There are two kinds of 2000-word articles. The first has twenty dense paragraphs — each naming specific entities, citing numbers, making concrete claims. The second has one good paragraph and nineteen filler — "it's important to note," "at the end of the day," "needless to say," followed by generic advice repeated in slightly different words.
Google's helpful-content system distinguishes between them. Users distinguish between them. AI answer engines pick the dense one to cite.
Paragraph Semantic Density measures the information payload of every paragraph on your page.
The five signals per paragraph
Noun-phrase density. Count of multi-word capitalized spans + "the X of Y" patterns per 100 words. High NP density = the paragraph is saying specific things about specific named entities, not floating generalities.
Entity density. Proper noun count per 100 words. Numbers, brand names, place names, person names. Entity-dense paragraphs are information-rich; entity-free paragraphs are opinion-mush.
Specific-claim markers. Count of numbers (42, 3.2%, 2026), year markers, and money amounts ($99, €15) per 100 words. Specific claims anchor a paragraph to verifiable reality.
Filler ratio. (stopwords + hedges × 3 + filler-phrases × 10) / word count. Hedges like "probably," "basically," "essentially" are weighted higher because they signal the author isn't committed to their own claims. Filler phrases like "needless to say" get 10× weight because their entire function is to fill space.
Word count floor. Paragraphs under 40 words usually don't carry enough payload to score high even if they're dense. Merge or expand.
What to do with the filler list
The AI rewrite prompt emits per-paragraph dispositions: cut / merge / rewrite. Cut when the paragraph is pure hedge-and-stopword content. Merge when it's a short transitional note that fits into the paragraph above or below. Rewrite when the paragraph says something true but vaguely — rewrite with one specific example, one number, one named entity.
Most articles can lose 20-30% of their paragraphs without losing meaning. The remaining paragraphs get a density lift just from no longer being diluted.
Why not just trust Flesch + word count
Voice & Tone measures Flesch and passive-voice ratio — readability concerns. This tool measures something orthogonal: information payload. A highly-readable page can still be full of filler; a dense page can still be readable. They're different axes.
Related reading
- Voice Cleanup — de-slop after density cleanup
- Voice & Tone — readability + passive voice
- Passage Retrievability — paragraph-level retrieval scoring
Fact-check notes and sources
- Helpful Content Update: developers.google.com/search/updates/helpful-content-update
- Information theory primer: en.wikipedia.org/wiki/Information_content
The $20 Dollar Agency covers editing-against-density as a skill. The tool is how you teach an editor to see filler without the editor already being world-class.
