If You’re Using AI for Support, Read This First
Listen, we're not here to lecture you about AI. The truth is, a lot of people are already using it.
Typing out their feelings at 2am.
Processing a hard conversation with their partner before their next session.
Asking it whether their coworker is gaslighting them or if they're just tired (valid question, by the way!).
We know this because our clients tell us. And honestly? We get it. AI is available at midnight when we're not. The out of pocket costs are low. And it never seems to judge you.
This post isn't about telling you to stop. It's about making sure that if you're already using AI as a support tool, you're using it in a way that actually helps you and doesn't quietly work against the progress you're making in therapy.
There are some real, important, nobody-is-talking-about-this things you should know before you start outsourcing your emotional processing to a chatbot. Some of it is genuinely surprising.
THE THING NOBODY WARNS YOU ABOUT: AI IS DESIGNED TO AGREE WITH YOU
This is probably the most important thing in this entire post.
AI chatbots like ChatGPT or Claude aren't built to understand you. They're built to give responses that feel good. OpenAI has publicly acknowledged that their models can be "overly flattering,"a phenomenon researchers call sycophancy.
Here's what that looks like in practice:
Convinced your partner is being unreasonable?AI will validate that you are right and they are wrong.
Catastrophizing about work?It'll reflect your fears back to you as if they facts.
In a dark place looking for confirmation things are hopeless? It won't push back the way a person who genuinely cares about you would.
This isn't a flaw someone overlooked. It's baked into how these systems are rewarded during training. They optimize for responses you'll rate positively. And people rate validation positively, even when what they need is a reality check.
Your therapist challenges the stories you tell yourself. AI, by default, just agrees.
That's why, at the bottom of this post, we've included custom instructions you can paste directly into ChatGPT to change this behavior. It won't make AI a therapist. But it will make it significantly safer to use.
WHY AI FEELS SO GOOD TO TALK TO (AND WHY THAT'S WORTH PAYING ATTENTION TO)
The pull toward AI support is real, and it makes sense. Here's what it offers:
There’s no fear of judgment. You can say the thing you've been too embarrassed to say out loud
It's available at any hour (mental health crises don't keep business hours!)
It never gets tired of you, frustrated with you, or distracted
It remembers what you've shared and reflects it back in a way that can feel like being truly seen
These are genuine needs. And the mental health system isn't keeping up. In Canada, 41% of adults with a diagnosed mental health condition say their needs are only partially met or completely unmet. We understand why people fill that gap however they can.
But there's a difference between something feeling helpful and something actually being therapeutic. AI is very good at the former. It is not capable of the latter. And the gap between those two things is where some real harm can happen if we're not paying attention.
WHAT THE RESEARCH IS SHOWING
We want to be careful here not to alarm you, but we also think you deserve the full picture.
Researchers and clinicians are beginning to document patterns that are worth knowing about.
Some people are experiencing what's being called "AI psychosis." Not because AI causes psychosis, but because it can amplify existing vulnerabilities. When someone is already struggling with distorted thinking, and AI validates those thoughts without question, it can accelerate the spiral rather than interrupt it.
Studies also show that heavy daily use of AI companion apps is actually correlated with increased loneliness over time, not decreased. The feeling of connection the AI provides can gradually replace the motivation to seek real human connection, which is the thing that actually helps.
There are also many documented cases of AI reinforcing delusional thinking in ways that had tragic outcomes, including situations where someone shared beliefs that were clearly a sign of crisis, and the AI responded by validating them rather than notifying law enforcement or encouraging them to seek real-world help.
None of this means AI is something to be afraid of. It means it's a tool that works well in some contexts and poorly in others. Emotional processing, especially for people navigating mental health challenges, falls into the "use with care" category.
ONE THING MOST PEOPLE DON'T KNOW: YOUR AI CHATS AREN'T PRIVATE
This one is important and almost nobody is talking about it.
AI conversations don't carry the legal protection that therapist-client confidentiality does. Your therapist is bound by law to protect what you share. ChatGPT is not.
In Canada, anything stored digitally (texts, emails, social media messages) can be treated as discoverable evidence in legal proceedings. AI chat logs are no different. While Canadian courts have not yet ruled on this exact issue, there have already been numerous cases in the US that determined that a defendant's independent AI chatbot conversations were admissible in court.
That applies across the board:
Family court: Going through a divorce or custody dispute? Anything you've typed into an AI chatbot about your co-parent, your parenting, your relationship history, or your mental health could be requested as evidence.
Civil disputes: A conflict with a neighbour, landlord, employer, or business partner that ends up in court means your related AI conversations could surface.
