All Tools

A PDF does not store paragraphs the way a word processor does. It stores text as a set of glyph runs, each with a font reference and coordinates placing it on the page. The tool uses PDF.js to open your file, then asks each page for its text content, which returns those runs together with their positions.

The extraction step reassembles them into readable text. Runs are ordered by position so that words appear in reading order, and the vertical coordinates are used to decide where one line ends and the next begins, which is what turns a scattered set of fragments back into sentences and paragraphs. The process repeats page by page until the whole document has been read, with progress shown as it goes.

If a page returns no text items at all, that is a reliable signal the page holds only images or vector drawings. The tool detects this case across the document and tells you directly that the file needs optical character recognition rather than text extraction.

Take a 24-page annual report where page 11 holds a revenue table you need in a spreadsheet. Select the PDF and the tool works through all 24 pages, reporting progress, then displays the complete text with page boundaries preserved.

Scroll to the page 11 section and you will find the table rendered as lines of text, with the cell contents in reading order across each row. A row that appeared in the PDF as Revenue, 4,820, 5,140, 6,005 comes back as those four values on one line. You can copy the whole extraction to the clipboard and paste the relevant block into a spreadsheet, then use a text-to-columns split to rebuild the grid.

The important limitation shows here too: column boundaries are not preserved as structure, only as spacing, so a table with merged cells or wrapped text inside a cell will need manual tidying. For a straightforward numeric table of the kind above, the result usually needs only a single split step to become usable.

Text-based PDFs produced by Word, LaTeX, InDesign or an export from a web page extract almost perfectly. What causes trouble is layout the extraction cannot infer. Multi-column pages may interleave columns depending on how the original was laid out, tables lose their grid structure, headers and footers appear inline with body text on every page, and hyphenation at line ends is preserved as written.

Scanned documents are the clear boundary. If someone photographed or scanned a page, the PDF contains a picture with no text underneath, and there is nothing to extract. The tool will tell you this rather than returning blank output. In that case you need OCR software, which recognises characters in the image, and results there depend heavily on scan quality.

Some PDFs carry permission flags restricting copying, and password-protected files must be unlocked before they can be read. Always proofread extracted text before reusing it, particularly numbers, and respect the copyright of the source document.