Getting a better conversion out of an awkward PDF

Converting a PDF is guesswork and sometimes the guesses are wrong. Most bad results come down to a handful of causes, and you can fix several of them.

The book is empty, or nearly

Almost certainly a scan. A scanned PDF is a stack of photographs. There’s no text in the file, only pictures of text, so there’s nothing to pull out.

Run it through OCR first, which we don’t do here. OCRmyPDF is free if you’re comfortable on the command line; Acrobat has it built in, and so do a lot of document scanners. Add a text layer that way, then convert the result.

We check for this before converting and say so, so it shouldn’t catch you out.

Every paragraph runs into the next

Usually a PDF that sets its lines tightly, with no extra space between paragraphs and no first-line indent. There’s then no visual signal that a paragraph has ended. The converter can’t spot one because there isn’t one to spot.

Not much to be done at conversion time. If the document exists as Word, HTML or LaTeX source anywhere, convert from that instead. It’ll beat the PDF every time.

Chapters are missing or in the wrong places

Chapters come from the PDF’s bookmarks when it has them. When it doesn’t, we guess from the headings, and that goes wrong on documents whose headings don’t look much different from the body text.

Two things to try:

• If we tell you the PDF has no bookmarks, expect the chapters to be roughly right at best. Adding proper bookmarks in a PDF editor first gives you an exact result.
• Or turn Split into chapters off entirely. You get one long document with no table of contents, which often beats a wrong one. Search and the progress bar still work.

Page numbers and headers appear mid-sentence

The Remove page numbers and running heads option deals with this by spotting text that repeats in the top or bottom margin across a lot of pages. It needs that repetition to be fairly consistent, so a document that alternates headers between odd and even pages, or switches them at every chapter, can hang onto a few.

If yours has an unusual header scheme and bits survive, convert with the option off and clean up afterwards in Calibre or Sigil. Tedious, but reliable.

The text is scrambled or interleaved

Probably a two-column layout being read straight across. We look for a clear blank corridor down the middle and read the columns in order, but pull quotes, sidebars or pictures spanning both columns break that. At which point there often isn’t one correct reading order to find anyway.

Magazines and heavily designed documents just aren’t good candidates. Read those as PDFs, in landscape.

Images are missing

A few possibilities:

• Include images was switched off.
• They’re vector drawings, not photographs. Vectors are drawing instructions rather than pictures, and we don’t extract them.
• They got filtered as page furniture. Anything repeating across most pages is treated as a logo or watermark and binned.
• They were too small. Under roughly an inch we assume it’s a bullet or a rule.

The result is fine but the title is wrong

Title and author come from the PDF’s metadata, which is very often a mess. Plenty of PDFs carry the name of whatever tool spat them out. Both fields are editable before you convert, and they’re what your device sorts by, so it’s worth the ten seconds.

When to give up on the PDF

If the document exists anywhere else in any other format, start there. A PDF is the worst place to convert a book from. It’s a print format, and everything we do here is an attempt to undo the flattening it applied. EPUB, HTML, Word, even plain text will all give you a better book.