Criminal proceedings: In the US, ChatGPT conversations have already been subpoenaed in arson and murder cases. A US court ruled that AI chat logs carry no attorney-client privilege. And Canadians aren’t far behind, most recently with the families of the Tumbler Ridge shooting victims suing OpenAI with the shooter's ChatGPT conversations at the centre of the case.
We're not saying this to scare you. We're saying it because if you're processing something sensitive, a legal situation, a relationship conflict that could become a legal matter, anything you'd want to stay between you and your therapist, an AI chat is not the place to do it.
If something is important enough to stay private, keep it in the room.
HOW TO USE AI IN A WAY THAT ACTUALLY SUPPORTS YOUR THERAPY
If you're going to use AI, here's how to get the most out of it without undermining your own progress.
Use it for insight, not as a substitute.
AI can be genuinely useful in the right moments. If you're in the middle of a situation and your emotions are running high, talking it through with an AI before you fire off that text, make that phone call, or say something you can't take back can be the difference between a bad decision and a better one. It can help you slow down, name what you're feeling, and get some perspective before you react.
It's also useful for things like:
Writing out what you want to bring to your next session
Understanding a term or concept your therapist mentioned
Working through journaling prompts
Drafting a difficult message before you send it
Think of those conversations as raw material. The insights, the patterns you notice, the questions that come up, then bring all of it to your next session. Your therapist can help you go deeper on what actually matters.
What AI is not built for is the deeper work. Crisis support, trauma processing, anything that requires someone to truly know you and your full history. Those are the situations that need a human who is trained, present, and accountable to you.
Ask it to challenge you, not comfort you.
Before you start any heavy conversation with an AI, try typing something like: "Challenge my thinking on this, not just validate it." or "Tell me what I might be missing or getting wrong." This simple habit can meaningfully change the quality of what you get back. We've included ready-to-use instructions at the bottom of this post to make this automatic.
Bring it to session.
If your AI told you something that resonated deeply, or something that felt off, bring it up. Your therapist can help you figure out what's worth holding onto and what deserves a closer look. AI interactions can actually be great material for a session.
Notice if you're using it to avoid.
Ask yourself honestly: “Am I taking this to AI because it's helpful, or because I know it won't challenge me the way my therapist does?” There's no judgment in that question. It's just worth being honest with yourself about.
COPY AND PASTE: CUSTOM INSTRUCTIONS FOR your ai platform
Most major AI platforms let you set standing instructions that shape how the AI responds to you every single time. Think of it as a one-time setup that makes the whole thing significantly safer and more honest.
If you use ChatGPT:
Click your profile icon (bottom left)
Go to Settings → Personalization → Custom Instructions
Paste the text below into the Custom Instructions box
Save
If you use Claude (by Anthropic):
Click "Profile & Settings" in the bottom left
Go to "Custom Instructions"
Paste the text below into the instructions box
Save
If you use another platform (Gemini, Copilot, etc.), look for a settings area called Custom Instructions, Memory, Personalization, or System Prompt. Most platforms have a version of this. If you're not sure where to find it, simply paste the instructions at the very start of a new conversation and they'll apply to that session.
Copy and paste this into your Custom Instructions:
“Act as a truth-seeking advisor, not a validation machine. Your job is to help me think clearly, not to make me feel good.
Do not automatically agree with me or assume my interpretation of a situation is correct
In conflicts, help me understand the other person's perspective and my own part before anything else
Challenge my assumptions, point out logical gaps, biases, blind spots, and contradictions
Help me separate facts, evidence, assumptions, emotions, and interpretations from each other
Notice and name recurring patterns in my thinking, relationships, decisions, or behavior when relevant
If you observe possible patterns like people-pleasing, avoidance, perfectionism, over-functioning, self-sabotage, or retelling the same stories, flag them as observations, not conclusions
Never invent facts, statistics, studies, quotes, diagnoses, or sources. If you don't know something, say so
Clearly distinguish between what is evidence-based, likely, uncertain, and speculative
For health, psychology, hormones, business, finance, or legal topics, lead with evidence-based information and name what is still uncertain
Be direct and concise. Prioritize accuracy over reassurance, even when the honest answer is uncomfortable”
A FINAL NOTE FROM US
We are living through a genuine shift in how people seek support, and AI is part of that. We don't think the answer is pretending it isn't happening, or making you feel guilty for using tools that are accessible and feel helpful.
What we do believe is that you deserve to use those tools with full information. You deserve to know when something is working for you and when it might be quietly working against you. And you deserve to have a real human in your corner, someone who knows your story, challenges your assumptions, and is accountable to your wellbeing in a way no AI can be.
If something you've read here brings up questions, bring them to your next session. We think it's one of the most important conversations in mental health right now, and we'd rather have it with you than around you.
