PDF Text Extractor | OCR, Markdown & JSON

Extract text from public PDF links and receive clean, structured results ready for research, search, archiving, document analysis and AI workflows.

The Actor supports normal digital PDFs and scanned documents. It can return plain text, Markdown or structured JSON, identify pages that may need OCR, highlight pages that look like tables, and optionally include page-level text blocks and coordinates.

Pricing at a glance

You are charged by the pages actually processed, not by the number of PDF files and not by the total number of pages reported by the document.

Processed page	Price
Page processed without OCR	$0.001
Page processed with OCR	$0.002

There is no separate charge for starting a run and no fixed charge per PDF.

Examples:

Work completed	Actor charge
1 normal page	$0.001
1 OCR page	$0.002
100 normal pages	$0.100
80 normal pages + 20 OCR pages	$0.120
1,000 normal pages	$1.000
1,000 OCR pages	$2.000

Each processed page is charged exactly once. If OCR is enabled and a page is read with OCR, that page uses the OCR price instead of the normal price. A failed download or a PDF that cannot produce a result does not create page charges.

You can also set Maximum cost per run in Apify run options. The Actor uses the actual processed-page count for billing and stops publishing new results when the configured charge limit is reached.

What this Actor does

For every direct PDF URL, the Actor:

1. Downloads the PDF from the supplied public link.
2. Reads its text and document structure.
3. Detects pages with little or no selectable text.
4. Uses OCR on those pages when OCR is enabled.
5. Returns the selected output format in the Dataset.
6. Reports the pages processed, pages flagged for OCR and pages that may contain tables.

The Actor processes only the PDF links you provide. It does not crawl links inside a document, create embeddings, build a vector database or automatically create a complete RAG system.

Best use cases

• Convert research papers, reports and public studies into searchable text.
• Prepare Markdown documents for AI assistants and knowledge bases.
• Extract text from manuals, regulations and technical documentation.
• Process scanned reports, faxes and image-based documents with OCR.
• Convert public product catalogues and brochures into structured data.
• Inspect page layout, text blocks and coordinates.
• Build document archives and search indexes.
• Process several PDF files in one run while keeping one result per file.

Quick start

1. Open the Actor's Input tab.
2. Add one or more direct PDF links.
3. Choose Markdown, JSON or Plain text.
4. Leave OCR enabled when the document may contain scans or image-only pages.
5. Set maxPages if you want to limit processing and page charges.
6. Launch the run.
7. Open the Dataset tab to inspect or download the results.

The default configuration is suitable for a first test: Markdown output, OCR enabled, a maximum of 50 pages per PDF and a maximum file size of 100 MB.

For integrations and automation, optional fields can be omitted when you want to use these defaults. Do not send null for an optional field; use a valid value or leave the field out of the input.

Input reference

Field	Type	Required	Default	Limits and description
pdfUrls	Array of strings	Yes	—	1–50 direct http or https PDF links. Duplicate URLs are processed only once.
outputFormat	String	No	markdown	markdown, json or text.
ocrEnabled	Boolean	No	true	Read pages flagged for OCR. When disabled, processed pages use the normal page price.
ocrLanguage	Select	No	eng	Choose one of the 20 supported OCR languages from the dropdown.
maxPages	Integer	No	50	1–5,000 pages per PDF. Only pages actually processed count toward billing.
includePageData	Boolean	No	false	Adds detailed page blocks, coordinates and layout data to the result.
maxDownloadMb	Integer	No	100	1–500 MB maximum file size per PDF. Larger files are rejected.

pdfUrls

Use a direct link that returns a PDF file, for example:

https://example.org/documents/annual-report.pdf

A page containing an embedded PDF viewer is not a direct PDF link and cannot be processed. The URL must be accessible without a login or personal browser session.

You can submit up to 50 URLs in one run. Duplicate URLs are processed only once. Empty or invalid URLs are rejected before processing.

outputFormat

• markdown: recommended for AI workflows, search and human-readable documents. Headings, lists and other detected structure are preserved when available.
• json: recommended for structured automation. The content field contains the parser's structured JSON result.
• text: recommended when you only need the extracted text as a string.

ocrEnabled and ocrLanguage

OCR is useful for scans, faxes and image-only pages. The Actor analyses every processed page and exposes candidates in ocrPages and pages[].needsOcr.

When ocrEnabled is true, pages identified as requiring OCR are processed with OCR and use the $0.002 page price. Other processed pages use $0.001.

When ocrEnabled is false, no OCR is run. All processed pages use the $0.001 normal page price, even if the analysis reports that a page could benefit from OCR. The page remains visible in ocrPages so you can identify where OCR would have been useful.

The Apify form provides a controlled dropdown, so you do not need to type a language code manually. The available choices are:

Code	Language
eng	English
spa	Spanish
fra	French
deu	German
ita	Italian
por	Portuguese
nld	Dutch
pol	Polish
rus	Russian
ukr	Ukrainian
ara	Arabic
hin	Hindi
jpn	Japanese
kor	Korean
chi_sim	Chinese (Simplified)
chi_tra	Chinese (Traditional)
tur	Turkish
vie	Vietnamese
tha	Thai
heb	Hebrew

For multilingual documents, choose the language that appears most often. OCR accuracy depends on scan resolution, contrast, rotation, font quality and the selected language.

maxPages

maxPages is applied separately to every PDF. If a document has 1,000 pages and maxPages is 50, the Actor processes at most 50 pages and can create at most 50 page charges for that PDF.

