PDF Text Extractor: OCR & Metadata

📄 Turn PDFs into text with OCR

For developers, researchers, and teams looking for PDF to text OCR, this Actor reads public PDF URLs and returns plain text, Markdown, page details, OCR details, metadata, bookmarks, and text chunks in your dataset. Choose pages when you need part of a document, or use OCR for scanned pages.

• Read a scanned document with Scanned PDF to Text and save its recovered text.
• Extract Text from PDF returns readable text from a public PDF URL.
• Turn a public document into plain text with PDF to Text.
• Read scanned PDF pages with Online OCR and review the returned text.
• Recover text from image-only pages with OCR PDF.

📚 PDF text, pages, and document details

Each saved dataset row describes one submitted PDF. It can include the source URL, file name, media type, file size, page count, extracted text, Markdown, text counts, page-level text, OCR details, PDF metadata, bookmarks, and page-tagged chunks.

▶️ Run PDF text extraction

1. Add one or more public PDF URLs.
2. Optionally enter page numbers or ranges and choose whether to use OCR for scanned pages.
3. Run the Actor and open the PDF results dataset.

⚙️ Input

Input fields

Field	Type	What it does
pdfSources	array of objects	Required. Adds one or more public PDF sources.
pdfSources[].url	string (URL)	Public URL of the PDF to read.
pages	string	Optional page numbers or ranges, separated by commas, such as 1,3-5. Leave empty to read every page.
useOcr	boolean	Uses OCR for image-only or scanned pages when true. Set it to false to skip OCR and use normal PDF text extraction. Defaults to true.

Default input

This is the smallest common input from a successful current-beta run:

{
  "pdfSources": [
    {
      "url": "https://www.w3.org/WAI/ER/tests/xhtml/testfiles/resources/pdf/dummy.pdf"
    }
  ],
  "useOcr": true
}

🧾 Output

Dataset fields

Field	Type	What it does
sourceUrl	string (URL)	Public URL of the PDF read for this row.
fileName	string	File name found for the PDF, when available.
mimeType	string	Media type reported for the source file, when available.
fileSizeBytes	integer	Source PDF size in bytes, when available.
pageCount	integer	Number of pages in the source PDF.
text	string	Text read from the requested pages, including OCR text when OCR is used.
markdown	string	Markdown version of the extracted text with layout cues when available.
characterCount	integer	Character count for the requested content.
wordCount	integer	Word count for the requested content.
pages	array of objects	Text and counts for each requested page.
pages[].pageNumber	integer	Number of the page in the source PDF.
pages[].text	string	Text read from that page.
pages[].characterCount	integer	Character count for that page.
pages[].wordCount	integer	Word count for that page.
ocr	object	OCR coverage and signals for the row.
ocr.pagesProcessed	integer	Number of pages read with OCR.
ocr.language	string	OCR language, when the source reports it.
ocr.confidence	number	OCR confidence reported by the source, when available.
ocr.remainingScannedPages	integer	Scanned pages left without OCR text, when reported.
metadata	object	Metadata stored in the source PDF, when available.
metadata.title	string	Title stored in the PDF metadata.
metadata.author	string	Author stored in the PDF metadata.
metadata.subject	string	Subject stored in the PDF metadata.
metadata.keywords	string	Keywords stored in the PDF metadata.
metadata.creator	string	Application named as the PDF creator, when available.
metadata.producer	string	Application named as the PDF producer, when available.
metadata.creationDate	string	Creation date stored in the PDF metadata.
metadata.modificationDate	string	Modification date stored in the PDF metadata.
bookmarks	array of objects	Bookmark or table-of-contents entries found in the PDF.
bookmarks[].title	string	Title of the bookmark entry.
bookmarks[].pageNumber	integer	Page opened by the bookmark, when it has a destination.
bookmarks[].level	integer	Nesting level of the bookmark in the PDF outline.
chunks	array of objects	Page-tagged text chunks for embedding and citation work, when produced.
chunks[].text	string	Text in the chunk.
chunks[].pageNumber	integer	Page that the chunk belongs to.

