Clones didn't catch on: Meta quietly buried its AI celebrities

Imagine you decided to spend tens of millions of dollars so your users could correspond with an AI version of Snoop Dogg that teaches them how to play dungeons and dragons. It sounds like a script from a satirical Silicon Valley show, but for Mark Zuckerberg it was a reality just a year ago. However, Meta's ambitious experiment to introduce celebrity "personas" into its messengers ended with a quiet admission of defeat. The company deleted all AI character accounts based on real people from Instagram, WhatsApp, and Messenger, effectively acknowledging that the project had failed.

When Meta announced this project at the Connect conference in 2023, the industry saw it as an attempt to capture the market for entertainment chatbots, where the startup Character.ai was already dominating. Zuckerberg hired a whole lineup of celebrities — from Kendall Jenner and Naomi Osaka to Tom Brady and Paris Hilton. Each of them allegedly received multi-million dollar fees for the right to use their appearance and voice to train models. It seemed that an army of these celebrities' fans would provide Meta with explosive growth in messenger activity. But reality turned out to be far more mundane than marketing plans and polished presentations.

The main problem was that these bots were frankly boring. Despite the faces of world stars, behind them lay a rather limited Llama 2 model at the time, wrapped in rigid corporate safety filters. Instead of lively and cheeky communication, users received sterile, predictable responses that only remotely resembled the speech patterns of the originals. Moreover, Meta inexplicably gave them fictional names. Snoop Dogg became "Dungeon Master," and Kendall Jenner became "Billie." This created an extra cognitive barrier: users had to get used to new character names that looked like their idols but behaved like polite consultant robots from a bank's tech support.

The quiet closure of the project speaks volumes. First, it's an acknowledgment that "stardust" doesn't work in the world of generative AI as effectively as it does in traditional advertising for sneakers or perfume. People come to chatbots for utility or for genuine, albeit simulated, empathy. Meta's project provided neither. Users quickly saw through the fake and returned to regular chats. Second, it indicates a sharp shift in strategy within the company. Now Meta is betting on AI Studio — a platform where any content creator or regular user can build their own bot. This is a transition from a centralized "we give you ready-made idols" model to a "create what you like yourselves" model.

It's interesting to observe how quickly Meta can cut costs and admit mistakes when it comes to hyped technologies. A similar fate previously befell NFTs on Instagram and many ambitious projects within the "metaverse." Now digital clones have joined that list. This is an important lesson for the entire industry: even if you have access to top-tier celebrities, unlimited server power, and the best engineers in the world, you cannot force people to communicate with something that doesn't spark genuine interest in them. Technology for technology's sake has once again lost to common sense and real audience needs.

What does this mean for the future of AI in social media? Meta is clearly trying to find a format in which algorithms become an organic part of communication rather than just a sticker on the interface. Now the company believes the future lies in millions of small, highly specialized bots created by users themselves, rather than a dozen expensive digital mannequins.

This is a more democratic but also more chaotic path. The success of competitors like Character.ai has shown that users care more about context, the ability to roleplay, and the absence of rigid constraints than about an official license from a Hollywood star's agent.

Zuckerberg had to learn this lesson at a cost of tens of millions of dollars.

Bottom line: The era of expensive AI toys from corporations is ending. Users have chosen functionality and flexibility over sponsored celebrity marketing. Will Meta be able to catch up with the market using its bot builder, or has the time already been lost?