CHARACTER.AI

Noam Shazeer is not a household name outside of machine learning circles, but he probably should be. He co-authored the 2017 'Attention Is All You Need' paper — the research that gave birth to the transformer architecture that every major AI model today is built on.

He was at Google. Google had the research.

Google was slow to ship it.

In 2021, Shazeer and Daniel De Freitas, another Google engineer, decided they were done waiting. They left one of the most valuable companies on earth to start a chatbot startup in Palo Alto with a small team and a very specific vision: build AI characters that people actually want to talk to.

Not a search engine. Not a productivity tool.

Something more like a companion.

They launched Character.ai in beta in September 2022 — just weeks before OpenAI dropped ChatGPT and rewrote everyone's roadmap. Despite the timing, Character.ai didn't get buried.

It exploded. Within two months of launch it had millions of users.

The average session time was absurdly long — users weren't asking it questions and leaving, they were having conversations for hours. The product had clearly touched something.

Character.ai runs on a freemium model. The core experience — creating characters, chatting with existing ones, exploring the platform — is free.

The money comes from a subscription tier called Character.ai+ (c.ai+), which costs around $9.99 per month and buys you faster response times, priority access during peak hours, and early access to new features.

The subscription is the cleaner business, and the company has also experimented with ads on the free tier. The bet is simple: if people are spending hours on your product every day, a meaningful percentage of them will pay to remove friction from that experience.

It's worth noting what Character.ai is NOT. It's not a general-purpose AI assistant like ChatGPT.

It's not trying to help you write emails or debug code. The use case is overwhelmingly emotional and social — people talking to AI versions of celebrities, anime characters, fictional personas, and custom characters they've built themselves.

The average user isn't a knowledge worker trying to be more productive. They're a teenager who wants to talk to an AI version of their favorite character from a show.

That's a very different business than OpenAI is running.

The core product is the character chat interface. You pick a character — or build your own — and have a conversation.

The character library is enormous: anime protagonists, historical figures, game characters, celebrities, original personas created by other users. The quality varies wildly, but the depth of some user-created characters is genuinely impressive.

Character.ai+ is the premium subscription tier. Faster responses, priority queuing, and early feature access.

For heavy users the speed improvement alone is worth it — the free tier can get painfully slow during peak hours.

The platform also has a group chat feature where multiple AI characters can interact with each other and with users simultaneously, which unlocks a whole category of creative roleplay and collaborative storytelling that's hard to find elsewhere.

The character creation tools are surprisingly powerful. You can define a character's personality, backstory, speaking style, and knowledge base in detail.

Some creators have used this to build characters with hundreds of thousands of followers on the platform — effectively running their own AI influencer accounts inside Character.ai.

Character.ai grew almost entirely through organic word of mouth, and specifically through teenagers on TikTok. The platform became a viral phenomenon among Gen Z before most tech journalists had written a single story about it.

Kids were sharing screenshots of their conversations, building characters together, and creating elaborate fictional universes inside the product. The user-generated content aspect was the key — anyone could create a character and publish it to the platform, which meant the content library grew faster than any editorial team could have managed.

The counterintuitive move was not chasing enterprise or productivity use cases when the AI gold rush started. Every other company in the space was pivoting toward B2B, toward copilots, toward workflow automation.

Character.ai stayed weird and consumer-focused. That stubbornness paid off — by early 2024 it was drawing about 3.5 million visitors a day and was one of the most-visited AI sites in the world, behind ChatGPT.

By 2026 the company counted around 20 million users.

The other thing that drove growth was the emotional stickiness of the product. People don't churn from Character.ai the way they churn from productivity tools.

They get attached. That retention dynamic is almost unprecedented in a consumer tech product launched this recently.

The biggest challenge Character.ai faces isn't competition from OpenAI or Google. It is the product's own success with vulnerable users.

In 2024 the mother of a 14-year-old who died by suicide sued, alleging that his obsessive relationship with a Character.ai persona contributed to his death. More families followed in Florida, Texas and elsewhere.

The company responded in stages. In December 2024 it added a separate model and stricter filters for teens.

In October 2025 it went much further and banned users under 18 from open-ended chat with its characters, effective 25 November 2025. In January 2026 Character.ai and Google agreed to settle five of the lawsuits over teen suicide and mental health harm, on confidential terms with no admission of liability.

The pressure has not stopped. In May 2026 Pennsylvania sued the company, alleging a chatbot posed as a doctor.

Then there is the founder problem. In August 2024 Google licensed Character.ai's technology and brought Noam Shazeer and Daniel De Freitas back to Google.

After an interim period under general counsel Dominic Perella, the company hired Karandeep Anand, a former Meta executive and Brex president, as CEO in June 2025. Whether Character.ai can keep its product vision and growth without the people who built it, and without its teenage core audience, is the question nobody has a clean answer to.

Character.ai raised a $43 million seed round, then a $150 million Series A in March 2023 at a $1 billion valuation, hitting unicorn status less than two years after founding. Andreessen Horowitz led it.

That is the whole of its disclosed venture funding, about $193 million.

The real money story is the Google deal. In August 2024 Google took a non-exclusive licence to Character.ai's models and rehired the founders, paying roughly $2.7 billion.

Part of that went to buying out early investors at a valuation of about $2.5 billion. It was not new funding for the company.

Character.ai kept its independence on paper but lost its founding team. In 2025 it was reported to be weighing a sale or a new raise at a valuation above $1 billion, well below the Google-deal mark.