How to tell if a PDF is scanned

A PDF can look like paper and still contain real text, or look clean and still be only a high-resolution scan. The first test is simple: open the file and try to select a word inside a paragraph. If selection snaps to characters and copies clean text, there is a text layer. If selection draws a rectangle or copies nothing, the page is probably image-only.

This is only a first check. Some OCR layers are misaligned, invisible, or incomplete. You may be able to select text but get nonsense characters because the PDF has broken font encoding. Conversely, a viewer can prevent selection even when text exists. Use command-line checks when the result matters.

pdftotext in.pdf - | wc -w estimates how many extractable words exist. A result near zero for a multi-page document strongly suggests an image-only scan. A result with many words means some text layer exists, but it does not prove the text is accurate or in reading order.

pdffonts in.pdf lists fonts used by text objects. If it reports no fonts, the file may be scanned or may use unusual drawing commands. If fonts exist but pdftotext is empty, the PDF may have text encoded in a way extraction cannot map back to Unicode.

Scanned PDFs often contain one large image per page. PDF inspection tools can show image objects, dimensions, and compression. A 300 DPI letter-size scan may have page images around 2550 by 3300 pixels. A born-digital report usually has many text and vector drawing objects instead.

Image-only pages explain why search, copy, and PDF-to-Word conversion perform poorly. The converter cannot recover paragraphs from vector text because there are none. It must run OCR on page images and then infer layout.

Many real documents are mixed. A bank statement might have born-digital account text plus a scanned signature page. A court filing might contain searchable cover pages and scanned exhibits. Testing only page one can give the wrong answer.

When diagnosing, sample several pages: the first page, a dense middle page, a page with tables, and any appendix. If only some pages lack text, run OCR only where needed when the tool supports it. That reduces processing time and avoids disturbing good text.

If the PDF has a good text layer, convert to TXT, DOCX, or Markdown with a text-based tool first. If it is scanned, use OCR before expecting editable output. ocrmypdf --deskew --rotate-pages in.pdf out-searchable.pdf can add a searchable text layer while keeping page images.

After OCR, verify a few copied passages, not just searchability. OCR can produce words that look plausible but are wrong, especially for low-resolution scans, stamps, handwriting, and skewed tables.

For intake automation, record the diagnosis with the file. A label such as image-only, text-layer, or mixed explains why one file went through OCR and another went straight to text extraction. It also helps support users who expected editable Word output from a faxed scan.

Do not judge scan status by file size alone. A compressed image-only PDF can be small, and a born-digital PDF with embedded fonts and images can be large. Inspect structure and text extraction rather than relying on storage size.