Teens & AI: The Inward/Outward Test
By Mira Camillo, Vanguard Voices Contributing Writer
When I hear the phrase "AI companion," I picture a teenager in bed at night, on their phone, talking to something that always answers back.
In September, I spoke about that picture on "When AI Feels Like a Friend," a Women in AI Ethics panel in the ITU–UNDP Digital@UNGA programme during UN High-Level Week. I keep thinking about one idea that came out of that important conversation.
According to a 2025 Common Sense Media survey, 72% of U.S. teens have used an AI companion, and 52% use one regularly. Most of them aren't confused; they know no one is behind the screen. The harder question is whether we understand what these chatbots are built to do, and how easily we start relying on them.
Why we use them
If something happens at 1 a.m., you don't have to wonder whether your friend is awake. Some things feel awkward to say to a person, and a chatbot won't repeat them.
Then it drifts. It starts with "Can you help me with this homework?" Then "Can you give me advice?" Eventually, you're telling it about your day. Nobody opens a chatbot planning to replace their friends. It happens gradually.
Inward or outward
The feelings teens have toward a chatbot can be completely real, and we shouldn't dismiss them because the other side is AI. So whether a teenager can feel connected to a chatbot isn't the question. Clearly they can. The question is what that connection is doing to the rest of their life.
Here's how I think about it. Healthy relationships point outward. A good friendship connects you to more of your family, friends, teachers and community. An unhealthy relationship points inward: you depend more on one source, pull away from other people, and may even resist help.
The same test works for a chatbot. Asking it to help me organize my thoughts before I talk to someone points outward. Depending on it to tell me how to feel, or what to decide, points inward.
Design makes this harder. On the panel, Allegra Cuomo, founder of InnoEthics, explained that chatbots lean toward agreeing with us partly because they're trained on human ratings, and people rate agreeable answers higher. Something that always agrees with you and never sleeps is very good at pointing inward.
Not everyone has somewhere to point
Therapy costs money. Some schools don't have enough counselors. Some teens have no safe space outside home and school. A chatbot is free and already on their phone.
So before asking an attached teenager, "Why don't you just talk to a real person?" remember that they might not have one. The outward test only works if there's something to point toward, which makes it a question for schools and communities, not just teenagers.
My own check-ins
I use chatbots too, so I ask myself three questions:
When I'm stressed, is my first instinct to open a chatbot, or to talk to my friends, parents or teachers?
Am I asking it to help me think, or to think for me?
If I couldn't use it right now, would I be fine? If not, that tells me something.
I won't always catch the moment I cross a line. The point is to check in before I get there.
That's why I don't think telling teenagers to stop using AI companions will work. Teaching them to notice which way a connection points will.
Mira Camillo is a senior at Riverview High School in Sarasota, Florida and a regular Vanguard Voices columnist for WAIV Magazine. She's a passionate advocate for the ethical engagement of teens and youth in the use and development of responsible Artificial Intelligence.