Per query, ranks search results by combined site- and content-quality signals. A core Upstill ranking primitive feeding the HCU-era site-quality lineage.
Patent Overview
- Inventor
- Sundeep Tirumalareddy, Trystan G. Upstill
- Assignee
- Google LLC
- Filed
- 2013
- Granted
- 2016-09-27
The Challenge
The Challenge
Per query, ranking must combine site-level and content-level quality signals coherently. Site-level signals (authority, trust, category position) and content-level signals (relevance, depth, originality) each contribute differently per query.
- Site-Level Signals Alone Underrank Content — Per query, good content on a moderate-authority site can underperform if only site signals weigh.
- Content-Level Alone Misses Trust — Per query, strong content on a low-trust site can over-rank if only content signals weigh.
- Combination Must Be Query-Aware — Per query type, site/content weighting differs.
- Per-Resource Combined Score — Per resource, combined score must reflect both dimensions.
- Calibration Against Held-Out Data — Per query type, calibration validates combination weights.
Innovation
How The System Works
The system computes per-resource site- and content-quality signals separately, applies per-query combination weights, produces composite score per (query, resource), ranks candidates by score, and calibrates against held-out data.
- Compute Site Signals — Per resource, site-level signals computed.
- Compute Content Signals — Per resource, content-level signals computed.
- Classify Query Type — Per query, classify type to drive weighting.
- Apply Per-Query Weights — Per (query type), site/content weights applied.
- Compute Composite Score — Per (query, resource), composite score produced.
- Rank By Composite — Candidates ranked by composite.
- Calibrate Continuously — Weights calibrate against held-out data.
Site Plus Content Combination
The patent's load-bearing idea is that site- and content-quality signals combine per query. Neither dimension alone suffices; per-query combination weights drive composite ranking.
Per-Query Weighting
Per query type, the right mix of site and content weights emerges. Per-query weighting beats uniform weighting.
- Site-Level Signals — Per resource, site-level signals computed independently.
- Content-Level Signals — Per resource, content-level signals computed independently.
- Per-Query Combination — Per query type, weights tune the combination.
Technical Foundation
Technical Foundation
The patent specifies the site-signal computer, content-signal computer, query-type classifier, weight applier, composite scorer, ranker, and calibration loop.
- Site-Signal Computer — Per resource, computes site-level signals.
- Content-Signal Computer — Per resource, computes content-level signals.
- Query-Type Classifier — Per query, classifies type.
- Weight Applier — Per query type, applies combination weights.
- Composite Scorer — Per (query, resource), produces composite score.
- Ranker — Composite score sorts candidates.
The Process
The Process
Per query, ranking pipeline runs in real time.
- Pre-Computed Signals — Site and content signals pre-computed at indexing.
- Receive Query — Query arrives.
- Classify Type — Query type classified.
- Apply Weights — Per-query weights applied.
- Compute Composite — Per candidate, composite scored.
- Rank — Candidates ranked.
- Return Results — Top results returned.
Quality Control
Quality Control
Combination weights determine ranking quality. The patent specifies safeguards.
- Per-Query-Type Calibration — Per query type, weights calibrated against labeled relevance.
- Signal-Validation — Per signal, validation against ground truth.
- Composite-Score Bounds — Per resource, composite score bounded.
- Engagement Validation — Per query type, engagement signals validate ranking quality.
- Continuous Recalibration — Weights refresh against fresh data.
Real-World Application
Combined site/content ranking is foundational for modern Google search. The per-query-type weighting pattern underpins how Google balances trust against content quality.
- Two-dimensional Signal Sources — Site and content signals combine.
- Per-query-type Weighting Granularity — Weights tune per query type.
- Composite Final Score — Composite score sorts results.
Why Strong Site Plus Strong Content Wins
Composite scoring rewards both dimensions. Pages strong on one dimension but weak on the other underperform pages strong on both.
Why Trust And Quality Compound
Per query type, trust (site signal) and quality (content signal) compound favorably under combination weighting. Investing in both compounds.
<\/section>What This Means for SEO
What This Means for SEO
Site-level and content-level quality signals are computed separately and combined with per-query-type weights into a composite score. SEO implication: you need both a trusted site and strong content, because excelling on only one dimension underperforms excelling on both.
- Strong Site Plus Strong Content Wins — Composite scoring rewards both dimensions, and per-query weighting combines them. Great content on a weak site or a strong site with thin content both underperform balanced excellence. Invest across both.
- Site-Level Signals Are A Foundation — Site authority, trust, and category position feed the site half of the score. A strong site lifts its content's composite score, so building site-wide trust pays dividends across every page you publish.
- Content Quality Cannot Be Skipped — Content relevance, depth, and originality feed the content half. Riding site authority with weak content still underperforms, because the combination weights content too. Each page must earn its content score.
- Weighting Is Query-Aware — Site and content weights shift by query type. Some queries lean on trust, others on content depth. Understand which dominates for your target queries and strengthen the dimension that matters most there.
- Trust And Quality Compound — Trust (site signal) and quality (content signal) compound favorably under combination weighting. Investing in both produces a multiplicative advantage, not just additive, so neglecting either caps your ceiling.
- Low-Trust Sites Cap Good Content — Strong content on a low-trust site is held back by the site signal. If trust is your weak point, fixing site-level signals can unlock the value already present in your content.
- Calibration Validates The Balance — Combination weights are calibrated against held-out data per query type. Because the balance is tuned to real outcomes, broadly investing in both site trust and content quality is the robust strategy rather than betting on one lever.