Character.ai had 12 million monthly active users and still filed for bankruptcy. Here's why I resisted copying their playbook for Maibel.
Their formula: fantasy characters, hours of roleplay. Look at those engagement numbers!
I understand the appeal. Women using Maibel have drawn fan art of our companion, sent daily good morning texts, and walked extra steps to unlock more of his world.
In a sworn bankruptcy declaration, Character.ai's CEO reported operating losses of approximately US$8-9 million a month as of August 2026.
This involved multiple pressures, including lawsuits alleging harmful chatbot interactions with minors. Inference costs alone don't explain the collapse.
Still, the economics are worth examining.
Every AI response costs tokens. Long roleplay scenarios can mean lengthy outputs and repeatedly processing conversation history. Add memory retrieval, image processing and infra costs?
Two subscribers might each pay $10 monthly. One sends 20 messages a day. Another sends 400.
Revenue stays flat while usage costs can multiply.
Without careful model routing and context management, your most engaged users could become your least profitable. And spending more doesn't guarantee a better experience.
In Maibel's earliest iterations, our companion sent THREE photos daily.
By Day 7, users accused him of spamming them. Oops.
We later tried one photo in three weeks. A top fan sent him ELEVEN photos of herself, hoping he'd send another back.
We’d generated more content and annoyed users. Then we sent less, and one user actively tried to get more.
That fascinated me.
Novelty becomes the baseline fast. I suspect generating more content eventually produces diminishing returns, while the cost of maintaining expectations keeps climbing.
Before Maibel, I coached women who paid four figures monthly for 1:1 support.
I wasn't spending HOURS convincing them to exercise. Effective coaching meant understanding their psychology, knowing which levers to pull, and when to intervene.
Sometimes a timely check-in. Sometimes sharing stories, laughing, building rapport. Sometimes giving her space.
The value was in my judgment, not hours spent yapping.
That's the methodology I've spent over a year translating into Maibel.
And yes, women WANT the fantasy.
They've been asking to personalise their AI oppa's looks and personality, inspired by their favourite K-/C-drama archetypes.
So we're building that next. More freedom to shape their companion, with our coaching methodology and safety boundaries intact.
We're testing paid access and character experiences early. What do women value, what will they pay for, and can we deliver it sustainably?
We haven't proven our per-user economics. But I'm glad we're testing assumptions while we're tiny.
Engagement alone isn’t proof of a sustainable business.
I learned long before building AI thoat the value of a good coach isn't how long she keeps you talking.
It's knowing what will get you moving.
Hi, I’m Mabel.
I’m the founder of Maibel. I write about emotional AI, women’s wellness and learning to build the things I wish existed.
More about me