GOOGLE DISCOVER ARCHITECTURE: CLUSTERS, CLASSIFIERS, OG TAGS, NAIADES - WHAT SDK TELEMETRY REVEALS

Summary from the TLDR Marketing newsletter

Google Discover reaches hundreds of millions daily, but its ranking system is mostly hidden. Content flows through a nine-stage pipeline. Missing og:image or images under 1,200px block or limit visibility, and dismissed items remain permanently suppressed. Collection-level filters can block an entire publisher before ranking. Discover uses 13 cluster types, multiple personalization layers, and about 150 concurrent A/B experiments with 51 feature flags, so similar users can see very different feeds.