Module 13: User Behavior Signals Patents
111 patents across CTR, user engagement, dwell time, personalization, intent modeling, behavioral ranking, A/B testing, and user satisfaction.
Overview
User behavior is one of the most important and most misunderstood ranking factor categories. These 111 patents reveal exactly how Google measures user signals, what it considers positive vs. negative, and critically — how it protects these signals from manipulation.
The 8 Categories of User Behavior Patents
| Category | Patents | Key Signals |
|---|---|---|
| CTR / Click Data | 20 | Click-through rate, click fraud detection |
| Engagement Metrics | 14 | Time on page, scroll depth, interactions |
| Dwell Time / Session | 10 | Long click, short click, pogo-sticking |
| Personalization | 25 | User history, preferences, location |
| Intent Modeling | 18 | Query understanding, intent classification |
| Behavioral Ranking | 12 | Post-click re-ranking, pogo-sticking |
| A/B Testing | 7 | Controlled experiment signals |
| Satisfaction | 5 | User satisfaction measurement |
CTR / Click Data (20 Patents)
US10229166B1 - NavBoost
Year: 2019 (Confirmed 2024 DOJ Trial)
The Most Important User Signal Patent:
NavBoost is Google's system for using click behavior to re-rank search results. Confirmed in the 2024 DOJ antitrust trial.
NavBoost Processing Flow:
Click Fraud Detection Patents
US7657626B1 - Click Fraud DetectionYear: 2007-2010
Detection Methods:
- Abnormal click volumes (statistical outliers)
- Bot-generated click signatures (device fingerprints)
- IP address pattern analysis
- Timing pattern analysis (too regular = bot)
- Geographic anomalies (clicks from unusual locations)
- Real-time fraud scoring
US7917491B1 - Click Fraud PreventionYear: 2007-2011
Additional Methods:
- Cross-platform pattern comparison
- Coordinated fraud campaign identification
- User behavior correlation across multiple touchpoints
- Verification procedure triggers
US20150032533A1 - Real-Time Click ValidationYear: 2015
Real-Time Pipeline:
- Every click validated within milliseconds
- Known bot signatures checked instantly
- Suspicious patterns flagged for review
- Invalid clicks removed from ranking signals
Engagement Metrics (14 Patents)
US10296642B1 - Engagement Scoring
Year: 2017-2019
Engagement Signals Measured:
- Page visit duration
- Scroll depth (how far down the page)
- Internal link clicks (multiple pages per session)
- Return visit frequency
- Query refinement after visiting (vs. not refining = satisfied)
US8626823B2 - Social Sharing as Quality Signal
Year: 2013-2014
Social Engagement:
- Shares on social platforms as quality signal
- Share velocity (how quickly content spreads)
- Authority of sharers (high-follower accounts)
- Cross-platform sharing patterns
Note: Social signals are correlative, not directly causal in ranking.
Dwell Time / Session Patents (10 Patents)
US9558233B1 - Selection Quality Score
Year: 2015
The Core Dwell Time Patent:
Selection Quality = f(time_between_click_and_return)
Long click: User stayed >2-3 minutes → Satisfied → Positive signal
Short click: User returned in <15-30 seconds → Unsatisfied → Negative signal
Intermediate: 30 seconds to 2 minutes → Neutral (query-dependent)Key Finding: The threshold for "satisfied" varies by query type. A weather query might be satisfied in 5 seconds. A research query might need 10 minutes.
US8255413B2 - Dwell Time Analysis
Year: 2012
Dwell Time Nuances:
- Measured per result, per query
- Aggregated across many users
- Compared to average dwell time for that query
- Relative to position in SERP (position 1 expected to have higher dwell)
US8838587B1 - Session Dwell Duration
Year: 2014
Session-Level Signals:
- Total session length
- Pages visited within session
- Queries reformulated (indicates dissatisfaction)
- Session ended after visiting your page (satisfaction signal)
US20170140049A1 - Session Context for Ranking
Year: 2017
Session Context Influences Results:
- Previous searches in session inform current query
- Pages visited before current search affect result weighting
- Time spent on previous pages in session
- Query refinement patterns show user needs
Personalization Patents (25 Patents)
US10810270B2 - Search Based on Browsing History and Emotional State
Year: 2020
Personalization Signals:
- Prior search history
- Pages visited (from Chrome/logged-in data)
- Inferred interests from behavior patterns
- Emotional context (detected from interaction patterns)
- Location and time-of-day
Key Insight: Your rankings are personalized. What you see is NOT what your customer sees. Use incognito mode for unbiased rank checking.
