Google's Discover job posting reveals the four building blocks of its recommendation engine

Search Engine Land identified a Google Careers listing for a Staff Software Engineer, Discover Ranking role that explicitly names four core components of Discover's recommendation system: retrieval, prediction, ranking, and embedding. Analysis of real Discover feeds over two years reveals three of those four layers already operating in production—retrieval pipelines that filter eligible content, prediction scores tracking attention and engagement separately, and embedding vectors maintained per user across multiple time windows. The research shows reader affinity learned by the model drives roughly 8x amplification differences between publishers at equal topic potential, outweighing explicit signals like Google's Follow button.