Meta AI celebrity chatbots have been exactly as popular as you'd expect

The way some companies with big AI dreams are thinking, smartphone users want nothing more than to get advice from, be entertained by, and interact with virtual chatbots. Meta is betting so big on this concept that it just launched a major effort to let people design custom AI chatbots, tailored precisely to their preferences. But as firms like Meta try to zero-in on the kind of AI interactions that are most engaging, they’re also picking up a lot of lessons about what doesn’t work. And as Meta apparently learned the hard way, that includes chatbots based on celebrities.

Meta introduced its celeb chatbots last September with a total of 28 accounts, all featuring the likeness of a famous person and given specific personas. The whole thing was a bit weird, not even using actual celebrity names: you might interact with “Lorena” the travel expert, based on Padma Lakshmi, or chat with Charli D’Amelio’s “Coco” the dancer. At least those track tonally with the people they’re based on, but others felt like wild swings: Paris Hilton as “Amber” the detective “for solving whodunnits,” or Snoop Dogg not as the resident cannabis sommelier but “Dungeon Master,” ready to help plan your next tabletop adventure.

Fast-forward ten months, and Meta is taking these chatbots back behind the shed for some Old Yeller action. The Information reports that the Facebook and Instagram pages for all these bots went offline earlier this week. The company confirmed to the site it had discontinued the feature, highlighting what it picked up from the experience, explaining, “We took a lot of learnings from building them and Meta AI to understand how people can use AIs to connect and create in unique ways.”

Based on the timing, we’d certainly hope that many of those lessons became the foundation for Meta’s new AI Studio offering, where rather than choosing from all these pre-packaged disparate personas, users can take the time to craft a custom experience. Maybe that one won’t last, either, but sometimes you have to learn what doesn’t work before you can figure out what does.