My PDF finds some words but not others. Why?

You almost certainly have a mixed document — some pages exported from software, others scanned and appended. Run OCR over the whole file; pages that already have text are unaffected in a well-behaved workflow, but check the result before discarding the original.

Search finds nothing even though the text is selectable.

The text layer may use a broken character encoding, which happens with some older generators and with certain CJK or Arabic embeddings. Extracting to plain text will show you the garbled characters, and re-OCR from a rendered image is the usual fix.

How to tell whether a PDF contains real text or just pictures

Try to select a word. If the cursor sweeps across the page without highlighting anything, you have a scan and need OCR before anything else will work.

Before you convert, search, compress or extract anything, answer one question: does this file contain characters, or photographs of characters? Almost every "the tool did not work" report traces back to skipping this check.

Check 1: try to select a word

Open the PDF in any viewer and drag the cursor across a line of text.

A selection highlight follows individual words — the file has a real text layer. Everything downstream will work normally.
A single translucent rectangle covers the whole page area, or nothing highlights at all — the page is an image. There are no characters to select because none exist in the file.

This takes two seconds and is right the overwhelming majority of the time.

Check 2: search for a word you can see

Use the viewer's find function and type a word visible in the first paragraph. No match on a word you are literally looking at is conclusive: there is no text layer, or the encoding is broken.

Do this on more than one page. Mixed documents are common — a report exported from software with a scanned, signed final page bolted on — and a single-page check will mislead you.

Check 3: extract the text

The definitive test is to pull the text out and read it. Run the file through PDF to Text and open the result.

Clean, readable text — the document is born-digital and in good shape.
An empty or near-empty file — no text layer. It is a scan.
Text present but scrambled, full of question marks or wrong letters — there is a text layer, but its character encoding is unreliable. This happens with some older PDF generators and with fonts embedded without proper Unicode mapping. Treat it like a scan: re-recognise from the rendered pixels.
Readable text in the wrong order, with columns interleaved — the text is fine, but the reading order has to be reconstructed from geometry and the layout defeated it. Extraction and conversion results will need manual reordering.

What each outcome means for your next step

You have real text. Search, copy, conversion to Word, text extraction and text-safe compression all behave. Do not run OCR — it would gain you nothing and risk replacing accurate characters with recognised guesses.

You have a scan. Nothing text-based will work until you run OCR on the PDF. Recognition adds an invisible text layer behind the existing page images, so the document looks identical but becomes searchable and selectable. Accuracy depends heavily on the scan itself — see how to get better OCR results.

You have a mixture. Decide per section. If most of the file is born-digital and only two appended pages are scans, it is often cleanest to extract those pages, OCR them, and merge the document back together.

A note on file size as a clue

Size is a useful secondary signal. A 40-page text document that weighs 300 KB is certainly born-digital. A 12-page document weighing 25 MB is almost certainly scanned images. It is not proof — a born-digital brochure full of photographs is also large — but combined with the selection test it gives you a confident answer in seconds.

Why this matters for privacy too

A scanned page hides nothing in a text layer, but it can hide plenty in the image: over-wide margins that captured the next document on the platen, or a low-contrast area that looks blank on screen. Conversely, a born-digital PDF can hold text underneath a black rectangle. Both cases are covered in what is hidden inside a PDF.