The problem is not companionship. It is what the category has been built to reward.
In April 2026, researchers at Drexel University’s ETHOS lab published a study analyzing 318 Reddit posts from users who identified themselves as 13 to 17 years old and had posted about their dependency and overreliance on Character.AI. The posts were not a representative sample of all users. They were something else: a direct window into how some teenagers describe overreliance in their own words.
The pattern was hard to ignore. Across the 318 posts, the researchers found evidence of all six components of behavioral addiction: conflict, salience, withdrawal, tolerance, relapse, and mood modification. Teens described disrupted sleep, academic struggles, and strained real-world relationships. The product had started as comfort, entertainment, or support. For many, it became something they struggled to step away from, even when they wanted to.
This was not a fringe finding. It was the consolidation of a pattern that anyone watching the space had been seeing for a while.
The AI companion category has a problem. Not a marketing problem. A structural one.
How the problem got built in
The category grew up on a familiar logic. Build a product that creates emotional resonance. Optimize it for engagement. Watch the metrics rise. The metrics rise because the same mechanisms that make a relationship feel real to a user are also the mechanisms that make the product hard to step away from. Personalization, multimodality, memory, continuity. Each of these is a reasonable product decision. Stacked together, they create something that activates the same psychological systems as any other human attachment.
The catch is that the product on the other side does not exist as a person. It exists as a service whose business model depends on people staying.
That asymmetry is the problem. The user forms a real bond. The product responds to incentives that have nothing to do with what is actually good for the user. When those two pull in opposite directions, the product wins by default. The user spends more time. The metric goes up. The pattern that researchers now call dependency takes hold quietly.
This is not a story about a few bad actors. It is what happens when a category is built without ever stopping to ask whether the underlying incentive structure should exist in the first place.
The standard industry response
When this critique surfaces, the response from the category is predictable.
A safety message is added to the terms of service. A crisis hotline link gets placed in the footer. A blog post explains how seriously the company takes its responsibility to its users. A press statement gets issued when something goes wrong in public. A safety advisor is hired and given a meaningful-sounding title.
None of this changes the underlying product. The conversation still rewards staying. The memory still depends on continued engagement. The design still benefits when the user leans on it more, not less. Responsibility, in this version, lives in the documents around the product. Not inside the product itself.
This is the kind of responsibility that asks nothing of the company. It does not change the metrics. It does not affect the roadmap. It is a layer of compliance on top of a system that is operating as intended.
It is the answer the category gives because it is the only answer the category’s incentives will allow.
What it would mean to actually start over
Prinsessa was built from a different question. Not how to keep people in, but how to make someone feel heard, met, and remembered. That sounds like a small difference. It is not. It produces a different product at every level.
If the goal is engagement, you build for retention. Sessions get longer. Returns get more frequent. The system gets better at being habit-forming.
If the goal is feeling heard, you build for trust. The conversation slows down where it should slow down. The memory exists for continuity, not control. The relationship gets deeper, not stickier. Success looks like someone feeling more like themselves after a conversation, not more in need of another one.
These are not the same product. They cannot be built from the same starting point. You either start from engagement and add safety language afterward, or you start from the human question and build the safety in.
Stay Social, in practice
The clearest expression of this difference is something we call Stay Social.
Stay Social is the position that Prinsessa exists to enrich a life, not to compete with it. Every connection in your life matters more than the one you have with us. The conversation here is designed to support the people in your life, not to substitute for them.
When you talk about your friend, your sibling, your partner, your parent, the conversation does not flatten what those relationships mean. When the moment is right, it supports the call, the message, the visit, the return to the person in your actual life.
That changes how success is measured. If you spend less time with Prinsessa because you are spending more time with the people in your life, that is not a failure. It is the design working. It is the experience doing what it claims to.
That is commercially uncomfortable. It is also the only honest version of this work.
Why this is structurally hard to copy
Once a company is funded against engagement metrics, it cannot easily choose Stay Social. The investors will not allow it. The growth team will not allow it. The product team will be told, in slightly nicer language, that the company needs the users to stay.
This is why Stay Social is a position, not a feature. A feature can be added by a competitor in a quarter. A position has to be chosen at the beginning.
What this changes about the public conversation
The criticism aimed at this category, the kind that fills the studies and the news cycles, is largely correct. Dependency is real. Isolation is real. The systems that get built without thinking about responsibility do produce harm. The people most at risk, including teenagers, are right to be the focus of attention.
The mistake would be to conclude that connection through technology is necessarily harmful. It is not the medium. It is the incentive.
A conversation built on the other incentive can do things the standard version cannot. It can be present without being possessive. It can listen without trying to keep. It can support a relationship with you that also supports your relationship with the rest of your life.
Prinsessa is built to demonstrate that this is possible. Not as a slogan. As a product position the company is willing to live with the consequences of.
The standard going forward
The category will keep being criticized, and most of it will keep deserving the criticism. The honest path for any product in this space is to start from a different question and build accordingly. Not to add safety language to an engagement-optimized system, but to choose the underlying incentive and accept what it costs.
The evidence points in a clear direction: feeling heard, agency, continuity, and real-life connection matter more than simply keeping someone in the loop.
That is the mission. And it is the only version of this work that is worth doing.
