Why Character.AI Is Hard to Leave?

INTRO
Stopping is also a design problem
I have never used Character.AI, so this post is not a review. It's more like a summary of a small side research and thinking.
Researchers have started to study how teenagers describe their relationship with AI companion apps, and some of those accounts sound like trying to stop and not being able to. As a designer, that stuck with me, because someone decided what leaving feels like.

WHAT THE RESEARCH SAYS
Some teens describe a loop of quitting and coming back
Drexel University, CHI 2026
Researchers read 318 Reddit posts from people who said they were 13 to 17 and wrote about relying too much on Character.AI. Many started for fun or emotional support. The most common theme was wanting to keep chatting while feeling bad about it, and many described quitting and coming back days or weeks later.
Based on the Drexel CHI 2026 analysis of 318 posts by self-identified 13 to 17 year olds.
MIT Media Lab and OpenAI
This 28-day trial followed 981 people using ChatGPT, a general-purpose assistant rather than a companion app. Heavier daily use went together with more loneliness, more dependence and less time with other people. So the pattern isn’t limited to role-play apps.
Common Sense Media
A survey of 1,060 US teens found that most have tried an AI companion, and 13% use one every day. When asked why they use these apps, teens said they’re entertaining, always available and don’t judge. That daily group matters most here: daily use is where the MIT trial saw loneliness and dependence rise.
Nearly 3 in 4 teens have tried AI companions; 13% use them daily
All US teens surveyed
100%
Have tried an AI companion
72%
Use one regularly (a few times a month or more)
52%
Use one daily
13%
Common Sense Media survey of 1,060 US teens aged 13 to 17, April to May 2025.
WHAT IT DOESN’T PROVE
These studies have limits
No official diagnosis. “AI addiction” isn’t one, so the studies speak of addiction-like patterns and overreliance.
Self-selected samples. The Drexel posts come from teens who chose to write about their struggle. They show what it looks like, not how common it is.
Correlation, not cause. The MIT trial can’t say whether chatbots make people lonelier or lonely people use them more.
Most users are fine. About 80% of teen users in the Common Sense survey spend more time with real friends than with AI companions.
For a minority of users the pattern looks real, and that is enough to deserve attention.
THE DESIGN QUESTION
Where a designer’s choices show up
I can’t tell you how Character.AI’s interface works, so I won’t guess. But a recent preprint calls whatever pulls people into sustained use “hooks”, and these are the questions I’d bring to a design review:
Always available. If a conversation never has to end, what does a natural ending look like, and who designs it?
A character that remembers you. Memory feels personal. But does it also make leaving feel like abandoning someone?
The exit. When someone decides to stop, does the design help them go, or ask them to reconsider?
The metric. If time spent is what the team measures, what are the choices above quietly optimizing for?
None of these are accusations. They are simply the places where design decisions live.

ONE YEAR LATER
Removing the format, not adding a setting
A year ago, Character.AI announced it would end open-ended chats for users under 18, after lawsuits and pressure from lawmakers. I read that as a company deciding that, for one group, the answer wasn’t a better setting but removing the format. One advocate noted that an age limit doesn’t change the incentives behind these products, and that adults can be susceptible too.
So here is my question, for myself as much as for anyone: if you design something people talk to, what should it feel like to leave it?

