How ranking works

Link Pulse ranks internal link candidates with a deterministic model: same content, same settings, same result. There is no model inference and no external call. Every score is a calculation you can read.

The eight signals

Each signal is scored from 0.0 to 1.0, multiplied by its weight, then summed. The weights sum to 1.00.

Taxonomy overlap Shared categories and tags between the source and candidate. 0.30 Title overlap Lexical overlap between post titles. 0.20 Body keyword overlap Keyword overlap in body text. 0.15 Heading overlap Overlap in heading text. 0.10 Link graph relevance Existing internal-link neighbourhood between the pages. 0.10 Excerpt overlap Overlap in excerpts. 0.05 Custom field overlap Overlap in indexed custom fields. 0.05 Priority score Manual priority weighting when set. 0.05

The weights sum to 1.00.

Why taxonomy leads

Shared words are weak evidence of relevance. A post titled “The Best Books of 2025” and a post titled “Best Camera Bags” share “best”; they share little else. Shared taxonomy is stronger evidence that two pieces of content are actually related. Weighting taxonomy at 0.30 — above title overlap at 0.20 — is what prevents shallow word collisions from outranking genuine topical relationships.

How the score is built

1. Score each applicable signal from 0.0–1.0.
2. Multiply by that signal’s weight.
3. Sum the weighted values.
4. Clamp the result to a valid match score range.
5. Apply the overlinked penalty (below) after the weighted sum.

Match strength badges (Strong, Good, Fair, Weak) are labels on that final score for the sidebar — they do not change the calculation.

Applicable-signal renormalisation

Not every site uses every signal. Renormalisation decides what happens to a signal’s weight when it is not meaningful on your install.

If your site does not use a signal at all — for example you have no custom fields indexed, or almost no posts are tagged — that signal is dropped from the calculation. Its weight is not left in place to permanently cap every score.

If your site does use a signal, a post that lacks it takes a real zero on that signal. The weight is not redistributed onto title matching (or any other signal). Pushing unused weight onto title overlap would reward shallow word collisions — the failure taxonomy-first weighting exists to prevent.

Overlinked penalty

After the weighted sum, Link Pulse applies an overlinked penalty: a target that is already heavily linked internally scores lower than the same target would with fewer inbound links. The penalty does not replace the signal model; it adjusts the final score so suggestions prefer targets that still need links.

What is not a weighted signal

Slug relevance is used to find candidates. It is not one of the eight weighted signals in Free.

Reproducibility

Same published content, same context-bank settings, same index → same ranking. Scores do not drift between runs. A suggestion you dismissed does not reappear with a different unexplained score the next day.

Related: Editor sidebar, Context bank, Features.