Standard PDF row

This complete row is from a successful current-beta run:

{
  "sourceUrl": "https://www.w3.org/WAI/WCAG22/working-examples/pdf-bookmarks/bookmarks.pdf",
  "fileName": "bookmarks.pdf",
  "mimeType": "application/pdf",
  "fileSizeBytes": 106567,
  "pageCount": 2,
  "text": "1 Contents Header One ................................................................................................................................................... 2 Header Two for List ................................................................................................................................... 2 Header Three Text ................................................................................................................................ 2 Another Header One ..................................................................................................................................... 2 Another Header Two for List..................................................................................................................... 2",
  "markdown": "1 Contents Header One ................................................................................................................................................... 2 Header Two for List ................................................................................................................................... 2 Header Three Text ................................................................................................................................ 2 Another Header One ..................................................................................................................................... 2 Another Header Two for List..................................................................................................................... 2",
  "characterCount": 776,
  "wordCount": 28,
  "pages": [
    {
      "pageNumber": 1,
      "text": "1 Contents Header One ................................................................................................................................................... 2 Header Two for List ................................................................................................................................... 2 Header Three Text ................................................................................................................................ 2 Another Header One ..................................................................................................................................... 2 Another Header Two for List..................................................................................................................... 2",
      "characterCount": 776,
      "wordCount": 28
    }
  ],
  "ocr": {
    "pagesProcessed": 0
  },
  "metadata": {
    "title": "Working example of creating bookmarks in PDF documents",
    "creator": "Acrobat PDFMaker 9.0 for Word",
    "producer": "Adobe PDF Library 9.0",
    "creationDate": "D:20110202105905-05'00'",
    "modificationDate": "D:20230718095527-07'00'"
  },
  "bookmarks": [
    {
      "title": "Header One",
      "pageNumber": 2,
      "level": 0
    },
    {
      "title": "Header Two for List",
      "pageNumber": 2,
      "level": 1
    },
    {
      "title": "Header Three Text",
      "pageNumber": 2,
      "level": 2
    },
    {
      "title": "Header Four",
      "pageNumber": 2,
      "level": 3
    },
    {
      "title": "Another Header One",
      "pageNumber": 2,
      "level": 0
    },
    {
      "title": "Another Header Two for List",
      "pageNumber": 2,
      "level": 1
    },
    {
      "title": "Links\r",
      "level": 0
    },
    {
      "title": "WCAG2.0",
      "pageNumber": 2,
      "level": 1
    }
  ],
  "chunks": [
    {
      "text": "1 Contents Header One ................................................................................................................................................... 2 Header Two for List ................................................................................................................................... 2 Header Three Text ................................................................................................................................ 2 Another Header One ..................................................................................................................................... 2 Another Header Two for List..................................................................................................................... 2",
      "pageNumber": 1
    }
  ]
}

OCR PDF row

This complete row is from a successful current-beta OCR run:

