How to copy text from a screenshot or photo

Updated July 22, 2026

You get a screenshot of an error code, a photo of a whiteboard after a meeting, or a receipt you need for an expense report, and now you need the words in it as real text: something you can search, edit, or paste into an email. Retyping it by hand takes a few minutes if you're careful, and you can still get a digit wrong. Software that reads text out of images, usually called OCR, can do that for you, but it's a lot pickier about what you feed it than most people expect.

What OCR is really doing

OCR software doesn't read the way you do. It looks at shapes and lines in the image, matches them against patterns it has seen before for letters and numbers, and strings the matches back together in something like the right reading order. That's why printed text in a normal font, sitting flat and well lit, is close to a solved problem: the shapes are consistent, and the software has effectively seen thousands of examples of each one already. Anything that breaks the shape (blur, an unusual font, a hand-drawn letter, a reflection cutting through a word) makes the match harder, and the software either guesses wrong or quietly skips that patch of the image.

In other words, it's pattern matching, not reading.

What reads well and what doesn't

Some sources are close to a sure thing. A clean screenshot of an error dialog or a chat message is basically already text as far as OCR is concerned: sharp edges, one consistent font, no lighting to worry about. A printed page photographed straight on (a receipt, a book page, a printed contract) usually comes out close to perfect too, especially in decent light.

Other sources are a real gamble:

Handwriting, especially cursive. Neat block capitals sometimes survive; joined-up handwriting usually turns into gibberish or just gets dropped.
Photos taken at an angle, where the top of a page is closer to the camera than the bottom. The software has to guess at letters that are compressed and skewed.
Glare, off a whiteboard, a laminated menu, or a glossy page. A stripe of reflected light across a line of text usually just erases that line.
Low light. Noise in a dim photo looks a lot like broken letter shapes, so the software starts inventing characters that were never there.
Tiny text photographed from across a room, like a warning label or a chart legend. If you can't quite make it out yourself at that zoom, the software probably can't either.
A photo of a screen, meaning your camera pointed at a monitor instead of a real screenshot. The pixel grid and moire pattern confuse recognition even when the text still looks readable to you.

How to shoot a photo OCR can read well

If you're stuck photographing something instead of taking a screenshot (a whiteboard, a printed page, a sign), a few habits make a bigger difference than any setting in the camera app:

Shoot straight on. Get your phone parallel to the page or board rather than tilted up from an angle at the bottom.
Fill the frame. Get close enough that the text is large in the shot instead of a small patch in the middle of a wide photo; you can always crop out clutter afterward.
Use even, diffuse light. Overhead light or daylight from a window usually beats a direct flash, which tends to throw a hot spot of glare right where the text is.
Hold still and let the camera focus before you tap the shutter. A half-second of motion blur can turn a clean line of text into static.
For a whiteboard, work around old ghost-marker streaks where you can, and stand somewhere the ink still contrasts with the board under the room lighting.

Before you bother running OCR on a photo, zoom into it yourself on your phone. If you can't comfortably read a word at that zoom, the software is going to struggle with it too.

What your phone or computer can already do, and where it stops

Most phones and computers already do a version of this without installing anything extra. On an iPhone or iPad, open the photo and tap the text; the system selects it for you to copy, a feature that's been standard for years now. Android's Google Lens does much the same thing from the Photos app or straight from the camera. On Windows, the Snipping Tool and PowerToys have both picked up a "grab text from screen" action in recent versions, and on a Mac, Preview and the Photos app do the equivalent trick.

These built-in options are genuinely good for a single grab: point, select, copy, paste, done. Where they run out of road is volume and control. Doing this one photo at a time is fine for one receipt; it gets old fast across a folder of forty. None of them hand you back a file, only a copy-paste, so there's no plain text document to keep or search later unless you build one yourself. None of them flag the words they weren't confident about, so a misread "8" as a "3" in a phone number or account number can slip straight through and you won't know unless you check every character by eye. And none of them treat a genuinely scanned PDF, a document that's really a stack of page images rather than real selectable text, quite as directly as opening the file itself would.

If you're doing this often enough that one-photo-at-a-time is wearing thin, or you specifically need a plain text file, a Markdown file, or a spreadsheet out the other end instead of clipboard text, Image to Text runs the same kind of recognition in your browser across a whole batch of files, or a multi-page scanned PDF, in one go, and underlines low-confidence words so you know what to double check before you trust a number.

When the photo is something you shouldn't hand to just anyone

Some of what people run through OCR is sensitive: a pay stub, an insurance form, a passport page, a diagnosis on a printed page. Before you drop something like that into a random web tool, it's worth checking whether the site actually processes the image in your browser or uploads it to a server somewhere. A lot of "free OCR" sites are really just a form that mails your file to a backend, and once it's uploaded, you're trusting a stranger's retention habits with a document you probably wouldn't photocopy at a public library. Image to Text is built to do the recognition entirely on your device, so the photo itself never leaves your computer or phone, which matters a lot more for a benefits form than for a grocery list.

If the plan afterward is to forward the extracted text somewhere (an email thread, a shared doc, a support ticket) and it contains account numbers or someone else's personal details you don't actually need to pass along, it's worth stripping those out first. PII Redactor can scan the text for that kind of thing before you send it anywhere.