The partial-match 'safe harbor': testing a comfortable assumption

The partial-match 'safe harbor': testing a comfortable assumption

A frequent claim is that partial-match anchors ('the best tools for X', containing the keyword in a sentence) are a low-risk way to keep keyword relevance. A 2024 dataset of ~3,500 URLs let someone segment by partial-match share specifically.

The finding complicated the comfort: URLs with very high partial-match concentration (>45% of the profile) showed a mild negative association with position — a median ~1.5 positions lower — even though no single anchor was a bare exact-match.

The interpretation: over-optimization may be assessed on keyword presence across the profile, not on the narrow exact-match bucket alone. Stuffing the keyword into many 'natural-looking' partial anchors can still concentrate the same signal.

On one hand, this aligns with the idea that algorithms phrase-match within anchors rather than checking for string equality. On the other, the association is weak (~1.5 positions), and partial-match-heavy profiles may simply belong to keyword-obsessed sites that are over-optimized everywhere.

Limitation: weak effect size, observational, no control for on-page over-optimization that likely co-occurs.

Confidence: low-to-moderate that partial-match isn't the safe harbor it's assumed to be.

Open question: is keyword density measured per-anchor, or summed across the whole anchor profile?