Keyword Density Score

Overview

You name the terms; we count them. Each keyword is analysed by the same word analyzer as the text — one normalisation form, one case fold, one of three token patterns, one minimum term length — and a phrase matches a run of tokens rather than a substring, so counting on the page and counting here agree. Every figure comes with what produced it: how the analyzer split your keyword, how many occurrences it found, which denominator the ratio used, and where the first match sits. Choosing which terms matter is a different job and a different skill; this one scores the list you already have.

Every keyword tool reports a percentage and none of them says what it divided by, so two tools disagree about the same page and neither can be checked. Here the analyzer, the denominator, the overlap rule and the rounding are all part of the answer, and the answer comes back with them attached.

When to use it

Scope

Non-goals

• Does not discover which terms matter — text.extract_keywords selects terms, and selecting is a judgement while scoring is arithmetic
• Does not use the model tokenizer: a density over subword pieces matches no number you can reproduce by counting words on a page
• Ships no built-in stopword list, because an English one would quietly move the denominator for text in every other language
• Does not accept a pattern of your own, since one over a quarter of a million characters is a denial-of-service waiting to be sent
• Does not rewrite, summarise or return your text

Properties

Capabilities

Cost & billing

Flat rate per invocation

File handling

How we handle your data

Input schema

Output schema

API contract

curl -sS -X POST "$BASE/v1/execute" -H "Authorization: Bearer $KEY" -H 'Content-Type: application/json' -d '{"skillKey":"marketing.keyword_density","input":{"text":"Our crm software helps teams close deals.","keywords":["crm software"]}}'

import { Squillify } from "@squillify/sdk";

const sq = new Squillify({ apiKey: process.env.SQUILLIFY_API_KEY });

const { output } = await sq.execute("marketing.keyword_density", {"keywords":["crm software","deal"],"text":"Our crm software helps teams. The crm software is fast and the crm keeps every deal in one place."});

Errors

Other limits

Limitations

• Does not discover which terms matter — text.extract_keywords selects terms, and selecting is a judgement while scoring is arithmetic
• Does not use the model tokenizer: a density over subword pieces matches no number you can reproduce by counting words on a page
• Ships no built-in stopword list, because an English one would quietly move the denominator for text in every other language
• Does not accept a pattern of your own, since one over a quarter of a million characters is a denial-of-service waiting to be sent
• Does not rewrite, summarise or return your text
• Working out which keywords a document is about
• Counting model tokens for a context budget
• Ranking documents against a query

How to call

POST /v1/execute (requires authentication)

{
  "skillKey": "marketing.keyword_density",
  "input": {
    "keywords": [
      "crm software",
      "deal"
    ],
    "text": "Our crm software helps teams. The crm software is fast and the crm keeps every deal in one place."
  }
}

Examples

{
  "keywords": [
    "crm software",
    "deal"
  ],
  "text": "Our crm software helps teams. The crm software is fast and the crm keeps every deal in one place."
}

{
  "analysis": {
    "rawTokenCount": 19,
    "analyzedTokenCount": 19,
    "stopwordTokenCount": 0,
    "distinctTokenCount": 15,
    "caseFold": true,
    "unicodeForm": "nfkc",
    "tokenPattern": "unicode_word",
    "minTermLength": 1,
    "denominator": "analyzed_tokens",
    "overlap": "non_overlapping_leftmost"
  },
  "aggregate": {
    "matchedTokens": 5,
    "density": 0.2632,
    "percent": 26.3158,
    "keywordCount": 2,
    "matchedKeywordCount": 2,
    "unmatchedKeywordCount": 0
  },
  "density": [
    {
      "keyword": "crm software",
      "tokens": [
        "crm",
        "software"
      ],
      "occurrences": 2,
      "matchedTokens": 4,
      "density": 0.2105,
      "percent": 21.0526,
      "occurrenceRate": 0.1053,
      "firstTokenIndex": 1
    },
    {
      "keyword": "deal",
      "tokens": [
        "deal"
      ],
      "occurrences": 1,
      "matchedTokens": 1,
      "density": 0.0526,
      "percent": 5.2632,
      "occurrenceRate": 0.0526,
      "firstTokenIndex": 15
    }
  ],
  "work": {
    "projected": 57,
    "budget": 8000000
  }
}

{
  "text": "growth growth growth",
  "keywords": [
    "growth growth"
  ],
  "overlap": "ALL_OCCURRENCES"
}

{
  "analysis": {
    "rawTokenCount": 3,
    "analyzedTokenCount": 3,
    "stopwordTokenCount": 0,
    "distinctTokenCount": 1,
    "caseFold": true,
    "unicodeForm": "nfkc",
    "tokenPattern": "unicode_word",
    "minTermLength": 1,
    "denominator": "analyzed_tokens",
    "overlap": "all_occurrences"
  },
  "aggregate": {
    "matchedTokens": 4,
    "density": 1.3333,
    "percent": 133.3333,
    "keywordCount": 1,
    "matchedKeywordCount": 1,
    "unmatchedKeywordCount": 0
  },
  "density": [
    {
      "keyword": "growth growth",
      "tokens": [
        "growth",
        "growth"
      ],
      "occurrences": 2,
      "matchedTokens": 4,
      "density": 1.3333,
      "percent": 133.3333,
      "occurrenceRate": 0.6667,
      "firstTokenIndex": 0
    }
  ],
  "work": {
    "projected": 6,
    "budget": 8000000
  }
}

Recommended use cases

• Check how often a target phrase appears in a draft before it is published
• Compare the same keyword list across several pages, using one analyzer so the figures are comparable
• Verify a supplier's keyword report by recomputing it from the published rules

Avoid for

• Working out which keywords a document is about
• Counting model tokens for a context budget
• Ranking documents against a query

Related skills

Composable with

Safety & privacy

• Runs without calling any external service.
• Your payloads are never stored permanently.
• Telemetry captures metadata only — never your payload contents.
• We never train models on your payloads.
• No human reviews your payloads.

Tags