Get something else out of a document

Glyphs at coordinates. A PDF records that a particular character from a particular font was drawn at a particular position, and nothing about what that character was for. There is no heading in the file, no list, no table and no paragraph — those are things a reader infers from size, weight, indentation and alignment, and it is what these converters infer too. The signals are strong and usually right; check headings and tables before you rely on them, which is why the Markdown tool shows you its output before it offers a file.

Take Word when somebody has to edit the text: you get real Word heading styles, real lists, tables where the columns genuinely line up, and bold and italic within a line. Take Excel when the document IS a table and the point is the numbers — that converter finds columns from the empty corridors between them rather than from where the text starts, which is the difference between a column of amounts and three columns of fragments. Take Markdown when the text is going into a notes app, a repository or another tool. Take images when the appearance is the point — a page as a picture is the only conversion that keeps the layout exactly, and it is the one that gives up the text.

A scanned page holds a picture of text and no characters at all, so there is nothing for a converter to extract and no way to recover a character that was never in the file. Reading letters out of an image is optical character recognition, and none of the converters in this group performs it — they name the pages that have no text layer and produce no file, rather than hand you an empty document that looks like its own bug. What the site does have is a tool for exactly that, in one step: OCR a PDF draws each page and recognises it.