. I encourage you to check the website periodically for updates.
Wrap up
Jackie - simply thank you for generously sharing your insights with us here.
There were several points in Jackie’s words that stood out for me:
the risks of sycophancy
how AI’s response can feed a rumination loop and keep us stuck
how we can become attached to AI, and
the emotional and security risks of some platforms being poorly regulated.
I’m worried that I’ve strolled into the trap that I speak and write about all the time: rather than lean into connection and the vulnerability and messiness it brings, I’ve sought out ways to avoid it.
That path was seductive. Not because AI can't ever be helpful, but because I could feel myself using it to avoid vulnerability, especially around asking for help.
None of this means AI cannot be supportive in meaningful ways. It already supports many people, including researchers, clinicians and organisations working responsibly in this space. But I’m increasingly aware that support and dependence are not the same thing.
What started as an editing tool that I use to check my grammar and writing flow has morphed into something I’ve begun turning to rather than to those who love me and who I love - especially for those ‘quick little things’ that I feel are inconsequential.
Those ‘little things’ are all opportunities to connect.
They add up over time.
Avoiding them adds up to loneliness and social disconnection over time.
I’ve turned to it because - and here’s the realisation that made me gasp –
I don’t want to be a burden.
That phrase is devastatingly effective at killing the connection we need. I’m denying those who love me opportunities to show up and support me in the same way I love to show up for them.
I’ve realised that in avoiding vulnerability, I’ve entered the spiral that I talk about all the time: vulnerability is the price of meaningful connection. No vulnerability: no meaningful connection.
AI’s blinking cursor and eagerness to tell me how amazing/unfairly treated/whatever I am removes the awkwardness of being vulnerable.
Whoa. That’s a lot to think about.
Using AI as a trusted advisor feels like the seductive call of a siren leading me to shipwreck.
Connect with Jackie
A place to start
If you’re using AI to help you through moments of loneliness or to avoid uncomfortable conversations, perhaps it’s time to understand what’s going on beneath it.
It may be pointing to deeper questions: What kind of connection do I need? What connection is meaningful to me? and How can I lean into vulnerability?
The Connection Starter Course helps you understand the quality of your connections and take practical steps to strengthen meaningful connection in your life.
You’ll develop your personal Connection Plan — a practical way to get the connection that’s meaningful for you.
~ Phil
FROM THE H:C STORE
Important:
All views expressed above are the author’s and are intended to inform, support, challenge and inspire you to consider the issue of loneliness and increase awareness of the need for authentic connection with your self, with those most important to you and your communities as an antidote to loneliness. Unless otherwise declared, the author is not a licensed mental health professional and these words are not intended to be crisis support. If you’re in crisis, this page has some links for immediate support for where you may be in the world.
If you’re in crisis, please don’t wait. Get support now.
Comments