The totalPages field shows the total page count reported by the PDF. It is informational. The processedPages field shows the number used for billing.

includePageData

Keep this option disabled when you only need extracted text. Enable it when you need detailed information about text blocks, positions, annotations, forms or layout.

Page-level details can make the Dataset considerably larger, especially for long documents. They do not create an additional per-page event charge; billing still follows processed pages.

maxDownloadMb

This limit protects the run from unexpectedly large files. A PDF above the configured limit is reported as an error and does not produce page charges.

Input examples

Recommended Markdown extraction

{
  "pdfUrls": [
    "https://example.org/reports/annual-report.pdf",
    "https://example.org/reports/market-study.pdf"
  ],
  "outputFormat": "markdown",
  "ocrEnabled": true,
  "ocrLanguage": "eng",
  "maxPages": 50,
  "includePageData": false,
  "maxDownloadMb": 100
}

Spanish scanned document

{
  "pdfUrls": [
    "https://example.org/documentos/informe-escaneado.pdf"
  ],
  "outputFormat": "markdown",
  "ocrEnabled": true,
  "ocrLanguage": "spa",
  "maxPages": 100,
  "includePageData": false,
  "maxDownloadMb": 200
}

Structured JSON with page details

{
  "pdfUrls": [
    "https://example.org/manuals/product-manual.pdf"
  ],
  "outputFormat": "json",
  "ocrEnabled": true,
  "ocrLanguage": "eng",
  "maxPages": 25,
  "includePageData": true,
  "maxDownloadMb": 50
}

Low-cost text extraction without OCR

{
  "pdfUrls": [
    "https://example.org/publications/digital-document.pdf"
  ],
  "outputFormat": "text",
  "ocrEnabled": false,
  "maxPages": 20,
  "includePageData": false,
  "maxDownloadMb": 25
}

Dataset output

The Actor normally creates one Dataset item for each PDF URL. Successful items have status: "ok". Failed files have status: "error" so one unavailable URL does not hide successful results from the same run. If a run-level charge limit is reached before a result can be published, that result is reported as not_charged in SUMMARY and is not added to the Dataset.

Successful result fields

Field	Description
status	ok when the PDF was processed successfully.
sourceUrl	The URL supplied in pdfUrls.
finalUrl	The final URL after redirects, when it differs from sourceUrl.
filename	A safe filename generated from the PDF URL.
sha256	SHA-256 hash of the downloaded PDF file.
fileSizeBytes	Size of the downloaded PDF in bytes.
format	markdown, json or text.
totalPages	Total pages reported by the PDF parser. This is not the billing count.
processedPages	Pages actually processed. This is the page count used for billing.
textChars	Number of extracted text characters.
ocrPages	Page numbers flagged as candidates for OCR by the page analysis.
tableLikelyPages	Page numbers that appear likely to contain tables. This is a detection hint, not a guarantee of perfect table extraction.
pages	Compact page-by-page analysis for the processed pages.
durationSeconds	Download and processing time for this PDF.
content	Extracted text, Markdown or structured JSON, depending on format.
pageData	Detailed raw page information when includePageData is true.

Page analysis fields

Every entry in pages describes one processed page:

Field	Description
page	Page number, starting at 1.
textChars	Characters extracted from that page.
needsOcr	Whether the analysis considers OCR useful for the page.
reasons	Reasons behind the OCR recommendation, when available.
tableLikely	Whether the page appears to contain a table.
columnCount	Estimated number of text columns.
contentBounds	Bounding box of detected page content, when available.

ocrPages and needsOcr describe the document analysis. They do not mean that OCR was executed when ocrEnabled is false.

content by format

• With markdown, content is a Markdown string.
• With text, content is a plain text string.
• With json, content is a structured object containing the parser's JSON representation, including page content and extracted text when available.

Example result

The following is an illustrative successful Dataset item. Hashes, timings, character counts and page classifications vary by document.

{
  "status": "ok",
  "sourceUrl": "https://example.org/reports/annual-report.pdf",
  "finalUrl": "https://example.org/reports/annual-report.pdf",
  "filename": "001_annual-report.pdf",
  "sha256": "7f4d8b7a8c0f3a1a4d2c8f9e7b6a5d4c3b2a190817161514131211100f0e0d0c",
  "fileSizeBytes": 284731,
  "format": "text",
  "totalPages": 12,
  "processedPages": 3,
  "textChars": 8421,
  "ocrPages": [2],
  "tableLikelyPages": [3],
  "pages": [
    {
      "page": 1,
      "textChars": 3120,
      "needsOcr": false,
      "reasons": [],
      "tableLikely": false,
      "columnCount": 1,
      "contentBounds": [55.2, 72.1, 488.4, 690.7]
    },
    {
      "page": 2,
      "textChars": 0,
      "needsOcr": true,
      "reasons": ["no-text"],
      "tableLikely": false,
      "columnCount": 1,
      "contentBounds": [48.0, 65.0, 500.0, 710.0]
    },
    {
      "page": 3,
      "textChars": 5301,
      "needsOcr": false,
      "reasons": [],
      "tableLikely": true,
      "columnCount": 2,
      "contentBounds": [48.0, 65.0, 500.0, 710.0]
    }
  ],
  "durationSeconds": 1.84,
  "content": "Annual report\\n\\nRevenue increased during the reporting period..."
}

