BM25F is a field-aware extension of the BM25 ranking function: it scores how well a term matches a page by weighting the page’s zones — title, headings, body, and so on — separately, then applies term-frequency saturation and corpus statistics before producing a single relevance score.
| Term | BM25F |
|---|---|
| Category | Statistics and Evidence |
| Also known as | Best Matching 25 Fields |
| Where it appears | Workbook BM25F, Settings → Factors, and BM25 Drafts |
What it means in RankGear
BM25F is the scoring model behind the BM25F workbook, and it can be enabled as a factor under Settings → Factors and referenced in BM25 Drafts. Plain BM25 treats a page as one undifferentiated bag of words, so a keyword in the title counts the same as one in the footer. BM25F recognizes that those zones are not equal: it weights each field before combining them, then dampens repeated occurrences and adjusts for how common the term is across the pages in the run. The number RankGear reports is that combined relevance score for your term against a page, computed the same way for every page in a run so the scores can be compared like with like.
How to interpret it
Read the score by its scale, its sample, and its spread rather than in isolation. BM25F has no fixed ceiling — the range depends on the corpus and the field weights in play — so a raw figure means little on its own. What matters is how your page compares with the pages already ranking for the keyword: a higher score means the term is more prominent across the weighted fields, not that the page will rank. Check how many SERP pages the corpus statistics were drawn from, since a score built on three pages is shakier than one built on ten, and confirm whether a zero is a genuine absence of the term or a field that was never measured.
| Component | What it contributes |
|---|---|
| Field weighting | Counts the term differently by zone — title and headings carry more weight than body or boilerplate. |
| Saturation | Dampens repeated occurrences so stuffing the same word yields diminishing returns. |
| Corpus statistics | Down-weights terms that are common across the ranking set and rewards the ones that distinguish a page. |
Example
Suppose you are optimizing for “fleet insurance” and two competitor pages mention the phrase the same number of times overall. One places it in the title and the H1; the other uses it only in body paragraphs. A plain term-frequency count would call the two pages equal, but BM25F scores the first higher because the title and heading fields carry more weight. Comparing your own page’s BM25F score against that set tells you whether the term is landing in the zones that count, not just how often it appears.
Important considerations
- The scale is corpus-relative: there is no universal “good” BM25F score, so always read your page’s number against the pages ranking for the same keyword rather than as an absolute.
- Field weights drive the result — the same term count can produce very different scores depending on which zones it lands in, so the placement matters as much as the frequency.
- Small SERP samples make the corpus statistics unstable; two runs can show a similar average score with very different spread, which changes how confidently you can pick a practical target.
- A higher BM25F score is a comparative relevance indicator, not a cause of ranking: statistical association is not causation, and this is a provider-computed comparative signal rather than one of Google’s own scores.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.