Quick answer
Try to select a sentence with your cursor. If it highlights, the PDF has a real text layer and can be parsed directly. If nothing selects, it is a scan and needs OCR first.
How to tell which kind of PDF you have
- Text-based: text selects and copies, search finds words, the file is usually small for its page count
- Scanned: nothing selects, search finds nothing, the file is large because every page is an image
- Mixed: common in real documents — a digital report with scanned appendices or signed pages
- Deceptive case: a scan that already carries an invisible OCR layer will let you select text, but the selection lands in the wrong place
Extracting from a text-based PDF
- Parse the text layer directly, keeping reading order rather than page coordinates
- Handle tables and multi-column pages separately, or two columns get merged into one sentence
- Keep headers, footers, and footnotes out of the body text
- Convert to Word or Markdown first if you need to edit before translating
Extracting from a scanned PDF
- Run OCR to produce a text layer, then verify it before translating
- Flatten multi-layer scans with Flatten PDF so OCR does not run over existing hidden text
- Expect tables and formulas to need manual repair
Why extraction order matters for translation
Text pulled out in the wrong order gets translated in the wrong order, and no translation model can recover the intended meaning from scrambled input. Extraction quality caps translation quality — see why PDF translation breaks.
Bottom line
Identify the PDF type first. Everything downstream — extraction method, cost, and achievable quality — follows from that one answer.