1. Reading the pages. EntityGap reads the title, headings, main content and JSON-LD of each page, skipping menus, footers, buttons and cookie banners. Sites that build their content with JavaScript are rendered on the server first. Each page is normalised and pages are averaged, so longer or more numerous pages don't dominate.
2. Finding entities. Proper names (including standards like “UNI EN 14411”) and recurring noun phrases are extracted with English and Italian word lists; generic words, verbs, shop vocabulary and product codes are filtered out; singular and plural forms are folded together. Mentions are weighted by position: title and H1 ×3, H2–H3 ×2, opening text ×1.5. Prominence is a transparent proxy, not Google's salience.
3. Knowledge Graph. The 40 most prominent candidates are matched to Wikidata by exact name; disambiguation pages and surnames are discarded; among homonyms the one fitting the page's vocabulary wins; ambiguous words without context stay unlinked rather than linked wrongly.
4. Gaps. A benchmark topic appears on at least half of the competitors (at least two) with a weight of at least 0.5%. Every topic is also searched in the full text of every site, so a passing mention is “too weak”, never “missing”. Brand names are excluded.
5. Target search and competitors. The search is matched to Wikidata entities. Each topic gets a relevance from 0 to 1 (search concept, shared words, co-occurrence above the page baseline) and gaps are ranked by competitor weight × (0.3 + relevance). “Find my competitors on Google” takes the top 10 organic results for the market and keeps the first four business pages.
6. Score. A diagnostic index, not a Google ranking factor: topic coverage 35%, search focus 15% (title, H1 and weight of search-related topics), Knowledge Graph recognition vs competitors 15%, brand visibility 15%, structured data 20%. Components that can't be measured are left out and weights rescaled.
Limits. Extraction models what engines see, it doesn't reproduce Google. Niche entities often aren't in Wikidata, and that's normal. Review the gap list with business judgment: a topic competitors cover is a candidate, not an obligation.
