How-to guide

How to redact a screenshot without leaving data behind

The safest redaction happens before the shutter clicks. Blur the sensitive parts in the page first, then capture, so the finished image holds nothing but frosted pixels, with no original hiding underneath to move, undo, or reverse.

Updated August 2026

Quick answer

To redact a screenshot safely, remove the sensitive content before you capture the image, not after, and cover short predictable values with a solid shape rather than a blur. Editing a saved screenshot leaves the original inside the file: a black rectangle added in an editor can sit on its own layer, and in a PDF the text underneath stays selectable. Redacting in the live page first means those pixels never enter the file at all. Then match the treatment to the content: a heavy blur is fine for prose, faces and imagery, but anything drawn from a small set of possible values (a PIN, a short code, a date) should be cropped out or covered, because it can be brute-forced regardless of blur strength.

With Blurr that is: open the toolbar with Ctrl + Shift + YCmd + Shift + Y (Cmd + Shift + Y on Mac), blur every private element at full intensity, verify nothing readable remains, then capture with your normal OS tool.

The important part

Why redacting in the page beats editing the image

The difference is where the original data ends up. Redact after capture and it may still be in the file; redact before, and it never enters the file at all.

Black boxes can come off

Drop a black rectangle on an image in an editor and the original pixels can survive underneath, in a saved layer or as a movable object in a PDF. Flatten it wrong and the data is one click away from anyone who opens the file.

Weak blur and pixelation can be reversed

A light Gaussian blur or a coarse mosaic over small, predictable text (like a number or a short ID) can leave enough signal for software to reconstruct it. Redaction has to destroy the information, not merely soften it.

Redacting before capture removes the data

When you blur an element in the live page and then screenshot it, the capture only ever sees the blurred result. A strong, high-intensity blur applied before capture replaces the underlying pixels in the image. There is no separate original layer to move, undo, or reverse.

One honest caveat. Blurring in the page solves the hidden-original problem: there is no layer, no text object, and no movable box left in the file. It does not make the blur itself irreversible. A blur is a low-pass filter, so it attenuates detail rather than destroying it, and short predictable text can be recovered by rendering every candidate and matching. Blur prose and images freely. Crop or cover anything short and guessable.

The mechanism

Why blur and pixelation can be reversed

Almost every guide on this topic asserts that blur "can sometimes be reconstructed" and then never explains how. The mechanism matters, because it tells you exactly when a blur is safe and when it is useless.

A blur attenuates information, it does not destroy it

A Gaussian blur is a low-pass filter. Each output pixel is a weighted average of its neighbours, so the fine detail is spread across a wider area rather than deleted. The signal is still present, just heavily attenuated and mixed together. That is the fundamental difference between obfuscation and redaction, and it is why a solid fill sits in a different category from every blur setting.

The real attack is brute-force rendering, not "unblurring"

Nobody runs your image through a magic deblur filter. They do something far more reliable: take the same font, at the same size, on the same background, render every plausible candidate string, apply the same blur or the same mosaic grid, and compare each result to your pixels. The closest match wins. This is how the published tools Depix (2020) and Bishop Fox's Unredacter (2022) defeat pixelated redactions, and the same logic applies to blur.

So the variable is entropy, not strength

Brute force only works when there are few enough candidates to test. That single fact decides everything. A 4-digit PIN has 10,000 possibilities and will fall at any blur radius. A US phone number, a date of birth, a short order ID, or a dollar amount sit in an uncomfortable middle. A paragraph of free prose is effectively infinite and is genuinely safe behind a heavy blur. Ask how many things the hidden text could be, not how strong the blur looks.

Deterministic distortions are simply invertible

The swirl effect once used to obscure faces is a reversible geometric transform: apply the inverse and the original returns. It was undone in real criminal investigations. Any effect that rearranges pixels rather than discarding them belongs in the same category.

Reference

Redaction methods, by what they actually guarantee

Two independent failure modes, tracked separately: whether the mask itself can be undone, and whether the original survives elsewhere in the file.

