AI Companionship

Roleplay/companionship is quickly emerging as a juggernaut use case by tokens for LLMs but you wouldn't know it if you follow the popular AI zeitgeist.

Just looking through the data on apps indexed by OpenRouter, roleplay/AI apps use 28.8% of total tokens while coding apps use 45.2%. And a glance at the top 10 apps on openrouter is all coding apps and roleplay apps:

Of course, this only considers apps indexed by OpenRouter, but you wouldn't think the roleplay category exists at all if you perceive reality through the mainstream AI discourse.

Why is that?

kids can't pay

Another interesting datapoint on OpenRouter is that roleplay apps use different models than programming apps:

Roleplay app model usage

Programming app model usage

Note that the vast majority of tokens are consumed by DeepSeek models. I don't think it's an accident that they use open source, free to host models for their inference.

And using these apps (I've been losing myself in janitorAI and dreamGen recently) is free. How is such a big token consumer not making any money?

The conclusion I've come to is that it's a userbase that can't pay for apps yet: kids. And kids aren't a great market unless their parents are the buyer; they simply don't have deep pockets.

AIM led to Facebook

The older members of my audience might remember AOL Instant Messenger, colloquially AIM. In middle school, AIM was how we stayed in touch after school.

It gave us a new affordance; the ability to spend time together without asking our parents. And of course, without access to a credit card.

Of course that meant that AIM didn't make any money despite being huge among teens of that era. Eventually AIM was discontinued, but the core communication and social primitives lived on and emerged in a new form years later for us in Facebook.

Facebook's status posts replaced AIM's away messages. And we all know how much money Facebook made.

Facebook appealed to something we wanted and learned we wanted years earlier in AIM but came to us at a time when our eyeballs started having value. Right before we became wallets to advertisers.

AIM was timed incorrectly and it was also prescient.

I'm sure my reason for this digression into early social media is obvious to you by now. I have a hunch that the same pattern might be repeating itself with AI roleplay apps right now.

The current crop might be AIM, but when its users become 'valuable' young adults, the Facebook will emerge and be enormously profitable.

what's next?

The simplest answer for why this area seems under-explored is that there isn't any money in roleplay yet.

However, as the (AI) industry matures, we'll likely see companion platforms evolve beyond character roleplay into more sophisticated emotional and social media. The generation currently using these apps for free will age into consumers willing to pay for AI relationships that feel less like games and more like genuine support systems and virtual worlds.

The winners won't be the companies building the most realistic characters, but those who understand that companionship, unlike productivity, isn't about optimization; it's about presence, consistency, and emotional resonance.

Perhaps the AI zeitgeist isn't just missing a market; it's missing a fundamental shift in how a generation relates to AI.