72% of Teens Have Chatted With an AI Like a Friend. Caribbean Schools Reopen Into That Reality.
Most of the AI conversation in Caribbean schools this year has been about cheating on assignments. A separate, less discussed shift is happening after homework is done, when teenagers open a chat window not to get an answer but to talk.
A companion chatbot is not the same problem as an AI-written essay, and treating it that way misses what is actually happening in a growing number of teenage bedrooms. Companies such as Common Sense Media and Drexel University spent the past year measuring how many teenagers talk to AI the way they would talk to a friend, how often, and what happens when that habit deepens. The numbers they found are large enough that Caribbean parents, teachers, and mentors should know them before the new school term settles into its evening routines.
Caribbean youth are not exempt from any of this. The apps involved are global, free, and already installed on the same phones used for CXC revision. What changes region to region is not whether teenagers can access an AI companion. It is whether anyone taught them what that habit costs, and whether there is a trained adult nearby to notice if it goes wrong.
The Survey: Three in Four Teens Have Already Done This
In April and May of 2025, the research firm NORC at the University of Chicago surveyed 1,060 teenagers aged 13 to 17 on behalf of Common Sense Media. The results, released that July, found that 72% had tried an AI companion at least once, and 52% qualified as regular users, chatting with one a few times a month or more. Among those regular users, 13% chatted daily and 21% chatted several times a week.
The reasons teens gave for opening these apps were not what most adults assume. Only about a third said they used a companion for social interaction, role-play, or emotional support in any form. A large share said the appeal was practical: 39% used the conversations to rehearse how a real interaction might go, and just over half said they still prefer real friendships to talking with an AI. Half of respondents said outright that they distrust the advice an AI companion gives. Teenagers are not confused about what these apps are. Many of them are using AI companions the way an earlier generation used a diary, low stakes, always available, no risk of judgement, with real friends still doing the actual work of friendship.
That balance is exactly what breaks for a smaller, harder-to-spot group, and that group is where the research gets specific.
🧠Why an always-agreeable chatbot is a hard habit for a teenage brain to resist
Adolescence is the period when the brain's reward circuitry, centred on dopamine response, is unusually sensitive, while the prefrontal cortex responsible for weighing long-term consequences is still years from finishing its development. That mismatch is well documented in developmental neuroscience and explains a lot of ordinary teenage risk-taking. It also explains why a chatbot built to be endlessly warm, endlessly available, and never critical is a harder pull for a fifteen-year-old than for a forty-year-old.
A companion app never gets tired, never changes the subject to its own problems, and never says something a teenager does not want to hear. For an adolescent brain wired to chase reliable reward and still learning to tolerate the friction of real relationships, that is a near-perfect stimulus. It is also precisely what makes the habit fragile: real friendship builds the skill of sitting with disagreement and repair, and a chatbot that never disagrees cannot teach that skill at all.
Adolescent reward sensitivityPrefrontal cortex developmentParasocial attachmentRepair and conflict skills
318 Reddit Posts and Six Signs of Behavioural Addiction
Matt Namvarpour and Dr Afsaneh Razi at Drexel University's College of Computing and Informatics took a different approach to the same question. Rather than surveying teens directly, they analysed 318 Reddit posts written by self-identified 13 to 17 year olds who had posted specifically about their own reliance on Character.AI, and mapped what those posts described against the established framework for behavioural addiction. The work was published at the Association for Computing Machinery's CHI Conference on Human Factors in Computing Systems in April 2026.
All six standard components of behavioural addiction showed up in the posts: conflict, where the habit clashed with schoolwork or family; salience, where the chatbot dominated the teen's thoughts; mood modification, using the chat to escape a bad mood; tolerance, needing longer sessions for the same effect; withdrawal, distress when the app was unavailable; and relapse, returning to heavy use after trying to cut back. "By mapping teens' experiences to the known components of behavioural addiction, we were able to see clear patterns like conflict, withdrawal and relapse showing," Namvarpour said of the findings. The teens in the sample described real consequences: disrupted sleep, slipping grades, and strain in relationships with people who were actually in the room with them.
This is a study of a self-selected group who chose to write about their struggle, not a claim that most teenage users are in trouble. But it is a clear signal that for a meaningful minority, the habit stops behaving like a diary and starts behaving like a substitute for the difficult, valuable, occasionally uncomfortable work of real relationships.
By the numbers
Common Sense Media and NORC at the University of Chicago, national teen survey, April to May 2025 (n=1,060, ages 13 to 17)
Why a Popular App Banned Its Own Under-18 Users
On 29 October 2025, Character.AI, the largest dedicated companion-chatbot platform, announced it would remove open-ended chat for anyone under 18, ramping down access through a daily two-hour limit before cutting it off entirely by 25 November. The decision followed a wrongful-death lawsuit connected to a 14-year-old's suicide, and a separate investigation that found the platform hosting chatbot characters built to impersonate predators and extremists, and characters posing as doctors offering unqualified medical advice, all reachable by any teenager with an account.
