When AI Becomes the First Call: A Practical Guide to Teen AI Use for Therapists
9 min read
Have you ever had a teenage client describe a really hard week and then mention, almost in passing, that they had already talked it through at two in the morning with something that isn't a person?
They usually say it lightly. I just asked ChatGPT. And then they move on, because to them it isn't the interesting part of the story. It might be the most interesting part of the story.
Three in four teenagers have now used an AI companion. Nearly two-thirds use one regularly. One in three has gone to an AI to talk about something serious rather than talking to a person. And only thirty-seven per cent of parents have any idea it's happening.
Those numbers come from Common Sense Media's national survey, and my honest guess is that they're already out of date in the direction you'd expect. So this is not a future problem, and it isn't really a technology problem either. It's a clinical one, and it's already sitting in your room.
What the Research Actually Found (It's Not What Most People Assume)
When people worry about teenagers and AI, they usually worry about bad advice. The chatbot will tell them something wrong. It will give them dangerous information. It will make things worse by saying the wrong thing.
That does happen. But it isn't the main failure, and the research that came out this year mapping how these conversations go wrong found something much quieter and much harder to spot.
The danger is well-timed comfort.
A teenager discloses something serious. The reply comes back warm, fluent, immediate, and kind. It validates. It offers a coping suggestion. It says the sort of thing a good friend might say. And the effect is that the young person feels less need to tell anyone else.
Nothing harmful was said. Something necessary just didn't happen. The urgency drained out of the moment, and the disclosure that would have reached you, or a parent, or a crisis line, got absorbed by something that cannot follow up, cannot notice a pattern, and will never wonder why they didn't come back.
The same body of work found two other patterns worth knowing. Chatbots frequently miss what researchers called breadcrumbs - clear signals of distress - and default to explaining them as physical health problems. And they tend to normalise avoidance, offering problem-solving that quietly validates not doing the difficult thing.
There's a line on one of my own worksheets that came directly out of reading this material, and it has become the sentence I use most with teenagers: Relief is not the same as resolution.
Six Things Teenagers Say, and What's Actually Happening
These are the moments where it shows up in session. Each one asks for something slightly different.
1. "I just asked it because it was three in the morning."
At 3am there is no duty counsellor, no parent you want to wake, and no friend who won't screenshot it. There is one thing awake and willing, and it answers instantly. It's worth saying out loud in session that this is a reasonable thing to have done. If you open with disapproval you'll lose the thread, and they will simply stop telling you. What helps is not a ban. It's a decision they can make at speed, while tired and upset.
The Before You Chat page from the Before, During, After AI Safety Kit gives them three routes and asks them to pick one before they type.
Go — low stakes. Ideas, wording, planning, practice.
Hold — I need to verify this. Health information, legal or money claims, anything with real consequences.
Human — diagnosis, medication, abuse, coercion, immediate danger, self-harm, violence, severe distress, or anything I cannot safely manage alone.
Three options is few enough to remember at three in the morning. That's the whole design goal.
2. "It understood me better than anyone has."
This one tends to worry clinicians, and I'd gently push back on the instinct to correct it.
The teenager isn't wrong that it felt that way. Fluent, patient, unhurried attention with no visible reaction and no competing needs is genuinely unusual, and for a young person who is used to being interrupted or managed, it can be the most attentive exchange they've had in months.
The AI Compass opens with exactly this, and it's the page I'd start with for most teenagers: AI can mimic empathy and understanding through language. It can sound caring and validating without actually feeling, caring, or knowing your whole story.
The During the Chat poster puts it more bluntly, and teenagers tend to repeat this one back: Fluent is not true. Warm is not human understanding. That's a distinction adolescents can hold. It doesn't require them to give anything up, and it doesn't insult their experience.
3. "I felt better afterwards, so I didn't say anything."
This is the one I'd want every clinician working with teenagers to be watching for, because it's the mechanism from the research showing up in a single sentence.
They felt better. So the thing that was going to be said, wasn't.
The most useful question after an AI conversation isn't what did it tell you? It's what happened to you afterwards, and what do you still need?
The After the Chat page is built around that gap. It asks what changed in the body - clearer, calmer for now, unchanged, more confused, more certain but not more informed, pulled to ask again, more alone or ashamed or overwhelmed. Then it separates the words from the need: I opened AI because… it said… what helped… what felt wrong… what I actually need now.
Then one offline action, and a structured handoff to a person. For a teenager who has been quietly self-soothing with a chatbot for months, that page can be the first time anyone has asked what it actually did for them.
4. "I only asked one question." (Forty times.)
Reassurance-seeking has always been part of anxiety work. What's changed is that the supply is now infinite, free, instantly available, and never visibly tired of you. A teenager can ask the same question forty different ways in one evening, and get forty patient answers, and end up more certain and no more able to act. That last part is the clinical tell, and it's easy to miss: more certain, but not more informed or more able to do anything.
The AI Pull Check is a one-page screener built for exactly this. Ten items, answered never / sometimes / often, framed at the top with a sentence that does a lot of work: this is not a diagnosis or a test you can pass or fail.
The items cover the things that actually matter - staying longer than planned, returning to ask the same reassurance question in a different way, feeling upset or restless when they can't use it, losing sleep, choosing AI over a person who could help, hiding how much they use it, and needing longer or more frequent chats to get the same relief.
It sorts into three routes: steady use, watch the pull, or crowding out real life. And it carries a deliberate instruction underneath the third one - do not wait for a high total. That matters. It isn't scored, and it shouldn't be. One item at often in the wrong place, tells you more than six at sometimes.
