Link Analysis Intelligence Vintage
Link analysis has been the backbone of Google's ranking system since PageRank was introduced in 1998. Bill Slawski's patent analyses reveal how link evaluation evolved from a simple vote-counting system into a sophisticated analysis of link context, trust, history, and user behavior.
Link Evaluation Framework
The Reasonable Surfer Model (Detailed)
The Reasonable Surfer patent describes how a hypothetical "reasonable" user would interact with links on a page. Links that a reasonable user is more likely to click receive more weight.
Link Value by Page Zone
User Data Integration
The patent also describes using actual user behavior data to calibrate the Reasonable Surfer model:
- Click probability — Which links on a page do users actually click?
- Mouse hover patterns — Where do users hover their cursor?
- Scroll depth — How far down the page do users go?
- Device interaction — Mobile vs. desktop click patterns differ
Font and Visual Properties
The patent mentions that visual properties of links affect their value:
| Property | Higher Value | Lower Value |
|---|---|---|
| Font size | Larger than body text | Smaller than body text |
| Color | Distinct from surrounding text | Same as surrounding text |
| Position | Above the fold | Below the fold |
| Formatting | Bold, highlighted | Unformatted |
Source: Google's Reasonable Surfer (2010)
Anchor Text Intelligence
Not All Anchor Text Is Equal
Bill's analysis showed that anchor text value depends on multiple factors beyond the text itself:
Co-Citation as an Alternative Signal
Bill explored how co-citation provides a link-like signal without requiring an actual hyperlink:
Co-citation scenario: Three news articles all discuss "Topic X" and mention both "Site A" and "Site B" in the same context, even without linking to either. Google can infer that Site A and Site B are related to Topic X and to each other.
This creates a form of "implied link" that does not require the cooperation of the linking page.
Source: Not All Anchor Text is Equal and Other Co-Citation Observations (2012)
Trust and Authority Propagation
TrustRank Model
Trust Signals
| Signal | Trust Indicator |
|---|---|
| Distance from seed set | Fewer hops = higher trust |
| Link source diversity | Trust from many independent sources > one source |
| Topical consistency | Links from same-topic pages carry more trust |
| Social endorsement | Links from trusted social profiles add trust |
| Temporal consistency | Long-standing links from stable sources indicate trust |
Reciprocal Links: Patent Truth
Bill debunked common myths about reciprocal links by analyzing what patents actually say.
Reciprocal Link Evaluation Matrix
| Pattern | Natural or Suspicious | Patent Treatment |
|---|---|---|
| Two related blogs cross-linking | Natural | Normal link value |
| Business + its blog on different domain | Natural | Normal link value |
| Complementary businesses linking | Natural | Normal link value |
| Blog roll with 50+ mutual links | Context-dependent | May be discounted |
| Automated link exchange at scale | Suspicious | Likely devalued |
| Simultaneous reciprocal link creation | Suspicious | Historical data flag |
The key insight: reciprocal linking is only problematic when it occurs at unnatural scale or in patterns that suggest automation rather than genuine editorial choice.
Source: What are Reciprocal Links? (2009)
Link Spam Detection
Patents describe multiple methods for identifying manipulative linking:
Spam Detection Methods
Manipulative Article Detection
From the 2007 webspam patent, Google can:
- Identify articles created primarily to carry links
- Group manipulative articles to detect link schemes
- Adjust PageRank calculations for links from detected schemes
- Remove pages from the index or reduce their rankings
Source: Google on Webspam, Doorway Pages, and Manipulative Articles (2007)
Key Takeaways
- Link position on the page is a major factor — Editorial links in main content are worth dramatically more than boilerplate links.
- Anchor text must be natural — Descriptive, varied anchor text signals editorial choice. Uniform exact-match anchors signal manipulation.
- Co-citation creates implied relationships — Being mentioned alongside authoritative sources builds relevance without requiring direct links.
- Trust flows from verified sources — The closer your link profile is to trusted seed sites, the more trust you receive.
- Reciprocal links are context-dependent — Natural reciprocal linking between related sites is fine. Scaled automation is not.
- Link spam detection is multi-dimensional — Google uses pattern, temporal, network, and content analysis simultaneously.