Comparison of screenshot redaction methods by reversibility and by suitability for short predictable text versus free prose.
Method Mask reversible? Short, guessable text Free prose & imagery Why
Solid opaque fill No Safe Safe Destroys the pixels outright. The only visual redaction with no recovery path, provided the file is flattened.
Crop it out of frame No Safe Safe The data is never in the image. Strictly the safest option, and the right answer for short values.
Replace with fake data No Safe Safe Swap in a fixture value before capturing. Also the most readable option for documentation.
Heavy blur (20px+ on 12-16px text) In principle Not safe Safe in practice Fine for paragraphs, faces and imagery. Not for a PIN, a code, or anything from a small set of possible values.
Light blur / soft haze Yes Not safe Not safe Leaves word shapes, character counts and line lengths. Treat as decoration, not redaction.
Pixelation / mosaic Yes Not safe Risky Each block averages a known region, which is exactly what makes brute-force matching work. Defeated by published tools.
Swirl / twirl distortion Yes Not safe Not safe A deterministic, invertible transform. Reversed in real criminal investigations.
Black box drawn over live text N/A Depends Depends The mask is opaque, but in a PDF or a layered file the text object underneath survives and stays selectable. Flatten, then verify.

Blurr applies a blur, so it sits in the "heavy blur" row: right for prose, faces, dashboards and imagery, and the wrong choice for a short guessable value. What it does add over an image editor is that no original layer is ever written to the file.

Step by step

Redact a screenshot in five steps

Blur first, verify, then capture. The whole flow adds about a minute and removes the guesswork.

  1. 1

    Blur the sensitive elements in the page

    Before you capture anything, open Blurr with Ctrl + Shift + Y (Cmd + Shift + Y on Mac) and draw a box over (or click) every private detail inside the page: names, emails, account numbers, and API keys. Anything outside the page, such as the address bar or a notification, has to be cropped out of the shot instead, because no extension can paint over browser chrome.

  2. 2

    Turn the intensity up high

    Drag the intensity slider to maximum, a 20px blur. That is well past the 12px to 16px text most pages render. A light haze or a coarse pixelation leaves recoverable traces, and so does any blur over a short, guessable string: crop those out instead.

  3. 3

    Verify nothing readable remains

    Zoom the page in, check the edge of each blur, and confirm no characters peek out around the mask. If anything is legible, resize the blur or push the intensity higher.

  4. 4

    Take the screenshot with your OS tool

    Capture with your native screenshot tool: Cmd + Shift + 4 on macOS, Windows + Shift + S on Windows, or Shift + Print Screen on Linux, which GNOME and KDE both map to a region capture. The saved image records only the blurred pixels; there is no hidden original layer behind them.

  5. 5

    Check the saved file before you share

    Open the screenshot at full size and scan it once more. What you see in the file is exactly what everyone else will see. No preview shortcuts, no trusting a scaled-down thumbnail.

Avoid these

Common redaction mistakes

Most screenshot leaks are not exotic: they come from small oversights. Run this checklist before anything with sensitive data leaves your machine.

  • Blurring too lightly: a soft haze that still hints at the shapes underneath. Push the intensity up until the content is genuinely gone.

  • Leaving the edges exposed, so a character or two peeks out around a too-small box. Resize the blur to fully cover the text.

  • Blurring a short, guessable value (a PIN, a 4-digit code, a date) instead of cropping it out. Strength does not help when the candidate list is small.

  • Forgetting the address bar, tab titles, desktop notifications, and autofill dropdowns that can appear in the frame.

  • Trusting a small preview instead of opening the finished file at full size before you send it.

  • Sharing the file straight out of a synced folder, so the unredacted original is already in the cloud.

Scope

What to redact, according to published standards

You do not have to invent a list. Two widely used standards already enumerate what counts as identifying, and they work as a checklist well outside the industries that mandate them.

Direct identifiers

Names, emails, phone numbers, home and IP addresses, account numbers, and photographs. HIPAA’s Safe Harbor method lists 18 identifier types that must be stripped before health data counts as de-identified, and it is a useful checklist well outside healthcare.

Anything that unlocks an account

API keys, tokens, session cookies, connection strings, recovery codes. Treat these as compromised the moment they appear in a shared image, and rotate rather than hope.

Card numbers, to the PCI rule

PCI DSS allows displaying at most the first six and last four digits of a card number. If your screenshot shows more than that, it needs redaction regardless of who you are sending it to.

The frame around your content

Address bar, tab titles, bookmarks, notification toasts, autofill dropdowns, and the profile chip. None of these are part of the page, so no extension can blur them. Crop, or share a single tab instead.

