How the text is put back in reading order
Inside a PDF, text is not stored as lines and paragraphs. Each word or piece of a word is placed at a position on the page, often in the order the program happened to draw it. The tool reads every piece with pdf.js, groups pieces that share a baseline into a line, puts the lines in order from top to bottom and adds a space wherever there is a visible gap between two pieces. A wide gap, as between table columns, becomes several spaces, so tables stay readable. A clearly larger gap between lines becomes a blank line, which keeps paragraphs apart. Pages turned with the rotate setting, and text drawn sideways, are read in their own direction.
Arabic and mixed Arabic-English PDFs
An Arabic line is read from right to left, but an English name, an email address, an IBAN or a phone number inside it keeps its own left-to-right order, and two numbers next to each other stay in the order they were written. Many Arabic PDFs store the joined letter shapes (presentation forms) instead of normal letters; the tool turns them back into ordinary Arabic letters, so you can search and edit the text. Brackets that some PDFs store reversed are put back the right way round. In our tests on an Arabic invoice and a 112-page official gazette, 98% or more of the Arabic words that Poppler's pdftotext found were found here too, in reading order. Some designed PDFs draw two letters as one shape (for example في); in those files the two letters can come out swapped. A PDF whose font maps letters to the wrong codes gives wrong letters in every tool, ours included.
Scanned pages and photos of documents
A scan or a phone photo saved as PDF is a picture of the page, with no text inside it. The tool checks every page that gives no text: if the page holds an image, it is listed as a scanned page, so you know why that part of the result is empty; a page with nothing on it is simply skipped. Getting text out of pictures needs optical character recognition (OCR), which is a different job: open the same file in Image to Text (OCR), which reads scanned pages in the browser too. A PDF that has a scan with an invisible text layer added by a scanner app is read like any other text PDF.
Page ranges, page markers and the .txt file
By default every page is read. To take only part of the file, type pages and ranges separated by commas, such as 1-3, 8 or 10- for page 10 to the end; Arabic digits work too. With page markers on, each page starts with a line such as "--- Page 4 ---", which helps when you quote or check the text later; with them off, pages are separated by a blank line. The download is UTF-8, which every current editor reads correctly. The BOM option adds three invisible bytes at the start of the file so that old versions of Windows Notepad recognise UTF-8 and show Arabic properly.
Privacy and honest limits
The PDF is read by this browser tab and is never uploaded, so contracts, payslips and bank letters stay on your device. A password-protected PDF has to be unlocked first with Unlock PDF. Text in two columns side by side is read across both columns line by line, and table cells come out as text separated by spaces, not as a spreadsheet. Files up to 150 MB work on most computers. If you need the content in an editable document with its layout, use PDF to Word instead. For a scanned PDF, image to text reads the pages with OCR.