We don't count keywords.
BERT scores meaning — then we move the score.
Every entity, heading and paragraph RankToOne ships is graded by production BERT embeddings against the live SERP. You don't get word-count targets. You get a cosine you can move — and an auto-repair pass that moves it.
Free credits · No card · Cosine you can audit
BERT engine by the numbers
Four reasons BERT is the engine, not a buzzword
These aren't post-hoc checks. They run inside extraction, brief generation, internal linking and URL audits.
BERT-grade embeddings, not keyword counts
Every entity, heading, paragraph, and SERP chunk is converted into a 768-dimensional BERT-family embedding. We compare meaning — not strings — so a brief about 'cold-pressed juice' aligns with a SERP that talks about 'raw, unpasteurized juice'.
Asymmetric query ↔ document compare
Seed keyword and entities embed as RETRIEVAL_QUERY. Brief body, headings and SERP chunks embed as RETRIEVAL_DOCUMENT. That's how BERT was actually trained for retrieval — and it lifts true alignment cosine ~3–8 points vs naive symmetric scoring.
Four-bar Semantic Alignment Score
Keyword ↔ Brief, SERP Coverage, Internal Cohesion and Entity Co-Occurrence — each scored 0–100 with calibrated BERT cosine thresholds. The brief doesn't just look good; it can be measured against the live SERP.
Auto-repair pass when cosine drops
If any bar falls below threshold, a repair pass rewrites the weakest headings and intro until cosine clears the bar — then we re-embed and re-score. Most tools tell you the brief is weak; we fix it before you see it.
Where BERT runs inside the product
Six concrete touchpoints — every one is a measurable cosine, not a vibe.
Entity extraction grounded in BERT relevance
After NER + topical extraction, every candidate entity is BERT-scored against the seed and the SERP. Entities below the relevance floor are dropped; survivors carry a 0–100 cosine-derived score that drives the H2/H3 skeleton.
Heading cohesion peak at 0.68
We measure pairwise heading cosine. Too low = scattered article. Too high (≥0.85) = redundant H2s. The cohesion peak sits at 0.68 — and the score punishes distance from that peak in both directions.
Entity co-occurrence sentence-scoring
BERT confirms which co-occurring entities sit in the same sentence as the primary keyword. That's the signal Google's passage indexing actually rewards — and we flag any missing co-occurrence the SERP rewards.
Internal links picked by cosine + reranker
Anchor candidates are BERT-embedded, top-N cosine, then passed through a semantic reranker for a final relevance_score. Only links above the threshold ship into the brief — no 'related posts' filler.
URL Alignment audit on live pages
Paste any URL — yours or a competitor's. BERT scores page ↔ keyword, entity semantic coverage (Strong / Weak / Missing) and cohesion. You see exactly which entities the page is failing on a vector basis, not a keyword basis.
Q&A Extractability scored with BERT
Generated Q&A pairs are BERT-checked for snippet-friendly first sentences (15–32 words, declarative, subject-noun fronted). Failing answers are auto-rewritten until the extractability score clears 70.
The four bars you'll see on every brief
Asymmetric kw-query × body-doc cosine
Heading vs top-SERP chunk max-cosine
Pairwise heading cosine vs 0.68 peak
Primary-kw sentence sharing entities
Most tools stop at term frequency. We start at cosine.
Most 'AI SEO tools' count keywords. We compare meaning.
TF-IDF and LSI tools still dominate the space. RankToOne uses production BERT embeddings end-to-end — extraction, ranking, internal links, alignment, and repair.
We expose the score. Then we move it.
You see the cosine. You see which entities are Missing vs Weak vs Strong. You see exactly where to write to close the gap — and the auto-repair pass writes it for you when you ask.
Calibrated, not generic.
Thresholds are tuned per use case (entity coverage ≥0.60, weak ≥0.40, cohesion peak 0.68, extractability ≥70). Generic cosine bars lie. Ours don't.
BERT, no hand-waving
See your brief's cosine. Then watch us move it.
Free credits, no card. Run one BERT-scored brief and audit any URL on the same engine.