Reading a screenshot on your phone is one of those small annoyances that never quite goes away, because the text is obviously readable to a human but not to anything that can select it. The same problem applies to a photograph of a receipt, a page in a book, a serial number on a label, or a note somebody handed you. Optical character recognition is what turns those back into text you can work with.
How it reads the page
The tool analyses the shapes in the image and matches them against the patterns of letters and numbers. Printed text in a clean image is the easy case, and the accuracy is high enough that the result is usually usable as it stands. What it does not do is read the page the way a person does. It does not know which block is a heading, and it has no way of knowing that a faint line is a footnote rather than a stray mark.
What makes a photograph easier to read
Distance and light matter more than anything else. Get the whole page in the frame so you are not relying on the edge of the image, and take the shot so the text fills as much of the picture as possible. Flat, even light beats a dramatic angle. And hold the camera parallel to the page, because perspective distortion is the single most common reason a photo of text comes out hard to read.
If you can screenshot on a device rather than photograph, do that instead. A screenshot is perfectly square, perfectly lit and completely sharp.
Handwriting and awkward cases
Handwriting works when it is neat. Slow, careful lettering in dark ink is often fine; quick shorthand, light pencil, and text crossing another line are not. It is worth knowing this before you take twenty photographs of something that will not be recognised.
Always check the result
Recognition makes mistakes that are easy to miss if you are reading quickly, and the usual pattern is predictable. Zeroes and the letter O are confused. The digit one and lowercase l are confused. Missing spaces are more common than wrong words. A glance before you paste is usually enough to catch them, and far cheaper than finding a mistake in something you already sent.
The processing happens in your browser, so a photograph of a document, a receipt with your address on it, or anything else personal stays on your device and is never uploaded.
Beyond copying text
Reading an image is more useful than it first appears, because it turns pictures into something you can search, sort and paste. A screenshot of an error message becomes text you can drop into a support ticket without retyping. A photograph of a whiteboard becomes notes. A photo of a book page becomes something you can quote from, using a copy operation rather than a careful retyping exercise.
The searchable version is the biggest practical gain. If you photograph paperwork, receipts or notes, having the text means you can find a specific line later instead of scrolling through pictures. A folder of images you cannot search is a folder you have to remember the contents of.
Handwriting deserves one more mention because it improves unevenly. Neat block capitals are the most reliable style by a wide margin. Cursive is difficult, joined-up printing is worse, and anything crossed out or written over gives mixed results.
Related reading
Guides that pair with this