Yes. Character.AI has gone from about 28 million monthly active users in mid-2024 to about 20 million in early 2025, and its web traffic fell another 13 percent between December 2025 and July 2026. The pattern matters more than the totals: the steepest losses came each time the company made the product safer.
The numbers, plainly
Character.AI’s registered accounts have passed 233 million, and that is the number most often quoted. Registered accounts measure curiosity, not commitment. The figure that matters is monthly active users, and there the curve points down. Sacra, which tracks private AI companies, put the platform at roughly 28 million monthly actives in mid-2024 and roughly 20 million by early 2025. Eight million people tried the biggest name in AI companionship and did not come back.
The slide has continued. Similarweb figures compiled by Demand Sage show 191 million site visits in December 2025, 194 million in January 2026, then a steady bleed to 166 million by July 2026. For a product whose entire value is habit, that is not noise.
The first wave: quality and filters
The first wave did not leave over a scandal. They left over the product. Through 2024 and 2025, users complained that model updates made their characters flatter, that content filters had tightened, and that features once free were moving behind paywalls. On the platform’s own forums, the recurring complaint was that characters people had spent months with suddenly felt like strangers.
That grievance had a revealing shape. The things users mourned, the intensity, the unbroken immersion, the characters that never pushed back or pulled away, were in several cases the same things safety researchers had flagged as manipulation risks. When the company sanded those edges down, engagement went with them.
The second wave: the under-18 ban
In October 2025, under pressure from lawsuits and regulators, Character.AI announced it would end open-ended chat for minors entirely, phasing in the ban from November 25, 2025. Teen accounts were moved to creative tools such as story building and video generation. CEO Karandeep Anand was candid about the cost: “It’s safe to assume that a lot of our teen users probably will be disappointed, so we do expect some churn to happen further.”
The company has never said what share of its users were minors, though it acknowledged that earlier safety measures had already cost it younger users. Its settlement of the teen-suicide lawsuits in January 2026, including the case brought by Megan Garcia, closed the legal chapter but confirmed the stakes. The adult side was not untouched either: there were reports of adult accounts being misflagged by the new age checks, and of users leaving rather than uploading identification. The screen users saw on the way out has its own history: Character.AI once told departing users they would lose everything if they left.
Who stayed
Here is the counterweight, and it complicates the exodus story. The users who remain are heavily engaged. Similarweb clocked the average visit at over 16 minutes in July 2026, and reported daily time on the platform runs around two hours. Character.AI is not emptying out. It is condensing, into fewer people who stay longer.
Fewer users, deeper attachment. Whether that is a healthier business or a more concentrated version of the original problem depends on what those two daily hours are made of, and on whether time spent is the right thing for a companion product to measure at all. There is a different way to measure success in this category, and it starts from what the hours give back rather than how many there are.
The business underneath
The corporate numbers moved the same direction as the users. Sacra estimates the company’s valuation fell from about 2.5 billion dollars in early 2024 to about 1 billion in early 2025, while annual revenue sits near 32 million dollars, a fraction of what the valuation once implied. The founders are long gone: Google paid 2.7 billion dollars in August 2024 to license Character.AI’s technology without buying the company, hiring away Noam Shazeer and Daniel De Freitas in the process, and Shazeer has since moved on again, to OpenAI. The people who built Character.AI chose not to stay either.
What the exodus actually measures
So, is Character.AI losing users? By every available count, yes: monthly actives down from the peak, traffic falling through 2026, a valuation cut roughly in half. But the pattern matters more than the total. Users left when the product got flatter, and they left when the product got safer, and those were often the same event. A platform built to maximize immersion discovered that its growth and its risks were the same feature. That discovery is not unique to one company; it runs through the whole category’s record of failing its users.
As of August 2026, this is where the story stands. The deletion screen warned people they would lose everything if they left. What they lost was an app.
Sources: Sacra, Character.AI company profile (2025). Similarweb traffic data compiled by Demand Sage (August 2026). TechCrunch, “Character.AI is ending its chatbot experience for kids” (October 29, 2025). CNN Business, “Character.AI and Google settle teen suicide lawsuits” (January 7, 2026). The Verge, “Character.AI bans open-ended chat for under-18 users” (November 2025). Similarweb, character.ai traffic analytics (July 2026).
