Per-user and per-query site preference signals. Captures sitewide trust and brand-preference behaviors and feeds them into ranking — the patent that operationalizes 'sites users like to return to'.
Patent Overview
- Inventor
- Paul Haahr, others
- Assignee
- Google LLC
- Filed
- 2009
- Granted
- 2018-07-17
The Challenge
The Challenge
Users have site preferences. They return to sites they trust; they avoid sites that disappoint. These preferences are per-user and per-query, and they're a strong ranking signal — but operationalizing them requires careful aggregation that respects privacy and resists manipulation.
- Per-User Preferences Carry Signal — Each user has preferences. Surfacing preferred sites improves their experience materially.
- Aggregate Preferences Reveal Trust — Aggregated across users, site preferences reveal trust patterns. Sites users return to are trusted sites.
- Per-Query Variation Matters — A user might prefer NYT for news, Amazon for products. Preferences vary by query type.
- Manipulation Defense Required — If 'preferred sites' boosts ranking, sites will manipulate the signal. Detection and aggregation strategy must resist.
- Privacy Must Be Preserved — Per-user preferences are personal. Aggregation must respect privacy boundaries.
Innovation
How The System Works
The system tracks per-user site interactions (visits, returns, dwell), aggregates into per-user preference profiles, layers in per-query type modulation, applies privacy-preserving aggregation, and feeds preference signals into ranking.
- Track Per-User Site Interactions — Per user, track site visits, returns, dwell time. Privacy-respecting telemetry only.
- Build Per-User Preference Profile — Per user, aggregate interactions into preference profile. Sites with strong return patterns rank highest.
- Per-Query Type Modulation — Preferences modulated by query type. Different sites preferred for different query categories.
- Aggregate Across Users — Aggregate preferences across user pool. Reveals general trust patterns at site level.
- Apply In Ranking — Per-user preferences and aggregate trust signal feed ranking. Preferred sites earn boost for the user.
- Privacy Preservation — Per-user signals stay user-private. Aggregations apply differential-privacy or comparable safeguards.
- Detect Manipulation — Pattern analysis flags suspicious aggregate-preference patterns. Manipulation filtered.
Site Preferences Are Trust Signals
The patent's load-bearing idea is that user site preferences — measured by return behavior — are a strong, structurally meaningful trust signal. Per-user and aggregate, they reveal which sites users actually value.
Return Behavior Is The Truth Signal
Users return to sites they value and avoid sites they don't. Return behavior is harder to fake than rating or review systems. The signal is structural.
- Per-User Tracking — Per user, site visits, returns, dwell tracked with privacy preservation.
- Per-Query Modulation — Preferences modulated by query type. Different sites preferred for different intents.
- Aggregate Trust Signal — Aggregate across users reveals general trust patterns. Sites with broad return appeal earn aggregate-trust boost.
Technical Foundation
Technical Foundation
The patent specifies the interaction tracker, per-user profile builder, query-type modulator, aggregator, privacy layer, and manipulation detector.
- Interaction Tracker — Per user, tracks site visits, returns, dwell time. Privacy-respecting telemetry.
- Per-User Profile Builder — Aggregates per-user interactions into preference profile.
- Query-Type Modulator — Per query type, modulates preferences. Different sites preferred for different intents.
- Aggregator — Aggregates per-user profiles across user pool into general trust signal.
- Privacy Layer — Differential-privacy or comparable safeguards. Per-user signals stay user-private.
- Manipulation Detector — Pattern analysis flags suspicious aggregate-preference patterns. Filtered.
The Process
The Process
Tracking runs continuously; per-user profiles update over time; aggregate trust signals update with new data.
- Track Interactions — Per user, site interactions tracked with privacy preservation.
- Update Profile — Per-user preference profile updates as interactions accumulate.
- Aggregate Across Users — Periodically, aggregate trust signal recomputes from user-pool profiles.
- Apply Privacy Layer — Differential-privacy applied to aggregations.
- Receive Query — Query arrives. Per-user and aggregate preferences consulted.
- Apply In Ranking — Preferences feed ranking. Preferred sites earn boost.
- Detect And Filter Manipulation — Aggregate-preference manipulation patterns flagged and filtered.
Quality Control
Quality Control
Preference signals are sensitive to manipulation and privacy. The patent specifies safeguards.
- Privacy Preservation — Per-user signals stay user-private. Aggregations use differential-privacy or comparable safeguards.
- Manipulation Detection — Pattern analysis flags suspicious aggregate-preference patterns. Filtered.
- Diversity Requirement — Aggregations require diverse user-pool support. Single-user or single-network patterns filtered.
- Per-Query-Type Calibration — Per query type, preference weights calibrate against held-out data.
- Continuous Recalibration — Per-signal weights and detection patterns recalibrate against fresh data.
Real-World Application
Site-preference signals underpin per-user personalization plus aggregate brand-trust signals. The pattern of return-behavior-derived trust is a foundational ranking input.
- Per-user Personalization Granularity — Each user has a preference profile. Personalized boosts for preferred sites.
- Per-query-type Modulation — Preferences modulated by query type. Different sites preferred for different intents.
- Aggregate-trust Cross-User Signal — Aggregate across users reveals general trust patterns. Sites with broad return appeal earn boost.
Why Brand Building Wins Long-Term
Aggregate-trust signals reward sites users return to. Building a brand users seek out and return to compounds as a ranking signal — far more durable than per-query optimization.
Why First Impressions Matter Structurally
If users don't return after first visit, preference signal degrades. Site-quality, UX, and content-fit on first visit directly shape long-term preference accumulation.
<\/section>What This Means for SEO
What This Means for SEO
This patent operationalizes site preference from return behavior, both per-user and aggregated into a brand-trust signal, modulated by query type. SEO implication: building a brand users seek out and return to compounds as a durable ranking signal that per-query optimization cannot match.
- Return Behavior Is The Trust Signal — Users return to sites they value and avoid ones that disappoint, and return behavior is harder to fake than ratings. Earning repeat visits is a structural trust signal, so optimize for users wanting to come back.
- Brand Building Wins Long-Term — Aggregate-trust signals reward sites with broad return appeal. Investing in a recognizable brand users seek out compounds over time and is far more durable than chasing individual query rankings.
- First Impressions Shape Long-Term Preference — If users do not return after a first visit, preference signal degrades. Site quality, UX, and content fit on that first visit directly determine whether preference accumulates or erodes.
- Preference Varies By Query Type — Users prefer different sites for news versus shopping versus reference. Being the trusted destination for your specific query category matters more than broad, undifferentiated reputation.
- Manipulated Preference Is Filtered — Suspicious aggregate-preference patterns and single-network signals are flagged and removed, and diverse user-pool support is required. You cannot fabricate return behavior; it must come from a broad, genuine audience.
- Personalized Boosts Reward Loyal Users — Per-user profiles boost sites that individual returns favor. Serving your audience well enough that they repeatedly choose you earns personalized lift with exactly the users you want.
- Privacy-Preserving Aggregation Means Breadth Matters — Aggregations apply privacy safeguards and need diverse support, so broad genuine appeal beats narrow intensity. Widening the base of users who return to you strengthens the aggregate trust signal.