Density is a ratio, not a count. This analyser divides how often a term appears by the total number of words that survived filtering, and reports the result as a percentage — which means the same word appearing five times is unremarkable in a chapter and dominant in a book description. Length is half the answer, and counting alone never tells you that.
Two controls decide what gets counted. A minimum word length filters out short fragments, and a minimum occurrence threshold hides anything appearing fewer times than you specify, which is what stops a long text producing a list of thousands of words used once. A built-in list of common function words is removed regardless, since the and of would otherwise top every result.
Multi-word phrases are extracted alongside single words, and this is usually where the real finding is. A phrase is kept unless every word in it is a function word, so genuine combinations survive while empty strings like of the are discarded. Readers search in phrases far more than in single words, and a description can contain every individual term you care about while containing none of the phrases anyone types.
The tool reports the number and does not prescribe one, because published targets are convention rather than measured fact and no marketplace documents a threshold. Density is most useful comparatively: measure text that already performs well and see where yours sits, rather than optimising toward a figure someone asserted.
Because the denominator changes. Density is calculated against the words that survive filtering, so raising the minimum word length removes short words from the total and every remaining percentage rises. Compare two texts only when both were measured on identical settings.
Most often the occurrence threshold. A phrase appearing fewer times than your minimum is hidden, and phrases naturally repeat less than single words, so a threshold that suits word analysis can empty the phrase list. Lower it and the phrases reappear.