{
  "sourceUrl": "https://raw.githubusercontent.com/tinytoolkit-org/pdf-sample-files/4fd733c47ead5446b06372ab910cb78d1c7da55e/out/sample-scanned.pdf",
  "fileName": "sample-scanned.pdf",
  "mimeType": "application/octet-stream",
  "fileSizeBytes": 432160,
  "pageCount": 3,
  "text": "Scanned-style sample (image-only pages)\n\nPage 1 of 3\n\nThis document is a machine-generated sample file. It contains no real data: every name, number, and\nfigure on this page is placeholder content produced for software testing.\n\nUse it to exercise PDF tooling — text extraction, page manipulation, rendering, compression — without\nworrying about licensing or privacy. The file is dedicated to the public domain under CCO.\n\nText on this page is real, extractable text set in a standard font, not an image. A text-extraction tool\nshould recover these paragraphs verbatim, in reading order, with no OCR involved.\n\nPage boundaries matter for testing. Each page carries a heading with its own page number so that split,\nextract, reorder, and delete operations can be verified against what the output claims.\n\nIf a tool you are testing reports a different page count, a different order, or drops one of these\nparagraphs, the tool — not this file — is the thing to investigate next.\n\nA reasonable test suite checks the boring cases first: one page, ten pages, a page with nothing on it, a\npage rotated sideways. The files in this collection cover each of those separately.\n\n©CO / public domain — generated sample, no real data",
  "markdown": "Scanned-style sample (image-only pages)\n\nPage 1 of 3\n\nThis document is a machine-generated sample file. It contains no real data: every name, number, and\nfigure on this page is placeholder content produced for software testing.\n\nUse it to exercise PDF tooling — text extraction, page manipulation, rendering, compression — without\nworrying about licensing or privacy. The file is dedicated to the public domain under CCO.\n\nText on this page is real, extractable text set in a standard font, not an image. A text-extraction tool\nshould recover these paragraphs verbatim, in reading order, with no OCR involved.\n\nPage boundaries matter for testing. Each page carries a heading with its own page number so that split,\nextract, reorder, and delete operations can be verified against what the output claims.\n\nIf a tool you are testing reports a different page count, a different order, or drops one of these\nparagraphs, the tool — not this file — is the thing to investigate next.\n\nA reasonable test suite checks the boring cases first: one page, ten pages, a page with nothing on it, a\npage rotated sideways. The files in this collection cover each of those separately.\n\n©CO / public domain — generated sample, no real data",
  "characterCount": 1219,
  "wordCount": 203,
  "pages": [
    {
      "pageNumber": 1,
      "text": "Scanned-style sample (image-only pages)\n\nPage 1 of 3\n\nThis document is a machine-generated sample file. It contains no real data: every name, number, and\nfigure on this page is placeholder content produced for software testing.\n\nUse it to exercise PDF tooling — text extraction, page manipulation, rendering, compression — without\nworrying about licensing or privacy. The file is dedicated to the public domain under CCO.\n\nText on this page is real, extractable text set in a standard font, not an image. A text-extraction tool\nshould recover these paragraphs verbatim, in reading order, with no OCR involved.\n\nPage boundaries matter for testing. Each page carries a heading with its own page number so that split,\nextract, reorder, and delete operations can be verified against what the output claims.\n\nIf a tool you are testing reports a different page count, a different order, or drops one of these\nparagraphs, the tool — not this file — is the thing to investigate next.\n\nA reasonable test suite checks the boring cases first: one page, ten pages, a page with nothing on it, a\npage rotated sideways. The files in this collection cover each of those separately.\n\n©CO / public domain — generated sample, no real data",
      "characterCount": 1219,
      "wordCount": 203
    }
  ],
  "ocr": {
    "pagesProcessed": 1,
    "language": "eng",
    "confidence": 94,
    "remainingScannedPages": 0
  },
  "metadata": {
    "title": "Scanned-style sample",
    "author": "tinytoolkit sample files",
    "subject": "Image-only pages — no text layer, use for OCR testing",
    "keywords": "sample test pdf cc0 pdftoolskit.org",
    "creator": "https://pdftoolskit.org/sample-pdfs",
    "producer": "pdf-sample-files generator (pdf-lib)",
    "creationDate": "D:20260719000000Z",
    "modificationDate": "D:20260719000000Z"
  },
  "chunks": [
    {
      "text": "Scanned-style sample (image-only pages) Page 1 of 3 This document is a machine-generated sample file. It contains no real data: every name, number, and figure on this page is placeholder content produced for software testing. Use it to exercise PDF tooling — text extraction, page manipulation, rendering, compression — without worrying about licensing or privacy. The file is dedicated to the public domain under CCO. Text on this page is real, extractable text set in a standard font, not an image. A text-extraction tool should recover these paragraphs verbatim, in reading order, with no OCR involved. Page boundaries matter for testing. Each page carries a heading with its own page number so that split, extract, reorder, and delete operations can",
      "pageNumber": 1
    },
    {
      "text": "be verified against what the output claims. If a tool you are testing reports a different page count, a different order, or drops one of these paragraphs, the tool — not this file — is the thing to investigate next. A reasonable test suite checks the boring cases first: one page, ten pages, a page with nothing on it, a page rotated sideways. The files in this collection cover each of those separately. ©CO / public domain — generated sample, no real data",
      "pageNumber": 1
    }
  ]
}

💳 Pricing

How charges work

The PDF text event is charged once for each PDF that is processed and returns extracted text. The OCR page event is charged for each scanned or image-only page successfully read with OCR and returned as text. Current tiered rates are shown in the Pricing tab.

🔌 Integrations

Start runs in Apify Console or through the Apify API, then read the PDF results dataset in JSON, CSV, or Excel. Use the dataset in your own script or document workflow.

❓ FAQ

Can this read scanned PDFs?

Yes. Set useOcr to true to read image-only or scanned pages with OCR. The OCR details in each row show the pages processed and any other signals the source reports.

Can I extract only some pages?

Yes. Enter page numbers or ranges in pages, such as 1,3-5. Leave it empty to read every page in each submitted PDF.

Can I submit more than one PDF?

Yes. Add one or more public PDF URL objects to pdfSources. The Actor returns a document result for each submitted source.

Can I upload a local or private PDF?

No. This input accepts public PDF URLs. It does not bypass login pages, paywalls, DRM, or other access restrictions.

Does it return tables or images as separate files?

No. The Actor returns extracted text and Markdown, plus document and page details. It does not extract structured table cells or image files.

📝 Changelog

v0.0 (30-09-2026)

• Initial release.

🆘 Support

For issues, questions, or feature requests, file a ticket and I'll fix or implement it in less than 24h 🫡

🔗 Related Actors

• Webpage Text Extractor - Extract public webpage text and Markdown before or alongside PDF work.
• Markdown to HTML Converter - Turn returned Markdown into HTML for a page or document.
• URL to BibTeX Converter - Create citation data from public paper and article URLs for research workflows.
• PDF Text Extractor — Markdown, OCR & Metadata - Compare another PDF text workflow with Markdown, OCR, and metadata.
• PDF Text Extractor - Multi-Column Layout, OCR & Metadata - Compare a PDF text workflow with multi-column layout, OCR, and metadata.

Made with ❤️ by Maxime Dupré