How to Copy Text From a Screenshot

Every major platform now does this for free, with software you already have. Before you look for a tool, check the one in your hand — for ordinary paragraphs of text it will almost certainly do the job.

Platform	What to use	How
iPhone, iPad	Live Text	Open the screenshot in Photos, tap the text-selection icon at the bottom right, select as normal
Android	Google Lens	Open in Google Photos, tap the Lens button, then Copy text. Circle to Search does the same from any screen
Windows 11	Snipping Tool	Open the capture, click Text Actions, then Copy All Text
Windows 10 or 11	PowerToys Text Extractor	Win+Shift+T, drag a region, text lands on the clipboard
macOS	Live Text	Open in Preview, Quick Look or Photos and drag across the text directly
Linux	Tesseract CLI	tesseract shot.png - prints the recognised text to standard output

Two details worth knowing. Live Text needs an A12 Bionic chip or newer, which means iPhone XS, XR or later — on an iPhone X or 8 the option simply is not there, and no setting will produce it. And the Snipping Tool’s Text Actions arrived in version 11.2308.33.0; recognition runs locally on your machine rather than in the cloud, which matters if the screenshot contains anything internal.

The trick that avoids OCR entirely

On an iPhone, if the screenshot is of a web page in Safari, do not recognise it at all. Take the screenshot, tap the thumbnail, and switch from Screen to Full Page at the top. Tap Done and save to Files, and you get a PDF of the entire page with a real text layer — selectable, searchable, exact. No recognition step means no recognition errors.

The same principle applies everywhere. If the text exists somewhere as text, getting to that copy always beats reading pixels. Print to PDF, Reader mode, view-source. Copying text that cannot be selected covers the routes for pages that are actively resisting you, and Telegram’s protected channels are the specific case where none of them apply.

Where the built-in tools fall down

They are good. They are not uniformly good, and the failures cluster in predictable places.

Too few pixels. This is the single biggest cause of wrong characters, and it is not a software problem. Recognition engines want roughly twenty pixels of lowercase letter height — the equivalent of 10pt text scanned at 300 dpi. Below about ten pixels, accuracy falls apart; below eight, small text is treated as noise and discarded outright. A screenshot of a phone, viewed on a laptop at half size and then re-captured, can easily land in that range.

The fix is upstream. Zoom in before you take the screenshot, so the characters are physically larger in the captured pixels. Enlarging afterwards adds no information.

Long random strings. Recognisers use language context to resolve ambiguous shapes, and a licence key, wallet address or verification code gives them none. Every character stands alone, so 0 against O and 1 against l become coin flips. This is also why phone numbers misread more often than the words around them — pulling phone numbers out of an image covers the structural checks that catch it.

Low contrast. Dark-mode interfaces, game HUDs, semi-transparent overlays, grey placeholder text. Contrast is what separates a glyph from its background, and thin contrast plus small type is the worst combination there is.

Compressed video frames. A screenshot of a paused video carries the codec’s blocking artifacts, which smear character edges in exactly the way recognition is most sensitive to.

Multi-column layouts. A screenshot of a PDF with two columns often comes back interleaved — line one of column one, then line one of column two. The characters are right and the text is unusable.

The proofreading gap

There is a design problem shared by all the built-in tools: they put the result straight on your clipboard. You paste it, it looks right, and you discover the error three steps later when the code is rejected or the reference does not match.

For prose that hardly matters — you would notice a wrong word. For anything you cannot verify by reading, it matters a lot. Our screenshot to text tool puts the recognised text in an editable field next to the image, so the comparison happens before the copy rather than after the paste.

That single change is most of the value. The recognition in a modern browser is not dramatically better or worse than the recognition on your phone; what differs is whether you get a chance to look at it. For long codes, references and anything numeric, the screenshot to text tool is worth the extra click for that reason alone.

Getting a better capture

Crop to just the text before recognising — surrounding interface chrome, icons and buttons all give the engine things to misinterpret. Capture at native resolution rather than screenshotting a scaled window. Straighten anything skewed. And if the source is a scrolling page, capture it in overlapping sections rather than as one zoomed-out image, because pixel height per character beats completeness every time.

The short version

• Try the tool you already have: Live Text on Apple devices, Lens on Android, Snipping Tool Text Actions on Windows 11.
• Live Text needs an iPhone XS or newer; older devices do not have it at all.
• In Safari, a Full Page screenshot saves as a PDF with real, selectable text — no recognition needed.
• Most wrong characters come from too few pixels. Zoom in before capturing, not after.
• Built-in tools copy straight to the clipboard with nothing to check; use an editable output whenever the text is a code or a number.