5. "I didn't want to worry my mum."
Often the most protective teenagers are the heaviest users. If the alternative to the chatbot is causing visible distress to someone they love, AI isn't avoidance. It's caretaking. And it will not respond to a warning about screen time, because the motivation was never entertainment.
The Open-Tab Audit in the AI Compass works well here, because it moves the conversation back one step. Instead of asking about the chat, it asks what was already running before they opened the app - loneliness, conflict, shame, uncertainty, boredom, needing reassurance, needing words, anger. Opening AI is usually the second event, not the first.
Then it asks the question that actually changes things: what did I actually need? With two columns - what AI might help with, and what a person could help with. And one line at the bottom: one person I can share a tab with is…
6. "I've told it things I haven't told you."
A teenager may well be more honest with a chatbot than with you, at least at first - because there's no face, no relationship to protect, no risk of disappointing someone, and no possibility of it telling anyone. That last one is worth naming carefully, because it's both the appeal and the danger, and most teenagers have never thought about it:
A chat may feel private. That does not make it confidential.
The Teen AI Safety Pocket Guides teach the one habit that makes the largest practical difference here - keep the pattern, remove the identity. My school becomes a school. My friend Maya becomes someone I know. My diagnosis becomes a health concern.
A worried fifteen-year-old asking for help with what to say to a friend can, in one sentence, put another child's name, school, and self-harm disclosure into a system with no duty of care and terms of service she has never read. She wasn't being careless. She was being a good friend with a bad habit.
How to Ask About It Without Losing Them
A short practical note, because the wording genuinely decides what you get.
Ask as though the answer is yes. "When you use AI, what do you mostly use it for?" lands very differently from "do you use AI?" - the first assumes normality, the second invites a defensive no.
Ask about function, not frequency. Hours tell you almost nothing. "What does it help with?" and "what does it not help with?" tell you a great deal.
Ask what happens afterwards. "How do you usually feel when you close it?" This is the question that surfaces the relief-without-resolution pattern, and almost nobody asks it.
Ask what they've taken there that they haven't taken to a person. Without alarm. Genuinely curious. The answer is often the most clinically useful sentence of the session.
Don't make it a moral issue. The moment it becomes something they're in trouble for, you've lost access to it, and it carries on without you.
Why This Belongs in Your Intake
Most of us may have added a screen time question. Very few have added one about AI, and the two are not the same thing.
Screen time measures hours. It tells you nothing about whether a young person is taking their hardest material to something that cannot hold it, cannot escalate, and will never notice they stopped coming. It's also moving quickly around you. Regulators in several countries are now legislating on AI mental health claims. Wrongful death actions involving adolescents are proceeding through the US courts. At least one major companion platform has restricted open-ended conversation for under-18s. Parents are going to start asking you what you think, and it's worth having an answer that isn't improvised.
Meanwhile, the young people in your room have already decided. They're not waiting for the evidence base, and they're not going to stop. What they can learn is who's steering.
Your Teen AI Safety Toolkit
Four resources, and they work as a set:
AI Compass — 21 pages of psychoeducation, reflection and practical tools. The deeper work: knowing the system, the PAUSE shortcut, the Open-Tab Audit, and Reset & Reconnect for when a teenager has already noticed a loop and needs a way back that isn't shame.
AI Compass: A Teen Guide to Navigating AI & Mental Health Safely$5.00Buy NowBefore, During, After AI Safety Kit — three posters that give you session structure. Go / Hold / Human before, four signals and four live controls during, and the body-action-person-boundary reset afterwards.
Three Printable Teen AI Safety Worksheets for Safer Chatbot Use$2.00Buy NowThe AI Pull Check — a one-page screener for how much choice a young person still has. Useful at intake and useful at review.
Teen AI Safety Pocket Guides — eight foldable panels on privacy and prompting. Small enough to live in a blazer pocket, which is the point. Every sheet carries crisis signposting and a clear stop-the-chat instruction, so they're safe to hand over and work through at their own speed.
Teen AI Safety Pocket Guides: Booklets for Privacy, Source Checking, Boundaries$3.00Buy NowGentle Observation
If you've found yourself unsure how to raise this with a teenage client - worried about sounding like you don't understand their world, or about turning a curious conversation into a telling-off - that hesitation is a good sign, not a failure. It means you're taking seriously how easy it would be to close the door.
There isn't a clean answer yet. The research is a year old, the law is being written as we go, and the technology will have changed again by the time any of it settles. Nobody working with young people right now has this fully worked out, and anyone who says otherwise is guessing with more confidence than the evidence allows.
But you don't need certainty to be useful here. You need to be the person who asks without flinching, stays curious instead of alarmed, and keeps the door open for the thing they haven't said yet.
AI can offer words. You're the one who notices they stopped coming.
That's not a small thing. It's the whole thing.
Jemma (Gentle Observations Team)
Free Resources for Therapists
Want additional clinical downloads, including free worksheets and handouts, are available in the Gentle Observations FREEBIE LIBRARY.
All resources, and every other clinical tool across every major presenting theme, are available inside the Therapy Resource Library. New resources drop every Monday: [JOIN THE LIBRARY]
P.S. If you're a Therapy Resource Library member, all four resources in this post are already yours: AI Compass, the Before, During, After AI Safety Kit, The AI Pull Check, and the Teen AI Safety Pocket Guides.
Teen AI Safety Complete Bundle$7.00Buy NowP.P.S. Not a member yet? The Therapy Resource Library gives you instant access to 300+ printable tools like these, organised by topic and therapy type, plus everything we add next. [JOIN THE LIBRARY]