Searchable PDF (OCR)
Make a scanned PDF searchable in your browser. OCR adds an invisible text layer so you can search, select and copy the words, and the scan itself is left untouched.
Drop your file here
or click to browse
300 is what OCR engines are tuned for. Higher is slower and rarely more accurate unless the type is very small.
Leave this on. OCR'ing a page that already has text leaves two overlapping copies of every word.
Words the engine is less sure about than this are left out, so searches don't match nonsense. Raise it for a clean scan, lower it for a poor one.
- ProcessingOn your device
- Uploaded to a serverNothing
- InternetNeeded once, to fetch the OCR engine
Your document still never leaves this device — but unlike every other tool here, this one downloads its recognition engine before it can start. Once that is cached, it runs offline too.
Worth knowing
This is the one tool here that needs the network on first use. Your document never leaves the device — but the page has to download the OCR engine and English language model, about 7MB, from this site. It is cached afterwards, and it is a request to us rather than to a third party, but it does mean this tool alone cannot run fully offline.
English only for now. Adding a language means shipping another model file, so more will follow based on what people ask for.
Accuracy depends almost entirely on the scan. Straight, sharp, 300 DPI black-on-white text reads very well. Faint photocopies, phone photos at an angle, handwriting, and text over background images read poorly — the reported confidence tells you which you have.
What a scanned PDF actually is
A scan is a photograph of a page. Open one and you see words, but the file contains no words at all — only a grid of coloured dots that happens to look like writing. That is why Ctrl+F finds nothing, why you cannot select a sentence, and why converting it to Word gives you an empty document.
OCR — optical character recognition — looks at those dots, works out which letter shapes they form, and writes real text. What this tool then does with that text is the part worth understanding, because it is where implementations differ most.
The invisible layer, and why it is the right approach
The obvious way to build this is to render each page to an image, recognise it, and assemble a new PDF from those images with the text behind them. It works, and it costs you something on every page: the scan is compressed a second time, so it gets softer, and the file usually ends up larger than what you started with.
This does not rebuild your file. It opens the original, leaves every page exactly as it is, and draws the recognised words on top in text rendering mode 3 — the PDF specification's instruction to render nothing at all. The words are genuinely there, so search, selection and copy all work, and not a single pixel of the page changes. Your scan comes out byte-for-byte as sharp as it went in.
Why the words line up when you select them
Tesseract reports a bounding box for every word it reads. Each invisible word is sized so its width matches the box of the word underneath it, rather than being dropped in at a fixed size. That is what makes a double-click select the right word and a search highlight sit on the right span instead of floating near it — the difference between a searchable PDF that feels right and one that feels broken.
Pages that already have text are left alone
Plenty of PDFs are part scan and part real text — a typed report with a signed page scanned back in, say. Running OCR across the whole thing would leave two overlapping copies of every word on the pages that were already fine, so searches return duplicates and copying a paragraph gives you it twice.
So pages with an existing text layer are detected and skipped, and the result tells you how many. You can switch that off if a page has a bad text layer you want replaced.
The honest bit about privacy
Every other tool on this site works with your network switched off. This one does not, and that is worth stating plainly rather than burying.
Your document still never leaves your device — the recognition runs in your browser through a WebAssembly build of Tesseract. But the engine and the English language model are about 7MB, and they have to be downloaded the first time you use the tool. They are served from this site, not from a third-party CDN, so nobody else sees the request, and they are cached afterwards. What that means in practice: your file is private, and the fact that you used an OCR tool is known to this site the same way visiting any page is. Claiming full offline operation would be a lie.
Getting a good result
- Scan at 300 DPI. It is what the engine is tuned for, and going higher is slower without being more accurate unless the print is very small.
- Straight matters more than sharp. A page rotated by even three or four degrees costs more accuracy than a slightly soft focus does.
- Black text on white paper reads best. Coloured backgrounds, highlighter and text printed over images all reduce accuracy.
- A phone photo will work, but flatten the page and avoid shadows — a curled page near the spine is the usual reason a book scan reads badly.
- Handwriting will not work. This engine recognises printed type; cursive is a different problem and not one it solves.
The result reports the average confidence the engine had. Above about 90% you can trust the text; below 75% it is worth rescanning before relying on it for anything that matters.
Once a scan is searchable there is a real text layer to work with, so converting it to an editable Word document or pulling a table into a spreadsheet both start working — neither can do anything with a scan on its own.
If the scan is crooked or the pages are the wrong way up, fix that first: rotating the pages before OCR is much more effective than after.
Questions people ask
Does this change how my scan looks?
No, and that is the point of doing it this way. Most tools rebuild the PDF out of freshly compressed page images, which softens the scan and often makes the file bigger. This keeps your original file and adds to it: the recognised words are drawn over the page in the PDF specification's invisible text mode, so they can be searched, selected and copied while not one pixel of the page changes.
Is my document uploaded for the OCR?
No. The recognition runs in your browser, in a WebAssembly build of Tesseract. What the page does download — once, then cached — is the engine and the English language model, about 7MB, served from this site rather than a third-party CDN. So your file stays local, but this particular tool does need a connection the first time you use it. Every other tool here works with the network switched off, and it would be dishonest to claim the same for this one.
What if my PDF already has text?
Those pages are skipped by default, and the result tells you how many were. Running OCR over text that is already there leaves two overlapping copies of every word, which makes search results duplicate and copy-and-paste return everything twice. If a page has a bad text layer you want replaced, switch the skip option off.
How accurate is it?
On a clean 300 DPI scan of printed text, typically 95% or better — the result reports the actual average confidence so you are not guessing. What hurts it: low resolution, skew, faint or broken type, shadows from phone photos, text over pictures, and unusual fonts. Handwriting is not realistically recognisable by this engine at all. Words the engine is unsure about are left out rather than added as nonsense, and you can adjust that threshold.
Why does the first page take so long?
The engine and language model load before the first page can be read, which takes a few seconds. After that each page is much quicker, and the model stays loaded if you convert another document in the same visit.
Can I get the text out as a file?
Once the PDF is searchable, PDF to Text or PDF to Word will pull the words out, because there is now a real text layer for them to read. That is the usual two-step route for getting a scan into an editable document.
How to use it
How-to guides · 21 August 2026 · 7 min
How to make a scanned PDF searchable
Ctrl+F finds nothing because your document is a picture of words rather than words. OCR fixes that by adding an invisible text layer — without changing a single pixel of how the scan looks.
How-to guides · 9 August 2026 · 3 min
How to scan a document properly with your phone
A photograph and a scan are the same capture done differently. Lighting, angle and framing are what separate them — and why your council rejected the last one.