Images · guide

How to extract text from image files, without uploading them

By Alberto Gulotta · Updated 23 September 2026 · 11 min read

How to extract the text from an image: Windows, iPhone and Google Drive all do it already, and two of the three never send the picture anywhere. The web converters that fill this subject work too — but they need a copy of your document, which for a photographed letter or payslip is the one thing worth avoiding.

1. On Windows, capture the image first. Press ⊞ Win + Shift + S and draw around the part you want. The capture opens in Snipping Tool.
2. Select Text actions. Microsoft: this activates “the Optical Character Recognition (OCR) feature”, and from there you can “select and copy specific text or use the tools to Copy all text”.
3. Redact before you copy, if the image holds anything personal. The same menu can “Quick redact any email addresses or phone numbers in the snip”.
4. On an iPhone, turn on Live Text once. Settings › General › Language & Region › Live Text. Then open the photo, touch and hold a word, adjust the grab points, and tap Copy — or Select All for the lot.
5. Use Google Drive only when the file is not private. Right-click the file in Drive and choose Open with Google Docs. It works on PDFs as well as pictures, and the recognition happens after the upload, on Google’s side.

The technology has a name — optical character recognition, OCR — and it stopped being specialist software some years ago. Windows ships it in the screenshot tool. Apple ships it in Photos and in the camera itself. Google runs it on anything you drop into Drive. None of these is a download, a trial, or a signup.

Which one to reach for is decided by two questions, and neither of them is about accuracy — all three are good now. The first is whether the thing is on your screen or in your hand: if you can see it, the screenshot route is fastest. The second is whether the document is yours to share. That is what separates the local tools from the Drive route, and it is the question the converter sites never put to you.

The limits above are not Google being modest. They describe how OCR works everywhere, including in the paid tools. Recognition reads shapes in a line; it does not understand that four columns are four columns. So a page of running prose comes back almost perfect, and a table comes back as a stream of numbers in the wrong order. If your image is a table, the extraction is the easy half and the rebuilding is the work.

Always read the result against the original. OCR does not fail loudly. It substitutes: a 1 for an l, a 0 for an O, rn for m. The output looks clean and reads clean, and the error sits inside an account number where nothing about it looks wrong. Anywhere the text is a figure rather than a sentence — an amount, a reference, a date — check it by eye before you rely on it.

Every quoted line on this page comes from Microsoft’s, Apple’s and Google’s own documentation, read on 22 August 2026 and listed below. The argument for starting with what you already own, and the warning about silent substitutions, are ours.

Not covered here. It will not send you to a converter site first. Three tools you already own do this, and one of them does it without the document going anywhere.

It will not claim OCR is solved. Tables, columns and handwriting still break it, and the documentation of the biggest provider says so in its own words.

And it will not skip the checking step. These tools substitute characters silently, which is fine in a paragraph and serious in an account number. What holds instead is simple: the settings and the limits above are Microsoft’s, Apple’s and Google’s own wording, read on the dates this page names.