Google's KScorer uses 56 annotated signals to decide what your search really means
VeryWellVersed · x · 2026-10-28
- RESONEO details how Google Search's internal KScorer (Knowledge Scorer) scores every possible interpretation of a query, ranks them, and drops the weakest — the winning interpretation decides whether you see a direct answer, knowledge panel, local pack, or Assistant action.
- Internal Google schemas list all 56 signals, each annotated by Google's engineers with its computation, formula, family, and pipeline stage; the 2024 Content Warehouse leak named only four, with no definitions.
- Three key findings: clicks predict the quality score human raters would give; the top 10 web results influence intent disambiguation (e.g. "jaguar speed" — animal vs car), with the #1 result weighing 10x the #10; context (distance, session, user) tips the balance.
- Signals span 12 families from clicks (UIS/AUIS) to geography (GEO); strongest families include UIS, REFRES, SALIENTTERMS, and ORBIT.
More from Apps
- AI Music Video Show ends its X livestream and moves to Roku as a TV channel — charis_ai · 2026-10-28
- Claude now works inside Google Docs, Sheets and Slides — setup guide and prompts — rubenhassid · 2026-10-28
- Gemini Notebook reportedly adding Connectors, with Google Drive as the only launch integration — testingcatalog · 2026-10-28
- Meta quietly launches Waddle World, an AI pet journal app with chatting pixel pets — The Verge AI · 2026-10-28
- Dev builds a 1975 yearbook app with ChatGPT Sites to nail era-perfect AI hair — craigsdennis · 2026-10-28
- A 7-point checklist for creating Obsidian notes, mostly automated via templates — dSebastien · 2026-10-28