Guide · Updated 23 July 2026 · Reading time 7 minutes

How to Make a Scanned PDF Searchable

You have a PDF open. You press Ctrl+F, type a word you can plainly see on the screen, and the search finds nothing. Or you try to select a line of text and instead get a blue box over the whole page.

Nothing is broken. That PDF does not contain text at all — it contains a photograph of text. This guide explains the difference, what OCR does about it, and how much of it you can realistically expect to work.

Two completely different things are both called "PDF"

The format allows both, and they look identical on screen.

	Born-digital PDF	Scanned PDF
Where it comes from	Exported from Word, LibreOffice, a browser's Print-to-PDF, an accounting system	A flatbed scanner, an office multifunction machine, or a phone camera
What is inside	Actual character codes plus font information	One image per page, nothing else
Ctrl+F	Works	Finds nothing
Select and copy	Works	Selects the image
Typical size	Small — tens of kilobytes per page	Large — often several megabytes per page

Ten-second test: try to drag-select a single word. If a neat text highlight appears, it is born-digital and already searchable. If the whole page highlights as one block, it is a scan.

A third case trips people up: a born-digital PDF that was printed and then scanned. It began life as real text, but the text was thrown away the moment it hit paper. What you have now is a scan, and it needs the same treatment as any other scan.

What OCR actually does

OCR — optical character recognition — looks at the pixels of a page image and works out which letters they represent. It finds regions that look like text, splits them into lines and then into individual characters, compares each shape against a trained model, and outputs its best guess as real text.

Two things follow from this, and they explain most of what frustrates people about OCR:

It is a guess, not a lookup. There is no hidden text being revealed. The software is inferring characters from shapes, so it can be wrong, and it is wrong more often on poor images.
It never touches how the page looks. Making a scan searchable does not sharpen it, straighten it, or clean it up. It adds an invisible text layer behind the existing image. The page looks exactly the same afterwards; it has simply become searchable and selectable.

What determines accuracy

OCR quality is decided almost entirely before you run it — by the image you feed in.

Helps a great deal

Resolution. Roughly 300 DPI is the practical floor for body text. Below that, characters like e, c and o blur into each other. This is why scanning at a higher resolution beats enlarging a small scan afterwards — upscaling invents pixels, it does not recover detail.
Straightness. Recognition works line by line. A page tilted by more than a couple of degrees makes line-finding harder and accuracy falls off noticeably.
Even lighting and clean contrast. Dark ink on a light background, with no shadow gradient across the page.
Ordinary printed fonts. Standard serif and sans-serif text at normal sizes is what the models are trained on.

Hurts, sometimes badly

Handwriting. Conventional OCR is built for print. Handwritten notes, signatures and margin annotations will mostly come out as nonsense. Expect nothing here.
Faint carbon copies, old typewriter output, heavy stamps. Broken or overlapping character strokes are exactly the failure case.
Dense tables and multi-column layouts. The characters may be read correctly while the reading order comes out scrambled — columns interleaved, table cells run together. Text is recovered; structure often is not.
Decorative or very small type. Anything below about 8pt in a scan is marginal.
JPEG artefacts. A heavily compressed scan has fuzzy halos around every character.

Doing it in Pennary

Open your PDF, then use the Convert bar. Three routes exist depending on what you want out.

Read text on screen

Extracts the text and shows it in an editable box you can correct and copy. If the PDF is born-digital, this pulls the existing text layer directly — that is exact, instant, and unlimited, because no recognition is involved. If it is a scan, recognition runs on the page images.

Searchable PDF

Produces a new file that looks identical to the original but has an invisible text layer added, so Ctrl+F and text selection work. If you feed it a PDF that already has a text layer, it will tell you so rather than doing redundant work. This is the option you want for filing and archiving, because the visual document is preserved exactly.

PDF to Word

Gives you an editable document. For born-digital PDFs this is free and unlimited, and it is text-accurate — but understand what that means: you get the words, not a pixel-perfect rebuild of the layout. Complex formatting, multi-column layouts and precise table structure will need fixing by hand. Any tool promising a perfect layout conversion is overselling.

Current limits, stated plainly: recognition is English only. Scanned-page OCR is free for the first 2 pages per session; beyond that it is a Pro feature. Extracting text from born-digital PDFs is free and unlimited and does not count toward that limit.

Getting the best result

Fix the image first. Straighten and clean the page before recognition, not after. Ten seconds of correction beats any amount of post-editing.
Check a sample page before committing. Run one representative page — ideally the worst-looking one. If that comes back as gibberish, the source image is the problem, and processing the remaining forty pages will not go any better.
Rescan rather than fight a bad scan. If the original is available, a better capture is almost always faster than correcting the output of a poor one.
Proofread anything that matters. OCR mistakes cluster in predictable places: 0 and O, 1 and l and I, rn read as m. If the document contains reference numbers, amounts or dates, check those specifically — they are exactly where an error costs you and where a human eye skims past it.

Common problems

Symptom	Likely cause	What to do
Output is meaningless characters	Resolution too low, or page badly skewed	Rescan at higher resolution; straighten before recognition
Most words right, numbers wrong	Digit and letter shapes are genuinely ambiguous	Proofread numeric fields by hand — always, on anything official
Text jumbled out of order	Multi-column or table layout	Expected. Recognition recovers characters, not reading order
Handwriting not recognised	Print-trained OCR	No fix. Type it manually
"Already searchable" message	The PDF is born-digital	Nothing needed — Ctrl+F already works
Searchable PDF slightly misaligned selection	Word boxes are positioned approximately	Cosmetic only. Search and copy still work correctly
Very slow on a long document	Recognition is computation-heavy and runs on your device	Process in smaller batches; keep the tab in the foreground

Why this runs on your device

Most online OCR services upload your pages to their servers. Pennary runs recognition locally in your browser instead. It is slower than a data centre — that is the honest trade — but the document never leaves your machine, which matters when the thing you are digitising is a contract, a medical record, an identity document or official correspondence.

Open the tool →