Google's Discover job posting reveals the four building blocks of its recommendation engine
Optimixed.com
Search Engine Land
72
Analysis
TLDR
Search Engine Land's two-year study of Discover feeds validates a Google job posting's mention of retrieval, prediction, ranking, and embedding as core recommendation-system components, revealing reader affinity as the dominant amplification lever.
Search Engine Land analyzed 42 million Discover cards over two years and matched three of four technical terms from a Google Staff Software Engineer job posting—retrieval, prediction, ranking, and embedding—to observable layers in the feed's behavior. The outlet found that reader affinity learned by the model drives roughly 8x amplification at equal topic potential, far outweighing explicit signals like the Follow button, and that attention and engagement scores operate as nearly independent dimensions, suggesting publishers should optimize for both separately.
Was this useful?