Beyond the pixels

The file is not the picture

You can redact everything visible and still leak. These four are the ones that catch careful people, and none of them are fixed by a better blur.

Metadata rides along

Screenshots carry creation timestamps, display scale, and on some tools the source app or window title. A phone photo of a monitor carries EXIF, sometimes including GPS. Strip it: Preview’s inspector on macOS, "Remove Properties" on Windows, or exiftool -all= from a terminal.

The filename leaks too

"Screenshot 2026-07-24 at 09.41.12 acme-corp-invoice.png" names the customer before anyone opens it. Rename before you attach.

Auto-upload beats you to it

macOS saves to a Desktop that iCloud may sync, Windows to a OneDrive folder, and several capture tools upload the original to their own CDN to generate a share link before you have edited anything. Check where your tool writes first.

Previews regenerate from the original

Some pipelines build thumbnails from the source file rather than your edited version. If you edited in place, confirm the preview updated too.

Before you send it

Verify your redaction in five checks

Thirty seconds of checking catches almost every real-world miss.

  1. 1

    Zoom to 400% and read along every edge of every mask. Stray characters hide at the boundaries.

  2. 2

    Select all on the page and copy. If redacted text lands in your clipboard, the mask is visual only.

  3. 3

    Open the saved file at 100%, not a thumbnail. What is in the file is what your reader gets.

  4. 4

    Screenshot your screenshot and look at that. It shows you the flattened result with no editor affordances.

  5. 5

    Ask someone else to look before it goes anywhere public. Fresh eyes find what yours have stopped seeing.

Terminology

Redaction, blurring, masking: what each word means

These get used interchangeably and they guarantee very different things.

Redaction
Permanently removing sensitive content from a document or image so it cannot be recovered. True redaction destroys the underlying data rather than covering it.
Blurring
Applying a low-pass filter that spreads each pixel into its neighbours. It attenuates detail rather than destroying it, so information technically remains and can sometimes be reconstructed.
Pixelation
Replacing blocks of pixels with their average colour. Because the transform is deterministic and the block grid is known, pixelated text can be brute-forced by rendering candidates and matching.
Masking
Hiding part of a value while leaving the rest readable, such as showing only the last four digits of a card number. Masking is about display; it does not alter the stored record.
Obfuscation
Making data harder to read without guaranteeing it cannot be read. Blurring, pixelation and swirling are all obfuscation, not redaction.
Anonymisation
Irreversibly altering data so no individual can be identified from it, directly or indirectly. Under GDPR, genuinely anonymised data falls outside the regulation entirely.
Pseudonymisation
Replacing identifiers with stand-ins while keeping a separate key that can restore them. Under GDPR Article 4(5) pseudonymised data is still personal data, because re-identification remains possible.

Honest limits

When a browser blur is the wrong tool

Blurr redacts what is rendered in a web page. Four situations need something else, and reaching for a blur in any of them is a mistake.

A PDF you are publishing

Use a redaction tool that removes the text objects from the content stream, not one that draws a rectangle over them. Acrobat’s Redact and Preview’s redaction do this; drawing a shape does not.

Legal discovery or FOIA

You need an audit trail, consistent redaction codes, and defensible process. Use purpose-built discovery software.

A dataset rather than a picture

Blurring a screenshot of records does nothing to the records. Anonymise or pseudonymise at the data layer.

Short, guessable values

A PIN, a 4-digit code, a date of birth. The candidate set is small enough to brute-force at any blur strength. Crop it out or cover it with a solid shape.

One more limit worth stating plainly: a blur is a visual layer. The text underneath is still in the page, still selectable, and still readable in DevTools. Blurr protects against eyes, cameras, and screen capture. It is not access control.

Automation

Redact the same things automatically

If you screenshot with real data all day, stop drawing the same blurs. Let Blurr catch them for you before every capture.

Keyword redaction, always on

32 built-in presets frost emails, phone numbers, card numbers, IBANs, SSNs, API keys, and crypto wallets wherever they appear, so a stray value never sneaks into a capture.

Auto-Blur media & tabs

Frost every image, video, avatar, and ad by category, and mask your tab titles and favicons with Tab Privacy, so nothing outside your content leaks into the frame either.

The blurs you draw stay on your device. Only your keyword and Auto-Blur rules sync to your account when you sign in for Pro. The screenshots you take never touch a server.

