How to Extract Text from a PDF (2026)
PDF was designed for fixed layout, not for text portability. That's why copying text from a PDF into a Word doc or an email so often produces a garbled mess — extra line breaks, missing spaces, jumbled column order, fi and fl ligatures showing up as question marks. Extracting text cleanly from a PDF is its own small discipline, and the right approach depends on whether your PDF has a text layer, how the original was structured, and what you plan to do with the extracted text. This guide covers both.
Two Types of PDFs, Two Extraction Paths
There are two fundamentally different kinds of PDFs when it comes to text extraction. A text-based PDF stores text as characters with positions, fonts, and styles — the same way a Word document does internally. Extracting from this type is fast, lossless, and produces clean output because the text is already there, just wrapped in PDF structure.
A scanned PDF is the opposite: it stores text as images with no underlying character data. Extracting text requires OCR — running the images through a recognition engine that reads pixels and outputs characters. This is slower and imperfect (in our OCR tests clean scans came back without errors, but a faint photocopy failed), and always requires a separate processing step.
The quick test: open your PDF, select a line of text with your cursor, and try to copy it. If clean text appears in the paste, you have a text-based PDF — use direct extraction. If nothing highlights or you get garbage, you have a scanned PDF — use OCR first via the PDF OCR tool, then extract from the OCR'd version.
Method 1: UDT PDF Text Extractor (Free, Browser-Based)
The UDT PDF Text Extractor pulls text from any text-based PDF directly in your browser using pdf.js. Upload (locally — nothing leaves your browser), and the tool outputs plain text ready to paste anywhere: email, Word, a terminal, another document. It keeps line breaks where they appear on the page and marks where each page starts.
The workflow: drop your PDF into the tool, choose output format (plain text, JSON with positional data, or Markdown), and download or copy. For scanned pages it reports that no text was found and points you to the PDF OCR tool; you can't extract text that isn't there.
Columns used to be where it fell short: on a two-column test page each output line joined the left and right columns ("L01 ... R01 ..."), as pypdf also did, while poppler's pdftotext in its default mode read the left column first. We added column detection: when an empty gutter runs down the middle of the page and both sides hold real lines of text, it now reads the left column, then the right, with a title across both columns placed first. Tables and right-aligned dates, which also leave gaps, stay row by row; all seven of our column and page-range test cases now pass.
Method 2: Acrobat and Other Alternatives
Acrobat Reader (free) can export PDF text via File → Export To → Text. Acrobat Pro adds Save As → More Options → Text (Plain) with additional formatting controls. Both handle text-based PDFs well. For scanned PDFs, Acrobat Pro runs OCR automatically during export; Reader doesn't.
Microsoft Word (2013+) can open PDFs directly, treating them as editable documents. Word's conversion is often the best-in-class for preserving formatting (fonts, tables, images), though simple text extraction is overkill if you just need the words. Google Docs has similar PDF import capabilities.
Command-line tools like pdftotext (from Poppler) are the workhorses for automated extraction. pdftotext handles layout preservation well, supports columns, and is what most document-processing pipelines use under the hood. Install via Homebrew on macOS, apt on Linux, or prebuilt binaries on Windows.
Handling Tables, Columns, and Layout
Tables are the hardest part of text extraction. A PDF table is just text positioned in a grid — there's no underlying "this is a table" structure the way there is in HTML or Word. Extraction tools guess at table structure by looking for aligned text runs, which works for clean tables and fails for complex ones with merged cells, nested columns, or inconsistent spacing.
For critical table extraction (financial data, scientific tables, anything where getting the structure right matters), specialized tools like Tabula or Camelot outperform general text extractors. They detect table boundaries, infer columns from x-coordinate clustering, and output CSV or JSON directly. General extraction tools give you text in reading order; table-specific tools give you rows and columns.
Multi-column layouts (magazines, academic papers, brochures) need column-aware extraction. A naive tool reads left-to-right across all columns on each row, producing nonsense. A better tool detects column boundaries and reads each column top-to-bottom in sequence. If your extracted text looks like interleaved fragments, you're probably using a tool that doesn't do column detection — switch to one that does.
Common Pitfalls and Cleanups
Ligatures are the most common complaint. PDFs often render "fi" and "fl" as single glyphs to look better typographically, and some extractors output these as question marks, boxes, or nothing at all. Modern extractors (including pdf.js-based tools) handle ligatures correctly; older ones or command-line defaults may not. If your output has missing letter-pairs, check ligature handling in your tool's settings.
Hyphenated line breaks are another trap. When a word wraps across lines as "how- ever" at the end of one line and start of the next, a careless extractor keeps the hyphen and newline in the output, producing "how- ever" in your text. Good extractors rejoin these automatically; if yours doesn't, a simple find-and-replace for "- " fixes most cases.
Header and footer repetition is the third common issue. Page numbers, running headers, and footers appear on every page, and naive extraction includes them in the output — producing "Page 1\nThe quick brown fox jumped...\nPage 2\n...over the lazy dog." Either skip headers and footers in the tool settings or strip them with a regex after extraction.