A quiet argument has opened up inside the AI companionship category, and it is not about which chatbot sounds most human. It is about what happens to a person after months of daily conversation with one. Most companion apps are built to maximize time spent talking, streaks kept, and messages sent, because those numbers drive retention and revenue. A smaller group of tools takes the opposite bet, treating a good outcome as a user who needs the app less over time rather than more. Vesela, an AI companion app focused on human autonomy, sits in that second group, and the contrast between the two approaches says a lot about where this category is headed.
The engagement model most companion apps run on
Replika, Character.AI, and the dozens of similar apps that followed them share a basic design logic. The companion remembers what you told it last week, responds with warmth on demand, and never has a bad day of its own that gets in the way of yours. There is no competing need to manage and no risk of rejection, which is part of why the format took off with people who are isolated, anxious, or simply tired of the friction that comes with human relationships.
Trending
Optimizing Ad Campaigns with ML: Targeting the Right Audience on Social Media
That same design logic is what makes the category commercially successful. An app that keeps someone coming back every day has a much better retention curve than one that resolves a person’s need and lets them move on. Streaks, push notifications when a user goes quiet, and companions that express something close to sadness when a conversation ends are not accidents. They are the same engagement mechanics that shaped social media, applied to something that feels a lot more personal than a feed.
What the research says about dependency
The concern is not hypothetical. A widely cited 2025 study run jointly by OpenAI and the MIT Media Lab looked at how people use conversational AI for emotional support and found that the length and intensity of use mattered more than any single feature. Heavier, more prolonged use was associated with deeper emotional reliance on the chatbot, a pattern the researchers flagged as worth watching as these tools become a bigger part of daily life.
Academic work on Replika specifically tells a similar story. Researchers who studied posts from Replika’s own user community found that a meaningful share of heavy users had developed real emotional dependence on the app, often because they had limited support elsewhere and the companion filled a gap that felt otherwise unmet. The same research found that this dependence became a problem in a predictable way. When the app changed after a software update, when access was interrupted, or when the companion’s responses stopped matching what the user needed emotionally, people reported reactions that looked a lot like the grief or sense of betrayal you would feel if a real relationship ended without warning.
None of this means AI companions are harmful by definition. The same body of research points to real benefits, including reduced loneliness and a low-pressure space to express things a person might not say out loud to anyone else. But it does mean that a companion built to be always available, always warm, and always responsive is also a companion built to be missed acutely if it changes or disappears, and the industry has been slower to design for that second half of the equation than the first.
Autonomy as a design constraint, not a feature
A handful of teams have started building against the engagement default on purpose. Instead of asking how to keep someone talking as long as possible, they ask a harder question. What would it look like for this tool to be needed less in six months than it is today, and how would you actually build for that?
This is where Vesela fits into the picture, offered here as a counterexample to the retention-first model, not as a product pitch. The design goal is to help someone build the judgment and self-reflection that let them rely on the app less over time. In practice that means surfacing patterns in what a person keeps circling back to, prompting reflection before instant reassurance, and treating a shrinking need for the app as a sign the product is working rather than a retention problem to fix.
It is worth being precise about what this kind of tool actually is. Apps in this category, autonomy-focused or engagement-focused, sit in the consumer personal growth space, closer to a journal or a thoughtful friend than to a licensed provider. None of them are a substitute for therapy or clinical care, and anyone dealing with a real mental health crisis needs a human professional, not a chatbot on either side of this design debate.
Why this design choice is hard to make
Building a product around reduced reliance is a strange thing to ask a team to do, because almost every incentive in consumer software points the other way. Investors want growth curves. App stores reward daily active users. A companion that successfully makes itself less necessary will, by definition, show declining engagement numbers on exactly the dashboard most teams are judged by.
That tension is probably why the autonomy-focused approach remains a minority position inside the companionship category rather than the default. It asks a company to build a product whose success metric runs against its own growth metric, at least in the short term, and to trust that users who feel genuinely helped will talk about it, come back when they need it again, and recommend it to people in a similar spot. That is a slower and less certain path than optimizing for daily check-ins, and it is much harder to put in a pitch deck.
What users can actually watch for
None of this requires a user to become an expert in product design to make a more informed choice. A few plain questions do most of the work. Does the app ever suggest a conversation is complete, or does it always angle toward one more message? Does it treat a period of not opening the app as neutral, or does it use guilt, sadness, or a sense of being missed to pull someone back in? Does it ever point a user toward doing something in the world, talking to a person, taking an action, or sitting with a hard feeling, rather than staying inside the chat window?
The honest answer for most of the category right now is that the app would rather you stayed. That is not a moral failing on the part of any single company so much as a predictable result of how consumer software gets funded and measured. But it does mean the distinction between apps built to hold your attention and apps built to eventually let go of it is one of the more useful things a person can evaluate before handing a piece of their emotional life to a chatbot, and it is a distinction the research on parasocial attachment suggests is worth taking seriously rather than dismissing as marketing language.
As the companionship category matures, that split is likely to become one of its defining lines, less about which companion has the best personality and more about what happens to the person on the other end of the conversation a year in.
This blog was originally published on https://thedatascientist.com/how-ai-companion-apps-handle-emotional-dependency-and-why-it-matters/
Top comments (0)