How to Verify Metadata Removal

Removing metadata from a photo is only half the job. The other half is confirming that the intended fields are gone and the output still behaves like a healthy image. A practical verification workflow uses three layers: the built-in output re-scan, an independent metadata reader, and visual or structural sanity checks.

The three-layer verification model

No single check answers every question. Stack these three layers instead:

1. Built-in output re-scan. After the AI Metadata Remover strips supported containers from an image, it scans the new copy again. The report separates removed, preserved, unsupported, and unverified results. This is the fastest first check, but the same scanner is inspecting its own output.
2. Independent metadata reader. A separate tool, ideally ExifTool, reads the output without knowing what was supposed to happen. This is the strongest single cross-check for ordinary file metadata.
3. Visual and file sanity checks. Open the new copy, confirm its dimensions and appearance, and make sure expected structural information remains. These checks are better at catching corruption than hidden metadata.

The confidence comes from agreement between the layers, not from a green badge alone.

Layer 1: Read the built-in report

The result report is deliberately more specific than “done.” Review four areas:

• Removed: supported EXIF and GPS, IPTC, XMP, PNG text, AI workflow, or other metadata containers that were found and stripped.
• Preserved: the compressed image data, color profile information, and an optional minimal JPEG Orientation tag when you chose to keep it.
• Unsupported: a recognized structure that the current cleaner does not handle.
• Could not verify: a result the scanner cannot responsibly confirm.

For the homepage image path, JPEG scan bytes, PNG IDAT chunks, and WebP image chunks are checked byte for byte. The original is never overwritten. A new copy is produced and inspected.

The report does not promise zero metadata or universal forensic sanitization. It reports the supported scanner categories and leaves uncertainty visible.

If you need to share the result, use the share-safe report. It omits the original file name, metadata values, and the actual SHA-256 digest.

Layer 2: Cross-check with ExifTool

ExifTool is a widely used open-source command-line utility for reading image metadata. It provides a useful independent view because it is not part of the browser cleaner.

Install ExifTool

• macOS: install with Homebrew using brew install exiftool.
• Windows: download the standalone executable from exiftool.org, then run it from its folder or add it to PATH.
• Linux: use your distribution's package manager, for example sudo apt install libimage-exiftool-perl on Debian or Ubuntu.

Run the check

Open a terminal in the folder containing the cleaned image and run:

exiftool -G1 -a -s clean-photo.jpg

The options make the output easier to audit:

• -G1 groups tags by metadata family.
• -a includes duplicate tags instead of hiding them.
• -s uses short tag names that are easier to search in documentation.

For a stronger comparison, run the same command on the original file and the cleaned copy, then place the outputs side by side.

Interpret what remains

A cleaned file will still produce ExifTool output. That is expected. Start by separating structural data from personal or workflow metadata.

Structural and display fields that commonly remain include:

• FileType, MIMEType, ImageWidth, ImageHeight, BitDepth, ColorComponents, or encoding details.
• ICC profile entries that describe how color should be displayed.
• A minimal EXIF Orientation value from 2 through 8 when Keep orientation tag was enabled.

These describe the file and its rendering. Their presence is not automatically a privacy failure.

Fields that should no longer appear when the corresponding supported container was removed include:

• GPS latitude, longitude, altitude, and related location tags.
• Camera, lens, serial number, software, and timestamp fields from the removed EXIF container, except the optional Orientation value described above.
• IPTC captions, bylines, contact details, and keywords.
• XMP creator, editing history, workflow, and descriptive fields.
• PNG text values used for comments or AI workflow exports.

If one of those groups still contains an original value, do not infer success from the file size or from another empty field. Return to the result report, confirm the file was supported, and investigate the remaining container.

C2PA and JUMBF need different wording

The browser scanner performs conservative C2PA/JUMBF marker detection. It does not cryptographically validate a manifest or retrieve a remote manifest. An independent scan can help confirm whether supported marker structures are still detected, but marker absence does not validate the former credential, its signer, or its history. The result must remain framed as a marker-level observation.

Layer 3: Quick system and visual checks

Operating-system viewers expose only a subset of metadata. They are useful for a quick spot check, not a complete audit.

macOS Preview

1. Open the cleaned image in Preview.
2. Choose Tools > Show Inspector, or press Command-I.
3. Open the information panel and look for EXIF or GPS tabs.
4. Compare the same panels in the original and cleaned files.

If an original GPS panel disappears, that is a useful signal. Preview may still omit XMP, PNG text, workflow fields, or provenance structures, so use ExifTool when the stakes are higher.

Windows File Explorer

1. Right-click the cleaned image and choose Properties.
2. Open the Details tab.
3. Check camera, GPS, author, and date rows against the original.

Blank rows are encouraging, but File Explorer also shows only a subset of the file's metadata.

Open and compare the image

Open both files and confirm that the cleaned copy has the expected dimensions, orientation, transparency, and color appearance. The homepage cleaner preserves compressed image data for supported JPEG, PNG, and static WebP files, while keeping display-critical structures such as ICC data. If the file will be used in a sensitive workflow, open it in the application that will actually consume it.

Common false positives

• The file is only slightly smaller. An original with little metadata may change by only a small number of bytes. File size alone proves nothing.
• The file happens to have the same displayed size. Rounded size labels can hide small byte differences. They still do not describe which tags remain.
• ExifTool still prints many lines. Container, encoding, dimensions, and color information are normal output.
• An ICC profile remains. The homepage cleaner preserves color profile information by design. ICC data is not automatically personal metadata.
• Orientation remains. A minimal Orientation tag can be intentionally preserved so the image displays correctly. The option is visible before cleaning.
• The whole-file hash changed. Removing a container changes the file bytes. A changed SHA-256 status does not say anything about visual classifiers, perceptual matching, or pixel watermarks.

What this verification does not cover

File-metadata verification has a clear boundary:

• Pixel-level watermarks, SynthID, and steganographic signals live outside ordinary metadata containers.
• Perceptual hashes, reverse image search, and visual classifiers work from image content.
• Platform-side labels may use metadata, account context, visual analysis, or other signals you cannot audit from the file alone.
• A PDF follows a separate document-properties workflow even though the homepage can route a detected PDF to it.
• No supported-file report is a universal forensic guarantee.

Use the Image Metadata Viewer when you want a read-only look at the original before cleaning.

Reusable verification checklist

• [ ] Review the built-in output re-scan and note removed, preserved, unsupported, and unverified items.
• [ ] Run exiftool -G1 -a -s clean-photo.jpg on the new copy.
• [ ] Confirm sensitive GPS, identity, timeline, and workflow values are absent.
• [ ] Treat ICC, dimensions, encoding details, and an intentionally preserved Orientation tag as expected structural or display data.
• [ ] Open the output and confirm its dimensions, orientation, transparency, and appearance.
• [ ] For sensitive use, repeat the check with an independent machine or reviewer.
• [ ] Keep the original. The cleaner never overwrites it.

For the project's repeatable corpus and browser results, review the test evidence. More evidence-led references are available in the guides hub.