A scanned contract that you cannot search is just a pile of pictures. You cannot press Ctrl+F to find a clause, cannot copy a paragraph into an email, and a screen reader cannot read it aloud. OCR, optical character recognition, reads the letters in the picture and gives you text. The OCR PDF tool does this in your browser, which means the scan of your passport, payslip or medical letter does not have to go to a server.
What you get back
You get two things. A searchable PDF: the page looks exactly like the scan, with invisible text laid over the letters so search, select and copy work. And the text itself, which you can copy or save as .txt. You do not get a Word file with the layout rebuilt; paste the text into Word or Docs if you need to edit it.
Step by step
- Drop a scanned PDF, or photos of pages (JPG, PNG, WEBP). Several files become one PDF.
- Pick the language. English is the default. Choose the language of the text, and a second one if the page mixes two.
- Press Run OCR and watch the page counter. You can cancel at any time.
- Download the searchable PDF, copy the text, or save it as .txt.
Why the first run is slow
The OCR engine and the language data are downloaded the first time you use a language, a few megabytes from a free content network. Your browser keeps them afterwards, so later runs start faster. You must be online for that first run. Only the engine and language data are downloaded; your file is not uploaded.
What decides the quality
OCR is not magic; it depends almost entirely on the picture. Clean, straight scans of printed text at around 200 dpi or better read well. These things make it worse:
| Problem in the scan | What you will see | What to do |
|---|---|---|
| Blurry or low-resolution photo | Wrong letters, such as rn read as m | Rescan at a higher resolution or take the photo closer, in good light |
| Curved page near a book spine | Lines bend, words break | Press the page flat; scan with the page edges parallel to the camera |
| Tables and columns | Text read in the wrong order or run together | Check the result; layout is not rebuilt |
| Handwriting | Mostly wrong or empty | Not supported |
| Wrong language selected | Gibberish with the right letter shapes | Choose the correct language, add a second if two are mixed |
Choosing the language
The language list includes English, Vietnamese, Japanese, Chinese, Korean, Spanish, French, German and more. Accents and non-Latin scripts work. Selecting the correct language matters more than anything else you can change, because the engine uses it to decide between similar shapes: it is the difference between "café" and "cafe", or between a Vietnamese diacritic and a smudge. If a page mixes two languages, such as English text with Vietnamese names, add the second language.
Improve a poor scan before running OCR
- If pages are sideways or upside down, fix them first with Rotate PDF.
- If the pages have wide dark borders, trim them with Crop PDF; margins and scanner edges can produce stray characters.
- If you only need some pages, take them out with Split PDF so OCR does not run on pages you do not need.
If you are starting from paper, the Scan to PDF tool straightens and cleans photos of pages, and a black and white filter helps OCR.
Speed and size
There is no fixed page limit, but your device does the work, so a big scan takes time: expect a few seconds per page, and keep the tab open until it finishes. A phone is slower than a computer; for a book-length scan, use a computer. You can cancel at any time if you started the wrong language.
A sensible test run
When you have a long scan, do not start with all of it. Split off two or three representative pages, one with plain text and one with a table or small print, and run those first. In under a minute you learn whether the language choice is right, whether the scan quality is good enough and whether the output is usable. If the sample reads badly, rescan before you spend an hour on the whole file. If it reads well, run the full document with the same settings and keep the sample result for comparison.
Mixed-quality scans
Real documents often mix clean pages with poor ones, for example a typed contract with a faded fax page in the middle. Run OCR on the whole file, then search for a word that appears on every page, such as a company name or a heading; pages where the search finds nothing are the ones to rescan or retype. This is a quick way to find the weak pages without reading everything.
Checking the result
Never assume OCR got numbers right. After a run, search for a number you know is in the document, an amount, a date or an ID, and confirm the match. For legal or financial documents, read the text against the page for every figure that matters. Mistakes tend to cluster in digits, in small print and in tables.
What to do next
- Make the searchable PDF smaller for email with PDF Compressor.
- Summarize the recognised text with Summarize PDF; scanned files must go through OCR first because that tool does no OCR.
- Translate it with Translate PDF, also after OCR.
- Convert to Markdown with PDF to Markdown, which needs a text layer.
Privacy
The recognition runs in your browser. The OCR engine and language data are downloaded, but your file is never uploaded. That makes it suitable for documents with personal data, although you should still keep the tab on your own device.
FAQ
Is there a page limit?
No fixed limit; big scans take time because your device does the work.
Can it read handwriting?
No. Handwriting gives mistakes or nothing.
Does it make a Word file?
No. You get a searchable PDF and plain text.
Why do I need to be online the first time?
The language data is downloaded once and then kept by your browser.