AI-assisted reread of Yandex leak: behavioral signals outnumber content signals 3 to 1
tracyingram · x · 2026-09-21
Ryan Jones re-examined the 2022 Yandex search engine code leak with modern AI tools and shared key takeaways:
- Behavioral signals outnumber content signals 3 to 1 (counting links as content). The single biggest ranking factor is whether users see you in the results and click; everything else just earns that opportunity.
- Ranking happens in four main stages: the first two are what SEOs focus on; the other layers are behavioral signals, ML re-ranking, personalization, and layout.
- Stage 1 is simple: mostly BM25 and cosine similarity over the inverted index. Query word coverage in title and snippet is heavily weighted — keep titles short so query terms carry high weight, and avoid all caps, which is down-weighted.
The leak remains the most direct first-hand material for studying search ranking mechanics, and AI-assisted code reading makes it far easier to mine.
More from Venture
- Smaller LLMs Aren't Always Cheaper: Review and Rework Costs Blow Up the Bill — zeuslac · 2026-09-21
- OpenAI Pays $300M for 37-Person Camera Startup GlassImaging to Power Its AI Hardware Push — 快鲤鱼 · 2026-09-21
- Brain-Computer Interface Firm Shuli Innovation Starts STAR Market IPO Coaching, Devices Priced at $270K Each — 创业邦 · 2026-09-21
- $40M Bet on Calibrated Confidence, Yet TypeSafe Publishes No Calibration Data — prakersh · 2026-09-21
- Indie maker claims Outrank grew SuperX impressions from 2k to 23k per day — tibo_maker · 2026-09-21
- China's top AI labs raised ~$35B in five months; combined haul could hit $60B by 2027 — FinanceYF5 · 2026-09-21