RL Quietly Rewrites Online Ads

In a 10‑week A/B test spanning 35,000 advertisers and 640,000 ad variants, Meta’s RL‑trained AdLlama increased click‑through rates by 6.7% vs. a supervised model. Reinforcement learning is now steering billions of impressions toward more engaging content. — Measured gains in attention optimization raise stakes for antitrust, consumer protection, and political ad policy as platform AI shapes what people see.