FAQ

Screenshot redaction: your questions

Can a blurred screenshot be unblurred?
Sometimes, and it depends on how many values the hidden text could plausibly have rather than on how strong the blur is. Blur is a low-pass filter: it spreads information out instead of destroying it. The practical attack is not "unblurring" but brute-force matching, where the attacker renders every candidate string in the same font at the same size, applies the same blur, and compares the result against your image. That works when the candidate set is small, which is why a blurred 4-digit PIN (10,000 possibilities) is recoverable and a blurred paragraph of prose is not. Published tools including Depix and Unredacter demonstrate the technique against pixelation. Use a heavy blur for prose, faces and imagery; crop or cover anything short and guessable.
Is pixelation safer than blurring?
No, and it is usually worse. Pixelation replaces each block with the average colour of a known region, which is precisely the structure a brute-force matcher needs. Both blur and pixelation are obfuscation rather than redaction. A solid opaque fill is the only visual treatment that destroys the information outright.
How strong does a blur need to be?
Blurr’s slider tops out at a 20px blur, which is comfortably past the 12px to 16px text most web pages render, so body copy is destroyed rather than softened. But strength is the wrong dial to reach for when the content is short and predictable. A blurred date, PIN or short code stays recoverable at any radius because the attacker only has to test a small number of candidates. Match the treatment to the entropy of the content, not to how important it feels.
What is the difference between redaction and anonymisation?
Redaction removes specific content from a specific document so it cannot be recovered. Anonymisation irreversibly alters a dataset so no individual can be identified from it at all. They operate at different layers: blurring a screenshot of a customer record redacts the image but leaves the record itself untouched. Under GDPR, genuinely anonymised data falls outside the regulation, while pseudonymised data (Article 4(5)) is still personal data because a key exists that can restore the identifiers.
Why not just draw a black box in an image editor afterwards?
A solid opaque box is the strongest visual redaction there is, because it destroys the pixels rather than rearranging them. The danger is not the box, it is the file around it: editors keep layers, PDFs keep the text object underneath as a separate item in the content stream, and a box nudged aside reveals what it covered. So flatten the image and verify by trying to select the text. Blurring in the page before you capture removes that risk from the other direction, because no original layer is ever written to the file.
Does the original text stay hidden inside the image file?
When you redact in the page before capturing, yes. The screenshot is a flat picture of what was on screen, and what was on screen was already blurred, so the file contains only the frosted pixels. There is no separate original layer stored in the image to be pulled back out.
What should I redact in a screenshot?
Two published standards give you a ready-made checklist. HIPAA’s Safe Harbor method enumerates 18 identifier types (names, geography smaller than a state, dates, phone and fax numbers, emails, SSNs, record and account numbers, biometric identifiers, photographs, and any other unique identifying number) that must be removed before health data counts as de-identified. PCI DSS allows displaying at most the first six and last four digits of a payment card number. Beyond those: API keys, tokens and connection strings, plus the frame around your content, since the address bar, tab titles and notification toasts often leak as much as the page does.
Does my screenshot contain metadata?
Usually yes. Screenshots carry creation timestamps, display scale and colour profile, and some capture tools record the source application or window title. A photo of a screen taken with a phone carries full EXIF, which can include GPS coordinates. Strip it with the Preview inspector on macOS, "Remove Properties" in Windows file properties, or exiftool -all= on the command line. Also rename the file: a name like "acme-corp-invoice.png" identifies your customer before anyone opens the image.
Is a black box over text in a PDF safe?
Not on its own. A PDF stores text as objects in a content stream, and drawing a filled rectangle simply adds another object on top. The text underneath stays selectable, searchable and extractable by copy-paste, which is how several well-publicised redaction failures in court filings happened. Use a real redaction function that removes the text objects, then verify by trying to select the covered area.
Which screenshot tool should I use after blurring with Blurr?
Any native one. Because the redaction already happened in the page, the tool only needs to capture what is on screen. Use Cmd + Shift + 4 on macOS, Windows + Shift + S on Windows, or Shift + Print Screen on Linux. Check where your tool saves: several capture utilities upload the original to their own server to generate a share link, which puts the unredacted file off your machine before you have touched it.

Launching soon on the Chrome Web Store

Blur it before you share it.

Be first to know when Blurr lands. Drop your email and we’ll send you the install link the day it goes live.

  • Free to use
  • No account for core features
  • Chrome, Edge, Brave & Arc