PDF Quote Finder & Page Citation Locator

What does PDF Quote Finder & Page Citation Locator do?

PDF Quote Finder & Page Citation Locator finds a supplied quote in one public PDF's embedded text and returns deterministic page-level locations. It is a focused PDF citation API for agents and workflows that need the passage, context, page number, page-local offsets, document hash, and retrieval provenance without downloading or exposing a PDF's complete text.

The Actor retrieves one public HTTPS PDF, parses only its embedded text with PDF.js, and searches a documented extracted-text representation. It does not use an LLM, OCR, browser rendering, uploads, authentication, batches, visual coordinates, or semantic verification. A no-match means only that the text was not found in this embedded-text representation; it does not prove that the text is absent from page images or that a source supports a claim.

Why use PDF Quote Finder & Page Citation Locator?

• Find a quote and its physical PDF page number before creating a citation.
• Return compact evidence instead of a full document-text payload.
• Use exact, Unicode/whitespace/punctuation-normalized, or deterministic fuzzy matching.
• Receive SHA-256 hashes, redirects, retrieval timestamps, PDF metadata, page coverage, and stable machine errors.
• Keep SSRF, redirect, decompression, PDF page, extracted-text, parser-time, and worker-memory limits fixed and safe by default.
• Invoke it through the Apify API, schedules, integrations, or Apify MCP, with standard monitoring and dataset export.

Example agent queries: “Find this quote and page number in a PDF,” “Does this PDF contain the quoted passage?”, “Locate this sentence in a PDF,” and “Return a PDF citation location for this text.”

What data does the PDF quote locator return?

Field	Type	Description
verification	object	Match decision, method, score, passage, context, global offsets, and page locations
verification.location	object	First match's one-based physical page(s) and local/global UTF-16 offsets
verification.locations	array	Up to 20 exact/normalized occurrences in document order
source	object	Redacted URLs, status, hash, byte length, timestamp, and redirects
pdf	object	PDF.js version, PDF format, physical page count, labels, and sanitized metadata
extraction	object	Embedded-text page coverage and representation facts
warnings / error	array / object	Stable machine-readable limitations and failure details

How to find a quote and page number in a PDF

1. Open the Actor's Input tab.
2. Enter a direct public HTTPS PDF URL.
3. Paste the text to locate.
4. Use normalized for typical quotations; select exact only for literal extracted text, or fuzzy for limited deterministic variation.
5. Start the run and inspect the single dataset item for verification.location.

Respect the source's terms, copyright restrictions, privacy expectations, and applicable law when retrieving a PDF.

How much does PDF Quote Finder & Page Citation Locator cost?

The release hypothesis is $0.01 per completed pdf-quote-location determination. A match and an embedded-text no-match are billable because both deliver a complete bounded determination. Invalid PDFs, password-protected PDFs, responses over the resource limits, empty embedded text, parser failures, and network failures are not charged.

The Actor is not ready for publication until a fixed private Cloud benchmark confirms p95 platform cost below $0.004, exactly one custom event per completed result, and no charges for free failures. The platform's live PPE settings—not this README—remain the source of truth before release.

Input

See the Input tab for the complete schema. The V1 contract deliberately has only three fields:

{
  "url": "https://example.com/document.pdf",
  "text": "Example quoted passage",
  "matching": "normalized"
}

url must be a public HTTPS URL with no credentials. A URL fragment such as #page=4 is ignored and reported as URL_FRAGMENT_IGNORED; it never changes matching behavior. text must contain non-whitespace characters and is limited to 5,000 UTF-16 code units. Exact matching compares PDF.js content-order output literally. Normalized matching uses NFKC, case folding, whitespace collapse, and equivalent quote/dash punctuation; it intentionally does not remove line-break hyphens. Fuzzy matching is a fixed token-F1 plus character-trigram-Dice algorithm and returns only its single best candidate.

Output

You can download the single dataset envelope in JSON, HTML, CSV, or Excel. Exact and normalized locations are exhaustive in totalOccurrenceCount, while the returned location list is capped at 20.

{
  "contractVersion": "1.0",
  "ok": true,
  "pdf": { "physicalPageCount": 12, "pdfJsVersion": "6.2.108" },
  "verification": {
    "match": true,
    "matchType": "normalized_exact",
    "searchScope": "embedded_text_only",
    "passage": "Example quoted passage",
    "totalOccurrenceCount": 1,
    "location": {
      "startPage": 4,
      "endPage": 4,
      "startPageOffset": 182,
      "endPageOffset": 204,
      "startOffset": 1398,
      "endOffset": 1420
    }
  },
  "warnings": []
}

Limits and accuracy

V1 accepts a single PDF up to 10 MiB after decompression, at most 200 physical pages, and at most 2,000,000 extracted UTF-16 code units. It uses one newline between trimmed physical-page strings; the separator belongs to neither page. Results never contain the complete extracted document text.

PDF layout order may differ from visual reading order, and hyphenation across a rendered line can prevent an exact or normalized match. This Actor does not reconstruct columns, tables, glyph positions, or text from images. It returns textual occurrence evidence only and never asserts citation correctness, semantic support, entailment, or factual truth.

FAQ, disclaimer, and support

Why did a scanned PDF return NO_EXTRACTABLE_TEXT?

The PDF has no usable embedded text. OCR and image-text claims are intentionally outside V1, so no textual no-match can be inferred from rendered page images.

Why is a result free but ok: false?

The document may be invalid, password-protected, too large, over the page/text limits, or lack embedded text. These are structured handled outcomes. Transport, timeout, worker, and unexpected parser failures also persist an error item where possible but fail the run because they are operational failures.

This Actor only retrieves public data that the user chooses to provide. Results may contain personal data from public PDFs; process them only with a legitimate purpose and in compliance with GDPR, other applicable regulations, source terms, and copyright law. For help or feedback, use the Actor's Issues tab. Programmatic callers can use the API tab and Apify MCP.