Google Discover's SDF filters stale articles before ranking — stale reports down 54.9% in two years
gaganghotra_ · x · 2026-10-24
1492.vision compares a Google CIKM 2026 paper with its own Discover data. The SDF system, "fully deployed in Google Discover," runs two pre-ranking filters: a supersession model removes articles made obsolete by newer coverage, and a lifecycle model removes articles once they've received 90% of predicted lifetime traffic. Stale-content user reports fell 54.9% over two years.
Their data shows leaks remain (a Michigan-UConn pre-game card served 15 hours after the final buzzer), article audiences drop after day three, and per-article peak audience for the same publisher fell about a third between May–June and August–September (0.68 in English feeds). Since SDF was live by the start of the year, it alone doesn't explain the reach drop recorded since July — but it shows the shape: articles keep their best day and lose their end of life.
More from Venture
- X won't pay creators for viral tweets posted before monetization approval — doooyle · 2026-10-24
- Jensen Huang: Nvidia software licensing could add ~$20B/year at $2,000 per enterprise GPU — sudoraohacker · 2026-10-24
- Neocloud premium is a speed premium: why Microsoft pays CoreWeave despite 5% borrowing costs — AccBalanced · 2026-10-24
- On-demand GPU rental indices hit all-time highs as even A100 prices rebound — AccBalanced · 2026-10-24
- Meta acquires analytics startup Sundial; founders join Meta — kevinakwok · 2026-10-24
- Tesla Energy Q3: deployments up 10% but revenue down 16% as prices collapse — pdamodaran · 2026-10-24