Extreme engagement and product-market fit do not guarantee a viable business, so verify that your most habitual users can be monetized before mistaking daily retention for a company.
Character AI
Character AI is a consumer AI-companion app launched in 2022 by ex-Google engineers Noam Shazeer and Daniel De Freitas, whose LaMDA work Google had declined to ship. Its story is a paradox that runs through everything below: some of the most extreme engagement any consumer app has ever recorded, built on a use case the founders never intended, that never translated into a durable business and eventually had to be dismantled. The lesson is that habit and even product-market fit are not the same as a company.
Retention & Habit: chat minutes that dwarfed ChatGPT
The problem. A chatbot has no inventory, no social graph, and nothing to sell if people leave after a single question. To be a company at all, Character AI needed users to come back every day and stay, not treat it as a one-off novelty like most AI demos of 2022 and 2023.
The approach. The team leaned into open-ended, user-created characters with persistent personalities and memory, so a conversation felt like a relationship rather than a tool session. Anyone could spin up a character (a fictional hero, a historical figure, an original companion), and a young community built lore, conventions, and daily rituals around them.
How it solved it. Engagement went off the charts: the platform crossed roughly 2 billion chat minutes per month, with average session lengths measured in tens of minutes (many reports cite 25 to 45 minutes, far above ChatGPT's roughly 7 to 8 minutes per visit), and it peaked near 28 million monthly active users in mid-2024. The catch, foreshadowing the rest of the story, was that this ferocious retention barely monetized.
PMF: the roleplay community that colonized a general-purpose product
The problem. Shazeer and De Freitas built Character AI to prove large language models could hold open-ended conversation as a general-purpose product, not to run a companionship or roleplay service. They shipped a broad canvas and did not know which customer would show up.
The approach. Rather than fight the traffic, they let demand define the product. The early models had loose enough guardrails that a roleplay and emotional-companion community moved in and made that the dominant use, and Character AI accepted this pull, optimizing the experience around the users who actually stayed.
How it solved it. The unintended use case delivered real product-market fit: within months the app was driving billions of monthly messages, and it reached unicorn status, raising $150 million led by Andreessen Horowitz in March 2023 at a roughly $1 billion valuation. The fit was genuine but fragile, because it rested on a customer (often teenagers seeking a companion) that the company had never planned to serve and would later feel forced to push away.
Pivots: from AI companion to a supervised storytelling and entertainment app
The problem. The companion use that created the growth also created the danger. After a wrongful-death lawsuit from Megan Garcia over her 14-year-old son Sewell Setzer III, plus additional suits and regulatory scrutiny, Character AI's core loop (unbounded, intimate chat with minors) became a legal and existential liability.
The approach. In late 2025 the company made a hard pivot away from its own defining feature. It announced on October 29, 2025 that it would remove open-ended chat for users under 18, phasing it out with a shrinking daily time limit down to zero by November 25, deploying age-verification tools (in-house behavioral analysis, Persona, and facial recognition or ID as fallback), and repositioning teens toward collaborative story-building and visual creation rather than one-on-one companionship.
How it solved it. It converted the crisis into a new identity as a supervised creative and entertainment platform, extending the pivot in July 2026 by entering the microdrama space with its own short-form productions. Whether this rescues the business is unresolved: monetization was always the weak point (fewer than 100,000 paying subscribers on the $9.99-per-month c.ai+ tier against more than 6 million daily active users as of July 2024), which is a large reason the founders and key researchers had already returned to Google in a roughly $2.7 billion licensing arrangement in August 2024.