Ranking Algorithm Patents Vintage
Google's ranking algorithm is not a single system but a collection of interlocking patents covering content relevance, link authority, user behavior, freshness, and quality scoring. Bill Slawski analyzed these patents more thoroughly than anyone in the SEO industry.
How Google's Ranking System Evolved (Patent View)
Five Years of Google Ranking Signals
Bill's landmark 2018 article cataloged ranking signals he had identified across five years of patent analysis. This is the most comprehensive patent-derived ranking signal list ever assembled.
Organic Search Signals from Patents
| Signal Category | Examples from Patents |
|---|---|
| Domain Signals | Domain age, rate of linking over time |
| Keyword Usage | Keywords in titles, headings, lists, body text |
| Related Phrases | Context terms, co-occurring phrases, n-gram language models |
| Page Speed | Load time as a ranking factor (confirmed in patents) |
| Watch Times | Time spent on page as a quality signal |
| Language Models | N-gram analysis for content relevance scoring |
Key Insight
Bill demonstrated that Google's ranking signals can be grouped into query-dependent signals (relevance of content to the query) and query-independent signals (overall page/site quality like PageRank). Modern ranking combines both via machine learning.
Source: Five Years of Google Ranking Signals (2018)
PageRank Evolution
The original PageRank patent treated all links equally. Over time, Google filed numerous patents that modified and extended this foundation:
What Changed from Original PageRank
- Not all links are equal — Position on page, surrounding text, link type (navigational vs. editorial) all affect value
- Velocity matters — A sudden spike in links looks different from gradual organic growth
- Trust propagates — Links from trusted sources pass more value than links from untrusted sources
- Context determines relevance — A link from a topically relevant page carries more weight
The Panda Quality Score Patent
In 2011, Bill identified what he called "The Birth of Panda" — a patent filed by Google engineers that described quality scoring using decision trees.
How Quality Scores Work (from the Patent)
The patent described a system that:
- Assigns quality scores to web pages and sites
- Uses decision trees to classify content quality
- Compares pages against templates to detect mad-libs-style keyword insertion
- Identifies cloaking and other deceptive practices
- Measures the ratio between useful content and low-quality content on a site
Panda Patent Key Findings
- Quality scores could be assigned at both page level and site level
- Google used human quality raters to train the decision tree model
- The patent referenced John Lamping's 2005 Berkeley presentation on information quality
- Ratio-based measurement: the patent looks at referral queries pages get optimized for vs. implied links (mentions without actual hyperlinks)
Sources:
- Google's Quality Score Patent: The Birth of Panda (2011)
- Is Ranking Search Results the Panda Patent? (2014)
The Historical Data Patent
One of the earliest and most important patents Bill analyzed (2005), this patent describes how Google uses the history of a document's changes to influence rankings.
Signals from Historical Data
| Historical Signal | How It Works |
|---|---|
| Document Age | When a document was first indexed; newer documents may get a freshness boost for certain queries |
| Content Changes | Frequency and magnitude of content updates — small vs. large changes tracked differently |
| Link Velocity | Rate at which new links appear — sudden spikes may indicate manipulation |
| Anchor Text Changes | Shifts in anchor text profiles over time as a trust signal |
| Link Lifetime | How long links persist — temporary links weighted differently |
| Page Freshness | Whether content is updated regularly or remains stale |
Critical Implication
The Historical Data patent means Google has a long memory. Manipulative link building that creates sudden spikes, or anchor text patterns that shift dramatically, can be detected by comparing current data against historical baselines.
Source: Google and Historical Data (2005)
The Reasonable Surfer Model
This patent fundamentally changed how SEOs should think about link placement. Not all links on a page pass the same value.
Links That Pass More Value (According to the Patent)
- Links in the main content area of a page
- Links with descriptive anchor text related to the target page
- Links that are prominently positioned (higher on the page)
- Links that users are likely to click based on surrounding context
- Links from pages with fewer total outbound links
Links That Pass Less Value
- Footer links and boilerplate navigation
- Links in advertising areas
- Links surrounded by unrelated content
- Links that appear in areas users are unlikely to interact with
- Links using generic anchor text like "click here"
Source: Google's Reasonable Surfer (2010)
Click-Through Feedback and User Behavior
Google patents describe systems that use click data, dwell time, and query sequences to refine rankings.
User Behavior Signals in Patents
Bill identified these user behavior signals across multiple patents:
- Click-through rates — Which results get clicked for a query
- Long clicks vs. short clicks — Time spent on a page after clicking
- Query sequences — What users search next after an initial query (refinement signals)
- Blocked sites — Sites users explicitly block from their results
- Pogo-sticking — Users returning to search results quickly after clicking a result
How User Behavior Feeds Rankings
Source: User Behavior Data Google May Use to Influence Search Rankings (2011)
Reranking: How Google Adjusts Initial Results
Bill documented over 50 ways search engines may rerank initial search results. Key reranking methods from patents include:
- Query-specific reranking — Different ranking factors weighted for different query types
- Location-based reranking — Results adjusted based on searcher location
- Freshness reranking — Recent content boosted for time-sensitive queries
- Diversity reranking — Ensuring variety in search results for ambiguous queries
- Personalization reranking — Adjusting based on user history and preferences
- Data center variation — Different data centers may emphasize different signals
Sources:
- 20 More Ways Search Engines May Rerank Search Results (2007)
- How Google Might Classify Queries Differently at Different Data Centers (2011)
Page Segmentation and Block-Level Analysis
Google does not treat a web page as a single monolithic block of content. Patents describe how pages are segmented into blocks, with different blocks receiving different levels of importance.
Vision-Based Page Segmentation (VIPS)
- Pages are broken into visual segments: header, navigation, main content, sidebar, footer
- The main content block receives the highest weight for ranking purposes
- Boilerplate content (headers, footers, navigation) is identified and weighted differently
- Content blocks are analyzed independently for relevance to a query
Source: Breaking Pages Apart: What Automatic Segmentation of Webpages Might Mean (2009)
Key Takeaways for SEOs
- Ranking is multi-signal — No single factor dominates. Patents show dozens of signals working together.
- Quality is measurable — Google has systematic, algorithmic approaches to measuring content quality, not just human judgment.
- History matters — Your domain's history of content, links, and changes creates a profile that influences future rankings.
- Link context matters more than link quantity — The Reasonable Surfer model means where and how a link appears affects its value.
- User behavior is a feedback loop — Click patterns, dwell time, and query refinements all feed back into ranking adjustments.
- Page structure affects ranking — Where content appears on a page (main content vs. sidebar vs. footer) changes how Google evaluates it.