Exact Match Position Variance

Exact Match Position Variance reports how spread out the exact search phrase is within a page’s body — it returns the average distance of each occurrence from their collective center, so a low number means the matches cluster together and a high number means they are scattered across the content. Despite the name, it is a mean absolute deviation, not statistical variance, and neither a low nor a high value is universally better.

What it measures

This factor reports the spread of exact search-phrase positions around their average. Rather than counting how often the phrase appears, it describes how evenly those appearances are distributed through the body: when the exact phrase surfaces in a few tight clusters the spread is small, and when it is peppered across widely separated passages the spread is large.

How RankGear measures it

RankGear finds every exact-phrase UTF-16 offset in the cleaned, lowercased body, computes the mean of those offsets, then averages the absolute distance of each offset from that mean. Despite “Variance” in the name, this is mean absolute deviation, not statistical variance. With no positions, it returns 0.

positions = [ offset of each exact-phrase match ]
mean      = average(positions)
variance  = average( abs(offset - mean) for offset in positions )
// returns 0 when positions is empty

How to optimize it

Treat this as an observation about phrasing distribution rather than a target to chase. Distribute the exact phrasing only where it reads naturally — use it in the key explanatory sections where it genuinely clarifies, and lean on broader vocabulary elsewhere to avoid repetitive writing. Because lower and higher are not universally better, there is no ideal number to engineer toward; let the spread follow from writing the phrase where it belongs.

Important considerations

• Positions are UTF-16 offsets — not words, sentences, pixels, or HTML byte positions.
• Cleaned body content excludes markup and may differ from the raw source, so the offsets are measured against the stripped text, not what you see in the editor.
• The generated variation set affects variation-based factors, so changes here can ripple into related distribution signals.
• These are descriptive distribution statistics; lower or higher is not universally better. The value relates to how a page sits within a measured result set — it is a prioritization signal, not a cause of ranking, and it guarantees nothing.

Related factors

• Exact Match Clustering Score
• Exact Matches in Search Result Link Text

Part of the Factors reference · how RankGear measures · glossary.