How does browser extraction read digital and scanned PDFs?
The extractor reads your PDF locally in memory using pdf.js for digital text layers and switches to self-hosted Tesseract OCR whenever a page contains only scanned images. Because all parsing and character recognition execute inside your browser tab, 0 bytes leave your device.
Reading embedded text streams with pdf.js
As soon as you drop a document, pdf.js opens the file and walks through every page sequentially. For digitally authored PDFs, it calls getTextContent() to collect text runs along with end-of-line markers in reading order.
Extra trailing spaces and runs of three or more consecutive newlines are normalized so the preview stays clean. A 50-page digital report typically finishes reading in moments without any manual button click before the preview appears.
When a page mixes selectable text with embedded screenshots, the extractor inspects the existing character count first. If a page has fewer than 24 non-whitespace characters, it supplements or replaces that sparse layer with OCR output.
Automatic local OCR fallback for scanned pages
When a page has little or no embedded text, the tool automatically triggers optical character recognition powered by the Tesseract OCR Official Documentation engine served locally from this site. Each scanned page is rendered onto a canvas capped at 4,000 pixels on a side and 16,777,216 total pixels.
Small scans narrower than 1,600 pixels are upscaled up to 3x before recognition so thin characters read more reliably. Words scoring below a confidence threshold of 30 are filtered out, and a notice tells you how many pages used OCR so you can verify names and figures. If you are starting from a screenshot or camera photo rather than a PDF, use browser-based image to text directly.
Controlling line breaks, paragraph wrapping, and word counts
PDF documents position every line at fixed coordinates rather than storing flowing paragraphs like a word processor. That difference explains why raw copy-paste from a PDF viewer often breaks sentences mid-line.
Keeping PDF line breaks versus joining paragraphs
By default, Keep line breaks as in the PDF is checked so every end-of-line marker from the page layout stays intact. That mode works well for poetry, addresses, code snippets, or numbered lists where each line should stay separate.
Unticking Keep line breaks as in the PDF joins single line breaks inside a paragraph into spaces while keeping blank-line paragraph breaks intact. Because the raw per-page text array stays cached in memory, toggling the checkbox updates the output instantaneously without re-reading the PDF.
This unwrapping behavior is especially useful when pasting academic articles or legal briefs into an email or translation tool. Sentences no longer split awkwardly at 70-character column edges.
Checking the live Words found counter
The options panel displays a live Words found count computed directly from the active output text. Watching that figure helps confirm that a scanned contract or multi-page appendix actually yielded readable text before you copy it.
If you need deeper metrics such as character counts with and without spaces, sentence totals, or estimated reading speed, paste the extracted text to check exact character and reading-time metrics. That workflow is handy for editors verifying manuscript limits.
Plain .txt output versus structured Markdown and table layouts
Plain text strips away bold styling, font sizes, headings, and grid borders, leaving universal character sequences as defined in the Library of Congress UTF-8 Plain Text Format Specification. Choosing the right output format depends on where the text is going next.
| Output format | Keeps headings and lists | Keeps tables | OCR on scans | Best destination |
|---|---|---|---|---|
| Plain .txt | No | No | Yes | Email, grep, diffs, raw notes |
| Markdown .md | Yes | Yes (pipe tables) | Yes | AI chats, Obsidian, GitHub |
| Word .docx | Yes | Yes | Yes | Office editing and redlines |
| CSV / Excel | No | Yes (rows/columns) | No | Spreadsheets and accounting |
Because plain text has no column delimiters, multi-column newsletters and financial tables collapse into sequential reading lines. When you need headings, bulleted lists, and aligned data tables preserved for notes or AI prompts, export headings, lists, and tables as Markdown instead.
Plain UTF-8 files remain the safest format for long-term readability because any operating system, terminal, or code editor can open .txt without proprietary software. Every character is stored without hidden styling tags or binary wrappers.
Copying to clipboard, saving UTF-8 files, and troubleshooting
Once the extracted text appears in the preview box, you have two ways to take it with you. Clicking Copy text writes the full string to your system clipboard via the MDN Web Docs Clipboard.writeText() API and flashes Copied for 1,600 ms.
Clicking Save .txt creates a local Blob URL encoded as text/plain;charset=utf-8 and downloads <filename>.txt straight to your device. After editing or cleaning up the extracted draft, you can turn edited plain text back into a printable PDF with clean margins.
Tip: If a scanned PDF returns an error saying no text could be read, the scan may be too faint, skewed, or low-resolution for OCR. Re-scanning at 300 DPI or higher contrast usually resolves recognition failures.
Documents protected by an open password cannot be decoded until unlocked, while PDFs that merely restrict printing or copying in desktop viewers still open normally here when no open password is set. You can click Change file at any time to clear the current document from memory and load another PDF.
Because each extracted file is encoded in UTF-8, accented characters, currency symbols, and non-Latin scripts transfer cleanly into modern editors. You do not need to worry about legacy code-page garbling when opening the downloaded .txt file.