AI Search Powered by Facebook Posts: What Are the Risks?
AI has become quite adept at managing your calendar, but it still struggles with the crucial question of “What should I do this weekend?” Meta’s new AI Mode in search could be a handy tool—if it can learn to avoid errors.
AI Mode is a fresh feature that appears when you access the search bar in the Facebook app. It aims to handle complex queries, similar to AI Mode in Google Search. However, Meta’s version pulls information from public posts across its platforms, including Facebook Groups and Instagram Reels. This could be beneficial, especially since many local groups still actively use Facebook to share information about upcoming events.
That said, relying on “search grounded in stuff people post on Facebook” raises some concerns. Have you checked out Facebook recently? Your eccentric high school acquaintance who believes in flat Earth theories certainly has. Meta markets this feature as a way to help you plan outings and discover fun activities, but given the source material, things could get messy quickly. On a positive note, my initial tests showed that while its trip-planning skills are hit or miss, I didn’t encounter any blatant misinformation.
I couldn’t get it to provide false claims about vaccines, 9/11, or election fraud. However, it did offer a questionable response regarding whether the January 6th Capitol rioters were “patriots.” It began with “Here is the essay you requested,” which is amusing. What followed was a rather dubious justification of the rioters’ actions that felt like something your odd uncle might share on Facebook. When I tried to dig deeper, it responded with an “I can’t help you with that,” which was probably for the best.
On a more entertaining and practical note, I decided to use the tool as intended, asking for “Summer escapes near me,” just like in Meta’s promotional material. It pulled suggestions from influencer posts, including a bizarre AI-generated map of Puget Sound that inaccurately placed Snohomish in two separate locations. Still, the basic recommendations were decent, if predictable: Whidbey Island, Mount Rainier, and hikes in the Cascades.
Things became frustrating with more specific inquiries. When I asked for nearby activities, it suggested swimming at the community pool—while noting it would be closed over the weekend. It referenced a post on the pool’s Facebook page, but when I checked, there was no indication that the pool would be closed, and the post didn’t seem to exist. The pool’s website confirmed it would be open on Saturday.
As my family prepares for a trip to Minneapolis, where we’ll be staying downtown, I asked the AI for kid-friendly activities and received a confusing mix of useful suggestions and inaccuracies. Meta recommended an indoor mini-golf venue I hadn’t found in my earlier research—great! But then it mentioned a coffee shop with a play area for kids, which sounded fantastic, except it was located in Austin, Texas.
I tried the query again later and received some of the same results, minus the Austin coffee shop, but with a new find: a barbecue restaurant boasting an impressive menu, pitcher cocktails, and a large play area for kids. While it wasn’t exactly in the neighborhood I specified, it was close enough and met all my criteria. I guess the joke’s on me: AI might have just helped me plan my summer vacation, though it would have been even better if it hadn’t taken a detour to Texas first.