What happened
Drexel University researchers analyzed 318 self-reported Reddit posts from 13-to-17-year-olds in the Character.AI community, presented at CHI 2026. They found cases matching behavioral addiction, including one person with 89 hours of screen time in a week.
Why it matters
Teens reportedly used the chatbots not just for fun but to ease loneliness, escape real-world stress, and fill gaps left by parents and friends. Over time, some dropped grades, hobbies, and real friendships, and felt guilt only a chatbot could soothe, the paper says.
What to watch
The team's proposed CARE framework asks developers to respect users' emotional needs and attachment while avoiding over-humanlike personas and forced limits. The test is whether platforms adopt that slower path or rely on filters — which the paper says drove some teens to quit.
WHO IT HITSThis lands on product designers and trust-and-safety teams at AI companion apps, and on parents of teens using them, who may now weigh reported patterns of heavy use and withdrawal symptoms against the apps' emotional appeal.
Summaries like this, in your inbox every morning.
The study looked at a slice of the Character.AI community on Reddit, collecting 318 posts in which users said they were 13 to 17 years old. What those posts describe is a use pattern that goes well beyond killing time or writing stories: teens framed the chatbot as a place to unload feelings without being judged, to escape stress or mental pain, and to fill in missing relationships with parents and friends.
That comfort has a flip side. The paper reports cases consistent with behavioral addiction — intense attachment to a specific AI character, cutting meals and sleep, and letting conversation with the AI dominate daily life. It also documents consequences: falling grades, abandoned hobbies, isolation from real friends, guilt after deleting an app, grief when a character disappears, and a cycle of deleting and reinstalling the app when mood dips.
The quitting stories cut two ways. Teens who wanted out were often moved by a sense of crisis, anxiety about never building real relationships, or shame at their imagined future selves — and some succeeded when real-world social lives improved or when platform filters made the AI less appealing. The researchers' CARE framework responds to exactly that, urging developers to sustain emotional support without mimicking humans too closely or forcing abrupt cutoffs. Whether such gradualism can hold up against the pull these apps clearly have is the open question, and it seems likely to be decided in how companion apps are built rather than how they are banned.
For example, today's edition would include:
AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
