How suggestion confidence is scored
Confidence on a suggestion is a 0–100% display of a deterministic weighted score. AI participates in finding neighbors and writing the Why—it does not assign the percentage. Rule-based rows may show “Rule-based suggestion” in the tooltip when a full breakdown is not attached.
Weights (must sum to 1 before penalties)
- 35% semantic relevance — embedding cosine between source and destination (same locale).
- 15% contextual overlap — shared product type, tags, vendor; blogs/pages pointing at products/collections get a commercial-path boost.
- 15% destination importance — how inbound-rich the destination already is versus peers.
- 15% commercial value — product highest, then collection, then article, then generic page.
- 10% link scarcity — boosts destinations that still need inbound links.
- 10% title mention — destination title already appears in the source body.
Penalties subtract a little when the destination already has many inbound body links (duplicates) or is a low-commercial page.
How this differs from Overview Score
Overview Score is store-wide graph health (orphans, broken links, underlinked hubs, dead ends). Confidence is per-suggestion relevance. Raising confidence does not automatically raise Overview Score until you apply links that change the graph.
Auto-implement (Pro)
Link Automation Pro can apply open suggestions at or above your minimum confidence (70–99%, default 85%). Lower-confidence rows stay on Suggestions for manual review.
Tips
- Start reviews at 70%+ for semantic/commercial links.
- Do not auto-implement until you have accepted a sample by hand.
- Edited anchors still keep the original score unless you regenerate.