Intent Modeling (18 Patents)
US20230334045A1 - BERT Intent Understanding
Year: 2023
Intent Signals from BERT:
- Full sentence context, not just keywords
- Implicit intent detection (what user REALLY needs)
- Ambiguous query handling (multiple intent possibilities)
- Follow-up query prediction
EP3005168A1 - Featured Snippet Intent Matching
Year: 2015
Intent-to-Format Mapping:
| Intent | Best Format | Why |
|---|---|---|
| Definition | Paragraph snippet | Quick factual answer |
| Process | Numbered list | Sequential steps |
| Comparison | Table | Side-by-side view |
| List of items | Bulleted list | Scannable format |
| Location | Map + address | Navigation intent |
Behavioral Ranking (12 Patents)
US9092510B1 - Pogo-Sticking Detection
Year: 2013-2015
Pogo-Sticking Definition:
Pogo-sticking = User clicks result → returns to SERP → clicks different result
Signal interpretation:
- Result 1 clicked, immediate return → Result 1 failed to satisfy
- Result 2 clicked, user stays → Result 2 satisfied
- Over many users: Result 2 gets promoted, Result 1 gets demotedUS8117209B1 - Ranking Based on User Behavior
Year: 2007-2012
Behavioral Ranking Signals Combined:
- Click-through rates
- Dwell time on pages
- Navigation patterns
- Return visits
- Query reformulation (suggests dissatisfaction)
- Session depth (more pages = more engaged)
A/B Testing Signals (7 Patents)
US11132700B1 - Direct and Indirect Effect Identification
Year: 2021
Google's A/B Testing on Ranking:
- Google runs continuous experiments on ranking algorithms
- Some results shown to "treatment" group, others to "control"
- Measures both direct (immediate) and indirect (behavioral) effects
- Validates algorithm changes before global rollout
Why This Matters:
- Algorithm updates are tested before launch
- Correlations between quality signals and user satisfaction are verified
- False signals (manipulation) fail to replicate in controlled tests
User Satisfaction (5 Patents)
US8442984B1 - Website Quality via User Behavior
Year: 2011-2013
The "R Metric" for Satisfaction:
R = Searcher Satisfaction Measure
Indicators of satisfaction:
- User does not return to SERP after visiting
- User completes apparent goal (order, contact, answer found)
- User visits multiple pages (exploration = engagement)
- Session ends naturally (vs. frustrated exit)Anti-Manipulation: Why These Signals Are Hard to Game
Statistical Protection Mechanisms
1. Aggregate signals across thousands of users
2. Histogram analysis compares patterns to baseline
3. ML classifiers identify and filter artificial patterns
4. Cross-reference with known manipulation signatures
5. Geographic + temporal analysis flags anomalies
6. Device fingerprint analysis catches bot trafficWhat Doesn't Work:
- Click farms (volume anomalies detected)
- Bot traffic (device fingerprint recognition)
- Coordinated clicking campaigns (pattern recognition)
- Purchased traffic (engagement depth analysis — bought traffic doesn't engage)
- VPN/proxy-based manipulation (IP analysis)
What Does Work:
- Genuinely good content that satisfies user intent
- Clear, immediate value delivery above the fold
- Fast page load (reduces bounces)
- Matching content format to query intent
- Fulfilling the implicit promise of your title/meta description
Practical Improvements
Improve Dwell Time
- Answer immediately — key information above fold
- Progressive disclosure — summary first, depth for those who want it
- Engaging format — headers, bullets, visuals break walls of text
- Internal linking — guide users to related content
- Fast load — every second of delay increases bounce rate
Reduce Pogo-Sticking
- Match search intent exactly — don't bait-and-switch
- Title/meta accuracy — deliver what the snippet promises
- Relevant above-fold content — first screen must match query
- Answer the question — state the answer, then provide depth
Improve CTR
- Compelling title tags — numbers, power words, curiosity
- Descriptive meta descriptions — clear value proposition
- Structured data — star ratings, FAQ, sitelinks
- URL readability — descriptive path segments
Key Patents Referenced
| Patent | Title | Year |
|---|---|---|
| US10229166B1 | NavBoost | 2019 |
| US7657626B1 | Click Fraud Detection | 2007-2010 |
| US9558233B1 | Selection Quality Score | 2015 |
| US9092510B1 | Pogo-Sticking Detection | 2013-2015 |
| US8117209B1 | Ranking Based on User Behavior | 2007-2012 |
| US10810270B2 | Search Based on History + Emotional State | 2020 |
| US11132700B1 | A/B Testing Signals | 2021 |
Next Steps
- Content Quality Module — Panda quality signals
- CTR Audit — Apply these signals to your audit
- Neural AI Search — ML context for these signals