Extract text and tables from PDF

Does your PDF have text?

A PDF made from Word, a web page or any other program stores real text: you can select it, search it and copy it. A scan or a photo saved as PDF stores only pictures of the pages, and there is no text to extract until someone reads it. The preview tells you which kind you have: if it says no text layer was found, use image to text, which recognises the letters in scans, or PDF to Word, which does the same while rebuilding the document.

Reading order or layout

Inside a PDF, text is stored as pieces placed at positions on the page, not as paragraphs, and the order of the pieces in the file is often not the order you read them in. The extractor sorts the pieces by their position: top to bottom, and left to right within a line, joining pieces whose baselines line up.

• Reading order, the default, gives the lines in the order you read them, with a blank line wherever a larger gap marks a new paragraph.
• Keep layout keeps each line where it was, with spaces standing in for the gaps between columns. Use it for forms, invoices and anything where the position of a value on the line matters.

Pages with two or three columns of text are the hardest case for any extractor. If lines from neighbouring columns come out interleaved, convert with PDF to Word, which detects columns, and copy from there.

Tables

With Detect tables on, the extractor looks for text that sits in aligned columns over several lines, such as price lists, statements and timetables, and rebuilds each one as rows and cells. Minimum rows, 3 by default, sets how many lines must line up before something counts as a table; lower it for short tables, raise it if ordinary text is mistaken for one. Each table can be copied as CSV straight from the preview, and is saved as its own CSV file that opens in Excel, Google Sheets or Numbers.

Simple grids work best. Tables with merged cells, text wrapping inside cells or several tables side by side may need tidying in a spreadsheet afterwards.

The files you get

• Plain text (.txt), in UTF-8, so accented letters, Arabic and Chinese come through intact.
• One CSV per table, named after its number and page, such as table-01-page-3.csv, and marked so Excel reads accented letters correctly.
• JSON, off by default: every line of every page with its position and text size, for developers processing many PDFs.

For a quick look at a single page, there is no need to download anything: copy all the text or one table from the preview.

If something goes wrong

• No text layer was found. The PDF is a scan. Use image to text or PDF to Word with text recognition.
• The text is gibberish. Some PDFs use fonts with no proper mapping to letters, so their text cannot be read out. Text recognition is the way around it.
• Words run together or split apart. The PDF positions single letters rather than words. Try Keep layout, which keeps the original spacing.
• It says the PDF is password protected. Unlock it first with unlock PDF, if you have the right to.