This lesson turns a hundred thousand raw entity mentions into one defensible verdict per page. It shows that the filtering is where the expertise lies, defines what counts as a page’s dominant entity, and produces the first of the five scores.
What this lesson covers
- Removing numbers and dates, generic nouns and leftover site chrome
- Tuning a chrome threshold by testing it rather than inheriting it
- A dominant-entity definition with a salience floor and a diffuseness flag
- The profile row and its intent-match column
- Phrasing findings so they get acted on
- Scoring Clarity from the demo site’s profiles
Three filters come first, in order. Structural types such as numbers, dates and prices are real entities but not topics. Generic nouns such as “data”, “content” and “course” appear everywhere and signal nothing. Site chrome is the one people miss: on the demo site the button labels “Purchase course” and “View course” appeared on around 11% of pages each. A rule catching entities that appear on a large share of pages and never as a proper noun finds them — but a 40% threshold caught only one entity, a genuine topic, and nothing else; catching the chrome meant coming down to about 10% and reviewing 129 candidates by hand. The lesson’s point is that no threshold does this for you.
With the noise removed, a page’s dominant entity has to clear a salience floor of 0.05, appear more than once and ideally appear as a proper noun, and a page with nothing above 0.12 is flagged as diffuse. Rolled up to one row per URL, the profile compares the detected subject with the intended one. Across 637 pages of the demo site, the dominant entity appeared in the title on 42% of pages, in the H1 on 30% and in the meta description on only 20%, while 26% of pages were too diffuse to read confidently. Loaded into the Scorecard, that is Clarity at one out of three — with the meta-description row marking the cheapest fix in the audit.
This lesson continues Module 2, Diagnostics: Extracting the Truth About Your Site, in The AI-powered Entity Audit, and is followed by a practical that runs the Extractor and the profile step on a whole site. It suits SEOs who need to turn entity data into page-level findings a client will act on.