When outputFormat is json, the same top-level structure is returned but content is an object instead of a string. When includePageData is enabled, the item also contains detailed raw information for every processed page.

Error result example

If a URL cannot be downloaded or the PDF cannot be processed, the Dataset item has an explicit error status:

{
  "status": "error",
  "sourceUrl": "https://example.org/reports/missing.pdf",
  "filename": "001_missing.pdf",
  "errorType": "HTTPStatusError",
  "error": "Client error '404 Not Found' for url",
  "durationSeconds": 0.42
}

Error text can vary by server and document. Failed files do not produce page charges.

Run summary

The Actor also saves a SUMMARY record for the run. It is useful when several PDFs are submitted together:

{
  "status": "partial",
  "processed": 1,
  "failed": 1,
  "total": 2,
  "format": "markdown",
  "ocrEnabled": true,
  "maxPages": 50,
  "durationSeconds": 3.18,
  "items": [
    {
      "sourceUrl": "https://example.org/reports/annual-report.pdf",
      "status": "ok",
      "totalPages": 12,
      "processedPages": 3,
      "textChars": 8421,
      "durationSeconds": 1.84
    },
    {
      "sourceUrl": "https://example.org/reports/missing.pdf",
      "status": "error",
      "totalPages": null,
      "processedPages": null,
      "textChars": null,
      "durationSeconds": 0.42
    }
  ]
}

Summary statuses:

• ok: every submitted URL produced a successful result.
• partial: at least one PDF succeeded and at least one failed.
• error: no PDF produced a successful result, or processing stopped before a result could be published.

Understanding billing with the output

Use these fields to reconcile a run:

1. processedPages is the total number of pages used for billing for that PDF.
2. With OCR enabled, count pages where needsOcr is true for the OCR portion.
3. The remaining processed pages use the normal page price.
4. With OCR disabled, all processedPages use the normal page price.

For example, a result with 80 normal pages and 20 OCR pages costs:

(80 × $0.001) + (20 × $0.002) = $0.120

The totalPages value is never used instead of processedPages. This is why a 1,000-page PDF limited to 50 pages cannot generate 1,000 page charges.

Important limitations

• The URL must be a direct, accessible PDF link.
• Login-protected, private, encrypted or password-protected PDFs may fail.
• The Actor does not use a personal browser session to access documents.
• OCR quality depends on the original scan and selected language.
• tableLikelyPages is a layout signal, not a promise of cell-perfect table extraction.
• Complex figures, handwriting, unusual fonts and heavily damaged scans may produce incomplete text.
• includePageData can make outputs much larger for long documents.
• The Actor does not follow links inside PDFs or process non-PDF web pages.

Frequently asked questions

Is pricing per PDF or per page?

Per processed page. There is no fixed PDF charge and no separate run-start charge.

What happens with a 1,000-page PDF?

Set maxPages to the amount you want to process. For example, with maxPages: 50, at most 50 pages are processed and billed. Without that limit, the charge follows the pages actually processed: $1.00 for 1,000 normal pages or $2.00 for 1,000 OCR pages.

Does every page cost the OCR price when OCR is enabled?

No. Only pages actually processed with OCR use $0.002. Other processed pages use $0.001.

What if I disable OCR?

The Actor processes pages without OCR and charges every processed page at $0.001. Pages that could benefit from OCR remain visible in ocrPages and needsOcr.

Why can totalPages be greater than processedPages?

totalPages is the complete page count reported by the PDF. processedPages is limited by maxPages and by the pages the Actor actually completed.

Does includePageData cost extra per page?

No additional page event is created. It can increase output size and processing work, so use it only when coordinates and detailed layout data are useful.

What happens if one URL fails in a batch?

The Actor keeps successful results, adds an error item for the failed URL and marks the run summary as partial when appropriate.

Can I submit a PDF viewer URL?

No. Submit the URL that downloads the .pdf file itself, not the page that embeds or displays the document.

Can the Actor create embeddings or a vector database?

No. It returns extracted content. You can download the Dataset and pass the text or Markdown to the search, indexing or AI system of your choice.

How can I keep a large run within budget?

Use maxPages, choose OCR only when needed, keep includePageData disabled unless required, and set Maximum cost per run in Apify run options.

Responsible use

Process documents that you are allowed to access and use. Do not submit confidential or personal documents unless your handling of that information is authorized and appropriate. Respect the terms of the website hosting each PDF and any applicable copyright, privacy and data-protection requirements.