Getting the most accurate result
OCR (optical character recognition) works best on sharp, evenly lit images of printed text. Photograph the page straight on, fill the frame with the text and avoid shadows and glare. Screenshots give the best results of all, because the letters are perfectly clean.
Crop away anything that isn't text, such as photos, logos or the edge of the desk, for a cleaner result. Always check names, numbers and amounts before relying on them.
What it can and can't read
This tool reads printed English text, including text in columns and tables, although the layout is simplified to plain lines. It does not read handwriting, and other languages and alphabets are not supported yet.
The recogniser and its English dictionary (about 7 MB) download the first time you use it and are then cached by your browser.