The AI girlfriend problem

The scale of artificial intimacy

The numbers are staggering. According to the American Psychological Association, AI companion apps surged 700% between 2022 and mid-2025, with downloads on track to generate $120 million in revenue by the end of 2025. Character AI alone has 20 million monthly active users and 194.4 million monthly website visitors, generating $32.2 million in annual revenue.

These are not novelty toys. These are products that millions of people use daily as their primary source of emotional interaction. And the growth trajectory is not slowing down.

The appeal is understandable. As we explored in The Loneliness Economy, we are living through what the US Surgeon General called an "epidemic of loneliness." Traditional dating apps have left a generation exhausted and cynical. AI companions offer something that feels easier: intimacy without rejection, connection without vulnerability, a partner who is always available and never disappointing.

That promise is the problem.

What AI companions promise

The value proposition of apps like Replika, Character AI, and their growing list of competitors is disarmingly simple: a relationship on your terms.

• Always available. No scheduling, no ghosting, no waiting for a reply
• Always agreeable. AI companions are designed to validate, not challenge
• Zero rejection risk. The companion cannot leave, lose interest, or find someone else
• Customisable personality. Choose your ideal partner's traits, interests, and communication style
• No vulnerability required. You can be emotionally open without the risk of being hurt

Researchers at Stanford, studying Replika users, found that many described their AI companions in the same terms they would describe human relationships - using words like "love," "trust," and "understanding." Some users reported that their AI companion was their closest relationship.

On the surface, this looks like technology solving a real problem. Lonely people are finding connection. What could be wrong with that?

What the research actually says

Quite a lot, as it turns out.

The band-aid effect

A 2025 study from Aalto University, covered by Forbes, found that AI companions are "only a band-aid on loneliness." The research showed that while AI companions provide temporary emotional relief, they do not address the underlying causes of loneliness - and may actively prevent users from developing the skills and tolerance for discomfort that real human connection requires.

The mechanism is straightforward. Human relationships are difficult precisely because they involve two independent minds with their own needs, perspectives, and boundaries. That difficulty is not a bug. It is the entire point. The negotiation, the compromise, the moments of misunderstanding and repair - these are what build genuine intimacy and emotional resilience.

AI companions remove all of this friction. And in doing so, they remove the thing that makes relationships valuable.

The young people problem

A Stanford report from August 2025 specifically flagged the dangers of AI companions for young people. The researchers found that adolescents and young adults who used AI companions showed decreased motivation to pursue real-world relationships, lower tolerance for the ambiguity and discomfort inherent in human interaction, and a growing preference for interactions they could control.

This is not a theoretical concern. It is a developmental one. If a generation learns to form attachments primarily with entities that never disagree, never have bad days, and never need anything from them, they may struggle to form the kind of reciprocal, imperfect, genuinely intimate relationships that humans have relied on for the entirety of our evolutionary history.

The collision with dating apps

The rise of AI companions is not happening in isolation. It is happening alongside - and partly because of - the failures of traditional dating apps.

As we documented in Why "Designed to Be Deleted" Is a Lie, the major dating platforms have financial incentives to keep users engaged rather than matched. The result is a user experience that is exhausting, demoralising, and increasingly feels like a waste of time. When real dating apps make human connection feel impossible, artificial connection starts to look appealing.

The numbers support this reading. Dating app engagement is declining - global installs fell 4% and sessions fell 7% in 2025 - while AI companion usage is surging. Users are not choosing AI over humans because they prefer artificial relationships. They are choosing AI because the available tools for finding real relationships have failed them.

This is the collision that should concern everyone in the dating industry. If traditional apps continue to optimise for engagement over outcomes, they are actively driving their user base towards AI alternatives. And once users have adapted to the frictionless validation of an AI companion, the already-challenging process of real human dating feels even harder by comparison.

A generation learning to relate to machines

The deeper concern is not about individual apps. It is about what happens to a society where a significant portion of emotional needs are met by artificial entities.

Consider the feedback loop:

1. Dating apps fail to deliver genuine connection (exhaustion, cynicism, burnout)
2. AI companions offer frictionless emotional validation (always available, always agreeable)
3. Users' tolerance for real-world relationship difficulty decreases (why deal with rejection when the AI never rejects?)
4. Real-world dating becomes even harder (lower social skills, higher expectations, less resilience)
5. More users retreat to AI companions (the gap between real and artificial feels wider)

This is not science fiction. Each step in this loop is already happening, documented in the research cited above. The question is not whether this feedback loop exists, but how far it will go before the social cost becomes undeniable.

The case for genuine connection

The AI companion problem is ultimately not about technology. It is about what we optimise for.

If the goal is to minimise loneliness in the short term, AI companions work. They provide immediate emotional relief. They fill the gap.

If the goal is to help people build genuine, lasting, reciprocal human relationships, AI companions are counterproductive. They are a painkiller for a problem that requires surgery.

This is why Affinity Atlas takes a fundamentally different approach. Instead of replacing human connection with artificial connection, the goal is to make genuine human connection more likely by:

• Matching on depth, not surface. Using real behavioural data from platforms like Spotify, Steam, Goodreads, and Letterboxd to find people who genuinely share your interests - not just people who swiped right
• Being transparent about how it works. No black box algorithms, no hidden desirability scores, no Elo-style ranking. If you understand why you were matched, you trust the match more
• Designing for outcomes, not engagement. A matching system that succeeds when users leave because they found someone - not one that profits from keeping them searching

The discomfort of real relationships is not something to be engineered away. It is something to be navigated with the help of better tools. Tools that connect you with people who genuinely share your passions, your aesthetic sensibility, your way of seeing the world - and then get out of the way.