How does WDF·IDF
analysis work?
We fetch the top 10 Google results for your keyword, lemmatize the texts (cases, synonyms), compute WDF·IDF + BM25, build embeddings, and let Claude Opus 4.7 write a brief. All in about a minute. Below is exactly what happens and what you get.
What you get for a minute's wait
Four things, all structured, all exportable (Markdown, PDF), and all actionable — you don't need another tool to start writing.
SEO score
A single number 0–100. Shows how ready your text is to rank for that keyword, based on term coverage, length, structure and title.
Term list
Must-have, should-have, already good, reduce. With target range per term (e.g. 2–4 times). Sorted by importance, not alphabetically.
AI brief
H1, meta description, outline (H2/H3), FAQ questions, tone, guidelines — all generated by Claude Opus 4.7 based on competitors.
Action lists
„Add”, „Boost”, „Reduce”, „Already good” — 4 groups with concrete repetition counts. Hand it to a copywriter and they're off.
Five steps behind the scenes
You see only the final report — we tell you this because we like you to know what you're paying for. Everything runs in parallel, done in 45–60 seconds.
We fetch the SERP
For your keyword we pull the top 10 organic Google results via the DataForSEO API, plus „People Also Ask” questions and related searches.
We clean HTML to plain text
We strip navigation, footer, sidebar, hooks. Only body content remains. Trafilatura does this without configuration.
We lemmatize (cases, plurals)
„Site”, „sites”, „site's” — for us, the same term. CLASSLA Stanza, trained specifically for sr/hr/bs, does this at 97% accuracy.
We compute WDF·IDF + BM25 + embeddings
For every term in every document we compute WDF·IDF weight. Plus BM25 for relevance. Plus LaBSE embeddings to catch synonyms („purchase” ≈ „buying”).
Claude writes the brief
All numbers + your keyword + SERP intent → Claude Opus 4.7 generates H1, meta, outline, FAQ, tone, guidelines specifically for your case. No template — real semantic output.
The formula behind the score
WDF·IDF is a product of two numbers. Each tells you something different. Together they yield a term weight that maps directly to action.
How often a term appears in ONE text. Logarithmic — the first occurrence counts a lot, the tenth counts little. So keyword stuffing doesn't win.
How RARE the term is across the entire corpus (top 10 results). A word that's everywhere (e.g. „the”) counts little. A word in 2/10 texts counts a lot.
Top 5 terms by weight
| Term | WDF | IDF | Weight |
|---|---|---|---|
| seo optimization | 0.78 | 1.34 | 1.05 |
| google ranking | 0.65 | 1.42 | 0.92 |
| keywords | 0.71 | 1.19 | 0.85 |
| link building | 0.42 | 1.86 | 0.78 |
| meta description | 0.38 | 1.62 | 0.62 |
What sets us apart from generic WDF·IDF
Many tools have WDF·IDF, but most are trained on English. We built specifically for our language — the quality difference becomes obvious.
Case lemmatization
„Sajt”, „sajta”, „sajtu”, „sajtovi” are four inflected forms of one Serbian word. We use CLASSLA Stanza, trained on sr/hr/bs corpora, and collapse all of them to the lemma „sajt”. Generic tools trained primarily on English count each inflected form as a different word.
Semantic similarity
„Purchase” ≈ „buying” ≈ „acquiring”. LaBSE multilingual embeddings recognize these as related concepts. Your text doesn't have to literally repeat the same words.
AI brief curator
The brief is not a template. Claude Opus 4.7 reads all the numbers + your keyword + SERP intent and writes H1, outline, FAQ specifically for your case. Output is structured JSON, not strings.
Focus: one keyword → action brief
Other tools cover other needs. If you're looking for any of the below, use a specialized tool — they do that job better and we don't pretend.
Rank tracking over time
Tracking keyword position day to day. Not our focus because it requires daily SERP scraping and a history database.
Backlink analysis
Discovering who links, anchor text, link gap analysis. Different backend, different specialization.
Technical SEO audit
Crawl whole site, broken links, page speed, schema validation. We work on content, not site infrastructure.
Keyword research
Volume, difficulty, seasonality. We work with one keyword already chosen.
Now that you know how it works —
try it on your keyword.
Five analyses in the Free plan. No card, no registration for the first test. In a minute you have a score and an action list.