Distinguish an image-only PDF from a PDF with a text layer

Where the extra work starts

Identify which you have, then test the actual output so selecting a tool does not depend on appearance alone.

ParseMill processes native invoice PDFs and clear printed English PDF scans. Try a typical supplier invoice and a difficult one. Check the extracted fields, correct exceptions and open the workbook before committing to a monthly plan.

Do not assume that all pages in one PDF share the same construction. A document can combine a native cover page, scanned attachments and pages that already have an OCR text layer.

A practical way to get it right

Open the approved source in a PDF viewer and try selecting a word inside the target table. Paste it into a plain-text area and compare it with the visible source. Repeat on another target page. Record whether you obtain readable characters, no text, an image or garbled text.

If text cannot be copied, inspect the document's restrictions and confirm the correct selection tool before concluding that the page is image-only. If readable text is present, continue with heading, alignment and row-boundary checks; text availability alone is not table fit.

For a clear printed English invoice PDF with textless pages, test ParseMill's built-in OCR and inspect the resulting characters, rows and source links. Existing text layers follow the native-text path, so incomplete or incorrect OCR text is not automatically replaced. Unreadable scans and unsupported layouts may still need a better source or reviewed manual transcription.

Adobe: Recognize text in scanned documents explains that OCR creates a searchable text layer and requires review afterward. Keep that preparation distinct from native-text table extraction.

What this looks like on a real task

Synthetic example. The people, references and figures below are invented to demonstrate the method.

The three pages require different handling. Calling the whole file “selectable” because page one works would conceal the other issues. Copy the page, observation and next-step columns and identify the actual target table pages. A file-level label is useful only when it accurately reflects the relevant content.

Catch the cases that create rework

OCR output can be selectable while still wrong. A successful text-recognition operation is not proof that identifiers or decimal signs survived. Conversely, a copying restriction does not prove the document lacks text; request an authorized usable copy rather than diagnosing a parser defect from access failure.

Avoid repeatedly uploading an unreadable file or changing labels in the hope of fixing it. For a clear textless invoice, use the OCR-supported trial once and inspect its warnings and output. If source quality or an existing text layer is the problem, obtain a better source or use a reviewed recovery method.

W3C: OCR for scanned PDF text distinguishes scanned images from actual text. Use the selection-and-paste observations to identify the relevant route, while retaining permission checks as a separate question.

Decide whether the change is worth keeping

End the diagnostic with a clear state: native text needing layout review, an existing OCR layer needing verification, a textless invoice ready for an OCR trial, an unreadable input needing recovery or access unresolved. Record sampled pages and evidence so the next handler can choose a concrete route and measure its result.