How to use it
- Choose the scanned PDF with Choose a file. Its page count appears under the name.
- Leave Pages empty for the first 20 pages, or type pages such as 1-3, 8. At most 20 pages are read in one run.
- Pick the Language of the text, then press Read the text. Each page appears under its own heading as it is read; Cancel stops the run at any time.
- Check any page marked as unsure against the scan, then press Copy, or Download to keep a .txt file.
How the pages are read
Each page is drawn from the PDF at 300 dots per inch, and a very large page is drawn smaller to keep it within the browser's memory. The picture is read by Tesseract, the open-source OCR engine, which runs in a web worker so the page stays responsive. The engine finds the lines of text, then reads each character with a neural network trained on printed text in the chosen language. The language files are the fast models of tessdata_fast.
What makes a scan read well
Straight, sharp, dark text on a plain page reads best, from a scan of 200 dots per inch or more. A page that is tilted, faint, or shadowed reads worse, and so does very small print such as footnotes. Handwriting and decorative fonts are not read reliably, because the engine is made for printed text. A page with no text, such as a blank page or a drawing, is reported as having none.
The limits of a run
A run reads at most 20 pages. It may take 2 minutes for the engine to start and 1 minute for each page; a run that takes longer is stopped, and the pages it has read stay in the box. The confidence the engine gives each page is its own estimate for its words, so a page below 60 is marked for a check. It is a guide, not a proof that every word is right. A PDF that already holds text needs no OCR: PDF to Text reads that text as it is stored.
Your file stays on your device
The PDF is opened and drawn in your browser by pdf.js, an open-source library, and the engine reads the pictures in a worker. Nothing is uploaded, and nothing is stored in your browser or on our server. The text box is empty again when you leave the page.
Example: a 3-page scanned letter
A 3-page letter scanned as pictures has no text that PDF to Text can read. Choose the file here and leave Pages empty: all three pages are read, and the box shows --- Page 1 ---, then the text of page 1, then --- Page 2 --- and so on.
To read only the second and third pages, type 2-3 in Pages. Those two pages come out under their own headings, and the run is shorter.
Frequently asked questions
Why does a word come out with a wrong letter?
The engine reads the shape of each letter, so a small, faint, blurred or tilted line can come out with a letter changed. Compare the page with the scan. If many words are wrong, scan the page again straighter and at a higher resolution.
How many pages can I read at once?
At most 20 pages in one run. For a longer file, read it in parts: after the first run, type 21-40 in Pages, or split the file first with Split PDF.
What happens when a run stops early?
A run stops at its time limit, when you press Cancel, or when the engine fails. The pages already read stay in the box under their headings, and the message says how many of the pages were read.
Can it read a PDF that already has text?
It works on the pictures of the pages, so it suits a scan. A PDF that already holds text is read more exactly by PDF to Text, which takes the stored text as it is.
Is my scanned PDF uploaded or kept?
No. The PDF is read in your browser and is never uploaded. Nothing is stored in your browser or on our server, and the text is gone when you leave the page.
Why is a page marked for a check?
The engine gives each page an average confidence for the words it read. A page below 60 means the reader was unsure of many words. A high figure is a good sign, but it does not prove that every word is right.
Sources
- Tesseract OCR, the open-source engine behind the reading (Apache License 2.0)
- tesseract.js, the JavaScript library that runs the engine in a worker (Apache License 2.0)
- tessdata_fast, the fast language models used here (Apache License 2.0)
- pdf.js, the open-source library that draws the pages (Apache License 2.0)
Figures and text last checked on Wednesday, November 11, 2026.