Free · no signup · never uploaded
OCR PDF
Read the text out of a scanned PDF. You get a searchable PDF that looks identical to the original, plus a plain-text copy.
Drop a scanned PDF here
or drop your file here.
Nothing is uploaded — it stays on your device.
Runs in your browser — your file is never uploaded.
Add a scanned PDF, then press “Make PDF searchable”.
Short answer
To OCR a PDF, drop your scan below and press Make PDF searchable. A scanned PDF is only a picture of a page — there is no text in it, which is why Ctrl+F finds nothing. OCR reads the picture and adds an invisible text layer over the words, so the file looks unchanged but becomes searchable and selectable. It all runs in your browser, which matters because scans are so often contracts, IDs and bank statements.
Why Ctrl+F finds nothing in a scan
When a scanner or a phone camera captures a page, what lands in the PDF is an image. To you it plainly reads as words; to the computer it is a grid of coloured pixels with no more meaning than a photo of a beach. That is why searching fails, why you cannot copy a paragraph out, and why the document is invisible to search indexes and document-management systems.
OCR — optical character recognition — closes that gap by examining the shapes in the image and deciding which letters they are. It is a genuine act of inference rather than a lookup, which is why it is slower than reading a text layer and why it is never quite perfect.
Try PDF to Text first
Many PDFs that feel unsearchable are not scans at all, and for those OCR is the wrong tool — slower and less accurate than simply reading the text already in the file. Run yours through PDF to Text first. If it hands back your text, use that: it is instant and exact.
If it comes back empty, that is your answer — the file is genuinely a scan with no text layer, and OCR is what you need.
The scan decides the result
OCR accuracy is governed far more by the input than by the software. A straight, high-contrast 300 DPI scan of printed text reads almost perfectly. Skew of a few degrees, faint toner, low resolution, or a phone photo taken at an angle all cost accuracy quickly — and handwriting is effectively unreadable, because the engine is trained on printed type.
This is true of paid OCR too. If your results are poor, re-scanning straighter and brighter will help far more than trying a different tool.
Related
- Make a PDF searchable — keep the layout, add the text layer.
- Scanned PDF to text — get the words out as plain text.
- PDF to Text — instant and exact, when a text layer exists.
- Compress PDF — useful afterwards, since OCR output is larger.
What OCR reads well — and what it doesn't
| Scan condition | Expected accuracy | Why |
|---|---|---|
| Clean 300 DPI text scan | Excellent | The case OCR is built for — near-perfect on printed text |
| 150 DPI scan | Good | Usually fine for body text; small print starts to slip |
| Photo of a page (phone camera) | Fair | Lighting and curvature hurt more than resolution does |
| Skewed or rotated scan | Poor | Straighten it first — a few degrees of tilt costs a lot |
| Faint or low-contrast scan | Poor | Characters merge into the background and drop out entirely |
| Handwriting | Very poor | Tesseract is trained on printed type, not cursive |
| Dense tables | Fair | Words are read, but column structure is not preserved |
| Non-English text | Not supported here | This tool ships the English model only |
OCR quality is decided almost entirely by the input, not the software. Straight, high-contrast, 300 DPI scans of printed text are read near-perfectly; faint, skewed, or handwritten pages defeat every engine including the paid ones. If results are poor, re-scanning straighter and brighter beats trying a different tool.
How we calculate this
Recognition runs locally in your browser — here is exactly how:
- Render. each PDF page is drawn to a canvas with pdf.js at 2× scale (roughly 144 DPI). Rendering rather than reading is the point: a scanned PDF contains a picture of a page, with no text in it to extract.
- Recognise. Tesseract (compiled to WebAssembly) reads the canvas and returns each word with a bounding box and a confidence score. One worker handles the whole document — creating one per page would re-download the model every time.
- Filter. words below 40% confidence, whitespace-only results, and zero-area boxes are dropped. Scan speckle, page edges and table rules all produce these, and writing them into the file makes its extracted text worse rather than better.
- Place. each word's box is converted from image pixels to PDF points, flipping the Y axis — Tesseract measures down from the top-left, PDF measures up from the bottom-left. Getting this wrong yields a file that looks perfect but whose hidden text is upside down.
- Write. a new PDF is built with the page image drawn full-bleed and the recognised words drawn over it at zero opacity. The text is invisible but selectable and searchable — that is what makes a scanned PDF 'searchable'. A plain .txt of the same text is offered alongside.
Assumptions
- Everything runs client-side; nothing is uploaded, and it works offline once the model is cached.
- The English (tessdata_fast) model is used — about 1.9MB rather than the 10.7MB standard model, trading a little accuracy for a download people will actually wait for.
- OCR is slow: expect a few seconds per page, and longer on an older device.
- Accuracy depends on the scan, not the software. Skewed, faint, or handwritten pages read badly in every engine.
- The output PDF embeds a rendered image of each page, so it is usually larger than the original.
Last reviewed: July 21, 2026
Frequently asked questions
Is my PDF uploaded anywhere?+
No. The pages are rendered and recognised entirely inside your browser tab, and the recognition engine itself is served from this site rather than a third party. Nothing is transmitted — which matters more here than for most tools, since scanned documents are so often contracts, IDs, medical records and bank statements.
What is a searchable PDF?+
A scanned PDF is just a picture of a page — you can see the words but your computer cannot, so Ctrl+F finds nothing and no text can be copied. A searchable PDF keeps that picture and adds an invisible text layer positioned exactly over the words. It looks identical, but now it can be searched, selected, copied and indexed.
How is this different from the PDF to Text tool?+
PDF to Text reads a text layer that is already in the file, which is instant and perfectly accurate — but it returns nothing at all for a scan, because there is no text layer to read. OCR is for exactly that case: it looks at the picture and works out what the letters are. Try PDF to Text first; if it comes back empty, your file is a scan and you want this tool.
Why is it so slow?+
Because the recognition is genuinely running on your device rather than on a server farm. Expect a few seconds per page, more on an older phone or laptop. The first run also downloads about 5MB of recognition data, which your browser then caches for next time.
How accurate is it?+
On a clean, straight 300 DPI scan of printed text, very accurate — near-perfect on ordinary body text. Accuracy falls off sharply with skew, low contrast, and low resolution, and handwriting is essentially not readable. This is true of every OCR engine, paid ones included: the input decides the outcome far more than the software does.
Which languages are supported?+
English only. Each additional language needs its own model file of a couple of megabytes, and shipping several would make the first run slow for everyone to benefit a few. If you need another language, a desktop tool with the full Tesseract language set is the better choice.
Why is the output file bigger than the original?+
Because the searchable PDF embeds a rendered image of every page, at the resolution used for recognition. If size matters more than the text layer, take the plain .txt output instead, or run the result through the PDF compressor.
It found no text at all — what went wrong?+
Usually the scan is too faint, too low-resolution, skewed, or handwritten. Re-scanning straighter and with more contrast fixes far more cases than switching tools does. It is also worth checking the file really is a scan: if it already has a text layer, PDF to Text will give you a perfect result instantly.
Related tools
Your PDF is rendered and recognised entirely in your browser and never uploaded. OCR is a best guess at the characters in an image — proofread the output before relying on it for anything that matters.