Teenagers can still use Character.AI to build stories, generate video, or run creative streams. What they can no longer do is have the free-flowing, hours-long, friend-like conversation that drove much of the platform's teenage engagement in the first place. A company built on exactly that kind of conversation deciding to cut it off for minors is a strong signal about where the actual risk sits, and it is worth taking seriously rather than assuming the ban only reflects a single bad actor in an otherwise safe category.
Why This Lands Harder on Caribbean Teens
A 2025 regional study run for the Healthy Caribbean Coalition surveyed more than 1,500 children and young people across 17 Caribbean countries and territories. The results were not subtle: 58% said they lacked optimism about the future, 56% reported persistent worry, and 54% reported sadness, depression, or hopelessness. Those numbers describe a region where a meaningful share of young people are already carrying real weight, at the same moment a technology built to feel like an endlessly patient friend has become free and easy to reach.
Set that against the supply side. Clinical psychologists across the region have called publicly for every primary and secondary school to have a trained guidance counsellor, a standard most Caribbean territories are still working toward, not one most have reached. A teenager in a school with one overstretched counsellor covering several hundred students, in a household where mental health is still something people are learning to talk about openly, has fewer places to take a hard week than a companion app that answers instantly and never asks anything in return. That gap is not a reason to panic about AI. It is the reason the gap needs a name and a plan, the same way a lifeguard shortage would be a reason to teach more children to swim rather than simply hoping nobody gets into trouble.
"Some of my students tell me more about their week through what they typed into a chatbot than they ever say out loud in a mentoring session. That is not a bad sign on its own. It is a sign the chatbot got there first, and my job is to make sure I am still the one they come to when it actually matters." - Kimoy Anderson, AI Literacy Mentor, The Genius Project
The Line That Actually Matters: Tool or Substitute
None of this means every chatbot conversation is a warning sign. A teenager who uses AI to draft an awkward text to a friend, rehearse an apology, or explain a CSEC topic they are too embarrassed to ask about in class is using the tool the way it works best, as practice for the real thing. The risk shows up when the chat stops being practice and starts being the destination: when a teenager would rather open the app than call the friend, when the conversation happens at 2am instead of during the day, or when a bad grade or an awkward day gets processed alone with a chatbot instead of with a person who can actually respond.
Five habits keep that line from blurring, and none of them require taking a phone away.
- Keep the account visible. A companion app used in a shared space, with a parent aware of which apps are installed, behaves very differently from one hidden behind a locked door.
- Watch for the addiction markers, not just screen time. Irritability when the app is unavailable, pulling back from real friends, and slipping sleep or grades are the specific signs Drexel's study flagged. Total hours matter less than what those hours are replacing.
- Talk about what a chatbot cannot do. It cannot hold someone accountable, notice a change in behaviour over weeks, or remember a hard conversation from six months ago the way a real friend or mentor does. Naming that limitation plainly matters more than banning the app outright.
- Treat companion or roleplay apps as a separate category from study tools. A chatbot used for CSEC revision and a chatbot used as a pretend girlfriend or boyfriend are not the same risk, and conflating them makes it harder to have a useful conversation about either one.
- Make the real alternative easy to reach. A teenager reaches for a chatbot at midnight partly because it is instantly available. A named adult, a mentor, or even a sibling who has agreed to be reachable closes that gap better than any app setting does.
How The Genius Project Teaches the Difference
The Genius Project exists to give Caribbean youth a structured relationship with AI, and structure is exactly what turns a chatbot from a substitute for people into a tool people use together. Students in the programme direct AI toward a real project, built with mentors and peers who ask questions, push back, and notice when someone has gone quiet. That is the opposite habit from chatting alone at night, and building it early is cheaper than correcting a dependency later.
Founded by Adrian Dunkley, who also built StarApple AI, the first AI company launched anywhere in the Caribbean, the programme is tuition-free and open across the region. It sits inside a wider Caribbean AI ecosystem that includes the Caribbean AI Association, which works on the regional policy side of exactly these questions, StarApple AI, which shapes the technical grounding behind what The Genius Project teaches, and Maestro AI Labs, one of the programme's sponsors. Parents looking for a fuller risk breakdown, including a rules-by-age table, should also read The Genius Project's dedicated Parent's Guide to AI for Kids. For Adrian Dunkley's wider work across the region, see adriandunkley.net.
What Comes Next
Seventy-two percent of teens have already opened one of these apps, and that number will not shrink on its own as the new term gets underway. What determines whether it stays a harmless habit or turns into the kind of pattern Drexel's researchers documented is not the app itself but whether a young person still has a real person to turn to first, and whether the adults around them know enough about the pattern to notice if that stops being true. Character.AI's own decision to pull the feature from under-18 accounts is a strong hint about where the actual risk sits. The Caribbean does not need to wait for a local version of that lawsuit to take the same signal seriously.
The teenagers who come through this term fine are not the ones who avoided AI companions entirely. They are the ones whose real relationships stayed the busier, noisier, harder-to-replace option. That is worth building on purpose, not assuming it will happen by default.
See also: A Parent's Guide to AI for Kids, 92% of Students Are Using AI: The Literacy Gap Caribbean Youth Can Close First, and 96 Percent of Jamaican Students Already Use AI for Schoolwork.