Scanned PDF or text PDF? How to tell, and why it matters
Two files that look identical can behave completely differently. How to tell which you have in three seconds, and what each one can and cannot do.
Updated 2026-08-03
A PDF containing a picture of a page and a PDF containing the text of that page look identical on screen and behave nothing alike. One is searchable, selectable, accessible to a screen reader and a few hundred kilobytes. The other is none of those things and forty times larger. Almost every frustration people have with PDFs — cannot search it, cannot copy from it, will not compress, will not read aloud — traces back to having the second kind and expecting the first.
The three-second test
Open the file and try to select a line of text with your cursor. If you get a text selection, there is a text layer. If you get a rectangular marquee over an image, there is not. Ctrl-F for a word you can see is the same test from the other direction.
| Text PDF | Scanned PDF | |
|---|---|---|
| Select text | Yes | No |
| Search inside it | Yes | No |
| Typical size, 20 pages | ~200 KB | ~8–40 MB |
| Screen reader can read it | Yes | No — legally significant in many contexts |
| Compresses well | Barely — already small | Dramatically, 80–95% |
| Reflows on a phone | Sometimes | Never |
| Extract a table from it | Awkward but possible | Not without OCR |
Where scanned PDFs come from
- An actual scanner or a phone camera. The obvious case.
- A fax gateway. Still common in healthcare, law and government.
- "Print to PDF" from an image viewer. The text was gone before the PDF existed.
- Aggressive PDF compression. This is the one that catches people: rasterizing compressors turn a searchable document into pictures of itself. The file gets much smaller and quietly loses everything else.
Turning one into the other
OCR recognizes characters in the image and produces text. It is genuinely good on clean, straight, printed documents — 98% or better — and it degrades sharply with skew, low resolution, unusual fonts, handwriting and background patterns. It does not repair a bad scan; it reads what is there.
- 300 DPI is the floor. Below that, accuracy falls off a cliff. 150 DPI scans are the most common cause of bad results.
- Straighten before you recognize. A few degrees of skew costs more accuracy than any setting will recover.
- Grayscale beats color for text, and beats a heavy-handed black-and-white threshold that eats thin strokes.
- Expect to proofread. 98% accurate means roughly one error every two lines. That is fine for search and wrong for a contract.
Which problem you actually have
| Symptom | Cause | Fix |
|---|---|---|
| Cannot search or select | No text layer | OCR it |
| 40MB for 12 pages | Page images, probably at high DPI | Compress — this is the case it is built for |
| Compressed it and now cannot search it | The compressor rasterized it | Go back to the original. This is not reversible |
| Copied text comes out as gibberish | Broken font encoding, not a scan | OCR a rendering of it — extraction cannot help |
| Screen reader reads nothing | No text layer, or no tag structure | OCR, then check the tagging |
| Text is selectable but in the wrong order | Multi-column layout | Extract per column, or accept the reflow |
The compression trap, stated plainly
If a PDF is large because it contains scans, compressing it is exactly right and the saving is enormous. If it is large for some other reason and it has a text layer, a rasterizing compressor will destroy that layer to save space you probably did not need — and the original is the only way back. Check for a text layer before you compress anything you might want to search later.
Questions
How do I know whether OCR already ran on a file?
Try to select the text. A scan that has been OCR'd has an invisible text layer behind the image, so selection works even though what you see is a picture. That is the standard "searchable PDF" format.
Why is my scanned PDF so enormous?
Every page is a photograph. A color page at 600 DPI is several megabytes before anything else, and twenty of them is a large file. Scanning in grayscale at 300 DPI typically cuts it by an order of magnitude with no loss of readability.
Can OCR read handwriting?
Standard OCR, essentially no — it is trained on printed type. Handwriting recognition is a separate and much harder problem, and results on natural handwriting remain poor outside constrained cases like form fields.
Does OCR change how the document looks?
No. The text layer is added behind the image, so the page renders exactly as before and is now searchable. Nothing visible changes, which is why people often cannot tell whether it has been done.