How to redact a screenshot before sending it to AI

A screenshot shows more than the thing you meant to ask about. A browser tab, a notification, a terminal above the error, a key in a config file: anything on screen travels with the image once you send it. Before you paste a screenshot into an AI chat, run through this checklist.

What to look for

Scan the image for these, in roughly this order:

  • Email addresses, including the one in the corner of a mail client or an account menu.
  • Phone numbers.
  • Card numbers, including the last digits in a billing screen.
  • API keys and tokens, in terminals, .env files, browser developer tools and dashboards.
  • Names, addresses and anything else about other people that the question does not need.
  • Background clutter: open tabs, chat popups and notifications.

The simplest protection is to capture less. Select only the region you are asking about instead of the whole screen.

How to cover it properly

The way you cover something matters as much as covering it.

  • Remove the pixels, do not just draw over them. A box, highlight or blur that is only a layer on top of the original image can sometimes be lifted off again. Use a tool that flattens the image, and check what it actually sends.
  • Use Pixelate or Blur for text, a solid fill for secrets. Light effects on small text can leave it readable, so look at the result at full size.
  • Look at the final image, not the editor. Open or zoom the result before sending.

Doing it in SnapPrompt

SnapPrompt has redaction built into the capture:

  1. Press Ctrl+Shift+1 and select the region.
  2. Press Alt+A to annotate, pick Blur (B) or Pixelate (X), and drag over what must not be shared.
  3. Or open the editor with Alt+E and press Auto-redact. It finds email addresses, phone numbers, card numbers, API keys and tokens in the recognized text and covers them. A notice tells you how many it hid, and one Ctrl+Z undoes all of them.
  4. Blurred and pixelated areas are removed from the image before it leaves the app.

Auto-redact depends on text recognition, which runs on your computer, and it only looks for the kinds of data listed above. It will not catch a name, a face or text it cannot read, so cover those by hand and always look at the image before pressing Send. Learn more about how the text recognition works.

A note on what you are sending

Whatever tool you use, a redacted screenshot is still a screenshot you are sending to an AI service. Read that service’s data policy, and if your organization has rules about it, follow them. SnapPrompt keeps captures on your computer until you press Send, and an organization policy lets an admin set limits such as image retention. The details are in our privacy policy.

Checklist you can reuse

  1. Capture only the region you need.
  2. Cover emails, phone numbers, card numbers, keys and tokens.
  3. Cover other people’s details.
  4. Check that the hidden areas are removed, not just covered.
  5. Look at the final image once before you send it.

All posts

Try SnapPrompt free for 3 days

Windows 64-bit. After the trial, $19 + applicable tax once, with 1 year of updates included. You use your own Claude account.