People also ask
In practice
These scores are pattern classifiers: trained on correlations between flagged sites and their characteristics — thin content ratios, TLD distributions, anchor patterns, ownership fingerprints — they estimate resemblance to known spam, not actual harm. Their honest use is triage: screening hundreds of candidate publishers cheaply (a high score shortens the queue for manual review) and surfacing anomalies in a monitored profile (a cluster of high-score domains appearing overnight is investigable).
Their dishonest use is automation: “disavow everything above 30%” is how useful links die — the scores flag legitimate sites in niches with spammy neighbours, small sites with natural profiles, and anyone linked from the wrong neighbourhood once. The verification that matters after a flag: open the site (is it a real publication or a storefront), read its outbound links (commercial anchors sprayed everywhere?), check traffic against content (audience or shell), and place the flagged site in context (one spam neighbour is noise; a network is a pattern).
The same scepticism applies to “toxic backlink” reports from link-audit tools: their lists are lead generation for disavow services as often as they are risk analysis — the profile-level judgement (patterns, not pages) belongs to a person who has seen what natural distributions look like. The standing rule in a quality programme: metrics filter, humans judge — and the disavow file is written rarely, surgically, with documentation.
See also: Toxic links and when to disavow, Site risk at scale, Link neighbourhood.
Related service: Link insertions.
