Capture the message, then copy recognized text
On Windows 11, capture the relevant message with Snipping Tool and use Text actions to select text or copy all recognized text. Paste it into a plain-text editor and compare it with the screenshot before putting it in a support ticket.
Microsoft's Snipping Tool instructions document Windows+Shift+S for capture and Text actions for OCR in the Windows 11 section. Microsoft states that this recognition runs locally on the device. The page also covers Windows 10, but its Windows 10 section does not list Text actions; do not assume an older installation has that button. The reviewed page gives no exact minimum app build for this workflow.
Open Snipping Tool from Start if you cannot find the captured image. Keep the message and enough context to identify it, while excluding unrelated windows. In the captured image, choose Text actions, copy the needed text, and paste it into your editor. Copying a screenshot as an image is a different operation from copying its recognized characters.
The official documentation was reviewed on September 16, 2026. This article does not claim a hands-on Snipping Tool or recognition test; the exercises below are fictional samples with acceptance targets.
Check a short error message character by character
A short message is a good first exercise because every important character can be checked without reviewing a whole page. For example, type the fictional line Upload failed: code E101; retry in 30 seconds. in a local text editor, display it clearly, and capture only that line.
Run Text actions and paste the recognized text into a second document. The expected transcription is exactly Upload failed: code E101; retry in 30 seconds.. That is the target, not a prediction that every OCR run will achieve it.
Check these details against the captured pixels:
- The code contains a capital E followed by the digits 1, 0 and 1.
ElO1is not equivalent. - The delay is 30 seconds, not 3 or 300. Confirm both the digits and the unit.
- The text describes a failed upload; no missing word should reverse or weaken that meaning.
For this synthetic exercise, compare the original typed line with the corrected transcription using Text Diff. Differences should disappear after correction. A matching comparison establishes agreement with your typed reference, not the accuracy of unrelated screenshots. For a real error, the screenshot or application's accessible text remains your reference.
Do not execute anything simply because OCR produced something that resembles a command. Here the output belongs in a ticket or search field after checking, and the fictional code has no diagnostic meaning outside the exercise.
Treat two columns as pairs, not a paragraph
For a settings panel, verify each value beside its label. OCR can return recognizable words while losing their relationship, which is especially misleading when several adjacent values look plausible.
Create a second illustrative sample with two rows: the left label Mode beside the right value Manual, and the left label Retry count beside 3. Your accepted transcription should preserve the pairs Mode: Manual and Retry count: 3, even if you add the colons yourself during review.
If the result reads Mode Retry count Manual 3, do not paste it as though the layout were preserved. Return to the screenshot and reconstruct one pair at a time. If recognition repeatedly mixes rows, capture each row separately while keeping both its label and value visible. Cropping away the labels can make the text easier to recognize but harder to interpret correctly.
This is a practical limit of the output: plain text cannot serve as evidence that a table's structure survived. Keep an authorized copy of the screenshot with your working notes when that visual relationship matters, and distinguish the corrected transcription from the original image.
Use a saved-image fallback when Text actions is unavailable
If your installed Snipping Tool lacks Text actions, use an approved OCR workflow for a saved screenshot. For a non-sensitive practice image, ToolboxHub OCR (Image to Text) provides a separate browser route.
Choose one image, select English for the samples above, and press Start OCR. Wait for a result for that image before using Copy or Download .txt. The result box is read-only: paste into another editor to correct characters or rebuild the two label-value pairs.
The current component passes the image to a browser Tesseract worker and displays returned plain text. It loads a third-party script and may download worker, core and language resources, so browser processing does not mean the page will work without a network connection. This is code inspection, not a network trace or a promise about every browser environment.
Changing the selected image or language clears the old result. Start recognition again and verify the new output before copying. The tool does not edit the screenshot, save DOCX or XLSX files, reconstruct a settings table, or confirm whether an error code is valid. Use an approved local process if the image cannot be handled in this browser workflow.
Diagnose missing controls and misleading output
Match the remedy to the symptom instead of treating every problem as poor image quality.
- There is no Text actions button: first confirm that you are in the captured-image editor rather than the capture overlay. Then check which app and Windows version you have. Use the saved-image fallback when the documented control is unavailable; do not keep looking for a Windows 11 feature in the Windows 10 instructions.
- You paste a picture instead of characters: return to Text actions and copy recognized text. A normal image-copy command does not create a text transcription.
- The output confuses zeros and letters: enlarge the original on-screen text before recapturing, keep characters sharp, and check the code manually. Enlarging an already blurred screenshot does not recover missing detail.
- Browser OCR stops before producing text: runtime or language downloads may be blocked. Check the allowed network environment or use another approved OCR route. Do not weaken organizational controls just to load a model.
- Words are correct but values appear under the wrong label: reduce the capture to one complete row and compare each pair with the original. Reformatting the same scrambled text is not independent verification.
Before submission, reopen the exact text attachment or review the ticket draft. Confirm the code, units and label-value pairs again, and remove unrelated information. The progress indicator reports processing progress, not recognition accuracy.
Frequently asked questions
Does Windows 10 have the same Text actions workflow?
The cited Microsoft page documents it in the Windows 11 section and omits it from the Windows 10 section. Use the controls actually available in your installation or a saved-image OCR alternative.
Can OCR turn my settings screenshot into an editable table?
The ToolboxHub result is plain text. Rebuild any table yourself after checking each label and value; the tool does not export an editable spreadsheet or preserve the screenshot's geometry.
Is a fully completed progress bar proof the code is correct?
No. Completion means the recognition operation finished. Compare every character in important codes and every number with its unit against the image.
Should I keep the original screenshot after correcting the text?
Keep it in an appropriate location when it is needed to verify the transcription, following your retention rules. A corrected text copy is useful for searching and tickets, but it is a separate representation and cannot replace all visual context.