Suggestion types and when to use them
Each suggestion has a type (rule code). Filter the queue by type so you fix graph damage before chasing semantic extras. AI embeddings mainly power semantic matches; rules power structural types. Keyword targets come from Opportunities.
Broken target
An existing body link points at a missing, deleted, or unavailable resource. Prefer these after catalog cleanups. AI cosine neighbors can help pick a replacement when embeddings exist.
Orphan inbound
The destination has little or no inbound body links. Suggestions propose sources that should link in. Body-link orphans only—menu links do not count.
Underlinked hub
A product or collection is below peer median inbound. Use this to strengthen money pages without stuffing every blog.
Missing commercial link
A blog or page discusses a topic but does not yet link a related product or collection. High commercial-value weight in confidence.
Semantic match
AI embeddings found similar copy (same locale). Confidence is mostly cosine + context. Best for natural blog-to-product paths when the phrase already exists.
Keyword target
From Opportunities (GSC or manual keywords) mapped to a destination. Requires Link Automation. Still scored; AI can help with destination/anchor.
Content gap
Advisory: propose a new article outline. Not an implementable link and not billed as credits.
