Keyword Coverage Gap Analyzer EN PL

Overview

Compare phrase presence in supplied own and reference texts with English or Polish rules. The Actor returns referenceOnly, ownOnly and shared groups, document counts, original evidence and aliases for duplicate texts. Use the worksheet for a source review before content planning. Language and corpus ownership are user-declared.

Input example

{
  "language": "en",
  "ownPages": [
    {
      "id": "own-home",
      "text": "Steel roof panels. Roof repair guide.",
      "sourceUrl": "https://example.com/own"
    },
    {
      "id": "own-copy",
      "text": "Steel roof panels. Roof repair guide.",
      "sourceUrl": "https://example.com/own-copy"
    }
  ],
  "referencePages": [
    {
      "id": "ref-solar",
      "text": "Solar roof panels. Solar installation guide.",
      "sourceUrl": "https://example.com/reference"
    },
    {
      "id": "ref-price",
      "text": "Solar panels price. Installation price guide.",
      "sourceUrl": "https://example.com/reference-price"
    }
  ],
  "minNgram": 1,
  "maxNgram": 2,
  "minimumDocuments": 1,
  "maxPhrasesPerGroup": 50,
  "maxItems": 1
}

Output

One complete report is one dataset item and GAP_REPORT JSON. GAP_CSV exports one row per returned phrase. REPORT contains a readable summary. The phrases array joins to groups by phraseIds. Counts refer to unique documents per corpus, with normalized text duplicates counted once and original alias IDs retained. suppliedOwnPages and suppliedReferencePages retain the input page counts.

NFC and lowercase normalize words, with a fixed EN/PL stopword list and no stemming. Eligible tokens start with a letter and contain 2 to 64 Unicode code points. Ngrams combine adjacent eligible tokens separated only by whitespace. Stopwords and punctuation break windows. Whitespace includes newlines and paragraph breaks, so this is a lexical comparison without sentence boundary analysis.

minimumDocuments applies to the larger of the two document frequencies. A shared phrase can therefore have fewer occurrences on one side. Groups use document frequency and stable phrase ordering. They retain eligibleCount, returnedCount and truncated. outputTruncated signals any limited group. Evidence preserves original UTF16 offsets and the first matching span from up to three documents per side. Every source ID remains available, and omitted evidence counts are explicit.

Pricing and cost

The proposed price is USD 0.03 for one complete keyword-gap-report comparing up to 50 own and 50 reference supplied pages. Release receipts and the Store determine active pricing. No AI, content requests, proxy or external service is used by the comparison. Cloud cost and margin require separate acceptance.

Limits and known scope

Input is limited to 2 MB, 50 pages per side, 20000 Unicode code points per text and 160 per ID. The index fails before delivery at more than 20000 distinct phrases. Output allows up to 200 phrases per group, with explicit truncation. A 45 second cooperative application budget is checked by the shared runner. It does not interrupt synchronous CPU work mid-function. Acceptance additionally applies a process timeout and records peak process memory.

Source URLs are identifiers. Queries and fragments are removed from output, and embedded credentials are rejected. Text coverage does not establish search volume, ranking difficulty, semantic equivalence, conversion opportunity or traffic growth. Inspect the original contexts and data rights before acting. Remove secrets and unnecessary personal data before upload. Storage retention follows your Apify settings.

Development and recovery

Use Node 22 or newer. The dependency lock is reused from the accepted factory runtime. Install with npm ci --offline --ignore-scripts --omit=dev --no-audit --no-fund when cache is available, then run npm test and npm start. Normal start and test:local request a 128 MiB V8 old generation heap and an 8 MiB semi-space. Total process RSS is measured separately in acceptance. Docker uses npm start and installs the same lock with package scripts disabled. Shared libraries are vendored from _factory/lib.

PREPARED precedes report effects. COMPLETE requires one physical report and the expected named charge count. Replay checks report, CSV, Markdown, runtime and pricing fingerprints, marker counters and physical dataset. Unknown or inconsistent effects require reconciliation before a retry. Preserve default storage and use a single writer for a job. A new intentional job uses fresh storage. CSV IDs are JSON strings inside quoted cells to preserve leading zeros.