TrustRank algorithm: seed-set proximity, Anti-Trust Rank, the modern equivalent.
TrustRank is a trust-propagation algorithm proposed by Gyongyi and Garcia-Molina at Stanford and Yahoo Research in 2004. Trust propagates from a manually-vetted seed set through the link graph, attenuating with distance. Anti-Trust Rank is the inverse, propagating spam-mass from a known-spam seed set. Google has not named TrustRank as a product, but TrustRank-equivalent scoring operates inside the broader SpamBrain detection system.
Four moves the trust-propagation surface runs against.
TrustRank propagates trust from a seed set through the link graph. Anti-Trust Rank propagates spam-mass from a known-spam seed set. Google's TrustRank-equivalent operates inside SpamBrain. Prospect vetting against the seed-set proximity proxy is the off-page workflow the campaign runs against.
TrustRank propagates trust from a manually-vetted seed set through the link graph.
TrustRank is a trust-propagation algorithm proposed by Zoltan Gyongyi and Hector Garcia-Molina at Stanford and Yahoo Research in their 2004 VLDB paper. The algorithm starts with a manually-vetted seed set of highly-vetted source domains (the original paper used a small set of well-known authority sites). The trust score then propagates through the link graph: domains close to the seed set in link-graph distance receive a high TrustRank score; domains many hops away receive a lower score. The propagation is dampened so that trust attenuates with distance. The scoring identifies which domains are likely to be trustworthy based on their proximity to known-good sources in the link-graph network.
Anti-Trust Rank is the inverse: spam-mass propagation from a known-spam seed set.
Anti-Trust Rank is the inverse construction proposed in subsequent research: a seed set of known-spam domains is identified, and spam-mass propagates outward through the link graph in the same way TrustRank propagates trust. Domains receiving a high spam-mass score from many hops back to known-spam sources fingerprint as spam-likely even when they look superficially clean. The two scoring directions together (TrustRank from the vetted seed set, Anti-Trust Rank from the spam seed set) identify the domains in the middle territory: high-trust-low-spam clusters are the editorial publisher set, low-trust-high-spam clusters are the network-scheme set, and the boundary between them is where the off-page program scopes prospect vetting.
Google has not named TrustRank as a product but the equivalent scoring operates inside SpamBrain.
Google has not publicly named TrustRank as a scoring product the way it has named PageRank. The pattern of behavior across two decades of ranking observation and the leaked Content Warehouse attributes (May 2024) suggest that TrustRank-equivalent scoring operates inside the broader SpamBrain neural detection system. The implementation differs from the 2004 algorithm in several ways: the seed sets are likely much larger and continuously updated, the propagation is likely integrated with the broader link-graph quality evaluation rather than computed as a standalone metric, and the scoring is integrated with brand-strength signals (siteAuthority and adjacent attributes from the leak). The operational shape of the scoring matches the TrustRank concept.
Prospect vetting against TrustRank-equivalent seed-set proximity is the off-page workflow.
The white-hat off-page program vets prospect domains against a TrustRank-equivalent proxy as part of the campaign workflow. Raw domain-rating (DR or DA or equivalent) is insufficient because it correlates with PageRank flow rather than with trust-propagation distance from known-good seed sets. The proxy combines several signals: link-graph proximity to known-good authority sites in the topical cluster, absence of link-graph proximity to known-spam clusters (the Anti-Trust Rank inverse), brand-strength signals in the leaked-attribute family, behavioral signals confirming the domain as a user-preferred destination. Prospect domains failing the proxy get excluded from the campaign even when their raw DR signal looks favorable.
Methodology questions we get during the audit conversation.
What is the TrustRank algorithm?
TrustRank is a trust-propagation algorithm proposed by Zoltan Gyongyi and Hector Garcia-Molina at Stanford and Yahoo Research in their 2004 VLDB paper. The algorithm starts with a manually-vetted seed set of highly-vetted source domains. The trust score then propagates through the link graph: domains close to the seed set in link-graph distance receive a high TrustRank score; domains many hops away receive a lower score. The propagation is dampened so that trust attenuates with distance. The scoring identifies which domains are likely to be trustworthy based on their proximity to known-good sources.
Does Google use TrustRank?
Google has not publicly named TrustRank as a scoring product the way it has named PageRank. The pattern of behavior across two decades of ranking observation and the leaked Content Warehouse attributes (May 2024) suggest that TrustRank-equivalent scoring operates inside the broader SpamBrain neural detection system. The implementation differs from the 2004 algorithm: the seed sets are likely much larger and continuously updated, the propagation is likely integrated with the broader link-graph quality evaluation, and the scoring is integrated with brand-strength signals (siteAuthority and adjacent attributes). The operational shape of the scoring matches the TrustRank concept even where the implementation differs.
What is Anti-Trust Rank?
Anti-Trust Rank is the inverse construction proposed in subsequent research: a seed set of known-spam domains is identified, and spam-mass propagates outward through the link graph in the same way TrustRank propagates trust. Domains receiving a high spam-mass score from many hops back to known-spam sources fingerprint as spam-likely even when they look superficially clean. The two scoring directions together (TrustRank from the vetted seed set, Anti-Trust Rank from the spam seed set) identify the domains in the middle territory: high-trust-low-spam clusters are the editorial publisher set, low-trust-high-spam clusters are the network-scheme set.
How does TrustRank affect prospect vetting in off-page campaigns?
The white-hat off-page program vets prospect domains against a TrustRank-equivalent proxy as part of the campaign workflow. Raw domain-rating (DR or DA or equivalent) is insufficient because it correlates with PageRank flow rather than with trust-propagation distance from known-good seed sets. The proxy combines link-graph proximity to known-good authority sites in the topical cluster, absence of link-graph proximity to known-spam clusters, brand-strength signals in the leaked-attribute family, and behavioral signals confirming the domain as a user-preferred destination. Prospect domains failing the proxy get excluded from the campaign even when their raw DR signal looks favorable. The trust-propagation axis is part of the standard scope for the off-page SEO services we run.
What's the difference between TrustRank and PageRank?
PageRank measures the equity flowing into a domain through inbound links, computed as a recursive eigenvector calculation across the link graph. TrustRank measures the trust propagating into a domain from a seed set of known-good sources, computed as a damped propagation from the seed set through the same link graph. The two measures are correlated but distinct: a domain can have high PageRank from many inbound links of mixed quality, but low TrustRank if those inbound links do not trace back to the vetted seed set through short paths. The white-hat off-page program treats the two as separate axes and scopes prospect vetting against both rather than treating raw DR as a proxy for either.
How does TrustRank relate to the Content Warehouse leak?
The May 2024 leak surfaced multiple attributes referencing trust-adjacent scoring at the site level, including siteAuthority and brand-name entity-confidence inputs. The leak did not name an attribute called TrustRank specifically. The pattern across the named attributes is consistent with a TrustRank-equivalent scoring layer operating inside the broader ranking system. The scoring contributes to the diffuse brand-strength signal architecture the leak documented rather than as a single weighted signal. The off-page program operates against the same signal architecture by vetting prospects on the trust-propagation axis as part of the campaign workflow.
Raw domain-rating correlates with PageRank flow. Prospect vetting against the TrustRank-equivalent proxy is the white-hat campaign surface.
The audit reads the prospect-list assembly against the trust-propagation proxy, surfaces the placement-set distribution, and scopes the campaign mix that compounds inside the SpamBrain detection surface.