By the end of 2023, Mark Zuckerberg expects 30%+ of Instagram and Facebook content recommendations will be served by Meta's AI, double the ~15% on Facebook now
Meta’s stated target marked an early shift from social graphs toward AI-selected distribution across Facebook and Instagram. Later coverage shows the program becoming more explicit: Meta published system cards for recommendation AI across Feed, Stories and Reels, while reporting measurable engagement gains from AI-driven feeds and video recommendations.
First-order effects
Facebook and Instagram creators face a larger share of distribution being determined by Meta’s recommendation models rather than existing follower relationships.
Meta gains a common AI-ranking agenda across its major consumer surfaces, with Feed, Stories and Reels all within the recommendation program described in later coverage.
Second-order effects
Creators and publishers must optimize for recommendation eligibility and watch-time performance as Meta expands AI-selected inventory, especially in short-form video where its newer architecture was tied to Reels watch-time gains.
Higher engagement from AI-ranked feeds strengthens Meta’s ability to commercialize the attention those feeds generate; the company later sought to extend AI from recommendations into brand ad creation and targeting.
Third-order effects
If Meta continues to improve engagement through ranking models, social platforms’ competitive advantage shifts further from the size of a user’s follower graph to the quality and scale of their recommendation systems.
The same user-data feedback loop that improves recommendations also raises governance pressure: Meta’s stated next AI step involved learning from a dataset larger than Common Crawl, a development linked in coverage to privacy concerns.
The trend: Social platforms are becoming AI-distribution systems, using behavioral data and recommendation models to allocate both audience attention and advertising value.
On Meta earnings call, Mark Zuckerberg says AI-driven recommendations from accounts you don't follow is roughly 15% in Facebook and a “little more” in Instagram. He says that percentage will more than double by the end of next year. Huge departure from the social graph era.
What strikes me about all of these Insta/FB quotes from Zuck/Mosseri is that they seem to attribute the growth in video to user desire or some phantom phenomenon when, in fact, *their choices* and algorithms are the sole things driving those numbers higher. https://twitter.com/..…
Funny, it feels a lot more than 15%. I think it's because I find I resent the AI recommended stuff in my feed far more than the ads. At least the advertisers have paid https://twitter.com/...
Interesting comment from Mark Zuckerberg just now on Meta's earnings call: Roughly 15% of content in FB feed, and a little more than that in the IG feed, are from accounts that people don't follow. He expects those numbers to double...
As long as you engage, Zuck doesn't care whether the content is from someone you follow or someone you don't. Heck, it doesn't even have to be from someone at all. This time next year, they'll be hosting their own bots spamming your timeline to claim “bot reporting engagement.” h…
One primary rationale that Meta used to give for poor integrity experiences was that the feed was just content from entities you chose to follow. You had the control so it was essentially your problem. That justification doesn't work anymore. https://twitter.com/...
Also the AI is just sometimes starkly wrong. I get a lot of formula one reels right now, even though the sport bores me to tears. Not sure how its come to that conclusion, except maybe as I follow a few football accounts it makes a misguided inference