Preskoči navigaciju
WDFIDF·Pro
Edu · explained without a PhD

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.

01

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.

02

Term list

Must-have, should-have, already good, reduce. With target range per term (e.g. 2–4 times). Sorted by importance, not alphabetically.

03

AI brief

H1, meta description, outline (H2/H3), FAQ questions, tone, guidelines — all generated by Claude Opus 4.7 based on competitors.

04

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.

01

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.

DataForSEO · 10 SERP slots · PAA + related
02

We clean HTML to plain text

We strip navigation, footer, sidebar, hooks. Only body content remains. Trafilatura does this without configuration.

Trafilatura · BeautifulSoup fallback
03

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.

CLASSLA Stanza · POS tagging · 97% accuracy
04

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”).

Pure Python · LaBSE multilingual · BM25
05

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.

Claude Opus 4.7 · structured JSON output · prompt cache

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.

WDFWithin-Document Frequency

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.

„optimization” 5× in text → WDF ~ 0.62
IDFInverse Document Frequency

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.

„link building” in 2/10 → IDF high
Example · keyword: „seo agency belgrade”

Top 5 terms by weight

TermWDFIDFWeight
seo optimization0.781.341.05
google ranking0.651.420.92
keywords0.711.190.85
link building0.421.860.78
meta description0.381.620.62

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.

A

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.

97% accuracy
B

Semantic similarity

„Purchase” ≈ „buying” ≈ „acquiring”. LaBSE multilingual embeddings recognize these as related concepts. Your text doesn't have to literally repeat the same words.

LaBSE · 109 languages
C

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.

Claude Opus 4.7
Honestly · what we DON'T do

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.

Try: Specialized rank tracker

Backlink analysis

Discovering who links, anchor text, link gap analysis. Different backend, different specialization.

Try: Specialized backlink tool

Technical SEO audit

Crawl whole site, broken links, page speed, schema validation. We work on content, not site infrastructure.

Try: SEO crawler

Keyword research

Volume, difficulty, seasonality. We work with one keyword already chosen.

Try: Keyword research tool

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.