Read Content Credentials without mistaking provenance for truth
Someone asks whether the website image is "verified" and you've found a little CR icon. Before that turns into "authentic photo" in an approval email, find out what the credential actually records.
What you'll make
a short review note and a publication label for an image, keeping four questions apart: what its credential records, whether that record checked out, whether the picture shows something real, and who has approved it.
What you'll need
the image file itself (the original if you have it, not a screenshot or a copy saved from social media), a Content Credentials viewer in your browser and an AI chat your organisation lets you use. Allow about 15 minutes.
Example result.
History: credential issued by Adobe; generative AI used; created in Firefly.
File checked: the original. The resized web export has no credential.
Shows a real place? No. The reception is generated, not photographed.
Approved for: website hero. Approval by marketing lead still pending.
Label: "AI-generated illustration"1. Check the file in a Content Credentials viewer
Open the Content Credentials Verify site or Adobe’s Inspect tool and add the file. If the viewer finds a credential, it shows a summary: who issued it, the app or tool used, whether generative AI was involved and a history of edits. If it finds none, Verify says “No Content Credential”, and Adobe’s tool offers to search for possible matches.
Check the original file and the export you plan to publish separately. Resizing, converting and downloading from social sites can strip the credential, so the two files can give different answers. Both viewers are public websites: if the image is confidential, check your organisation’s rules before uploading it.
Why this works: a credential is a signed record attached to the file. Only a viewer can read and check it; a filename, a watermark or an AI looking at the pixels can’t.
What you should see: for each file, either a credential summary you can copy into a note, or a clear “none found”.
2. Ask AI to draft the note and the label
Paste what the viewer showed, plus anything you know about where the image came from, into your AI chat:
Here is what a Content Credentials viewer showed for an image, and what
I know about its source: [viewer result and source notes]. The image is
for [intended use]. Write a five-line review note for [approver] with
these lines: History, File checked, Shows a real place or person?, Approved
for, Label. Quote the viewer's wording. Mark anything I haven't given you
as "Not known". Then suggest a short publication label that matches
the evidence.[viewer result and source notes] is the copied summary and anything you know, such as “made in Firefly, resized in Canva”. [intended use] is where it’s going. [approver] is who signs it off, such as “the marketing lead”.
Why this works: the separate lines stop a credential from answering questions it can’t answer. A visible issuer doesn’t mean the scene is real, and a missing credential doesn’t mean the image is fake.
What you should see: the note from the example result, with “Not known” wherever you left a gap.
3. Check the picture for its use
A credential records how a file was made. It says nothing about whether a portrait still looks like the person, whether a product shot shows the real product or whether the person in it agreed. Compare a portrait with its source photo and get the person’s approval. Check a product scene against the product. Add the outcome to the note’s “Approved for” line yourself.
Why this works: these are the questions approvers usually mean by “is it verified?”. Answering them separately stops the credential from being stretched to cover them.
What you should see: a completed “Approved for” line with a name, or a clear “pending”.
Check the result
- Every line of the note matches the viewer’s wording or your own source notes.
- The note says which file was checked, and whether the publishing export was checked too.
- “Shows a real place or person?” is answered from what you know about the source, not from the credential.
- The label describes what the image is, without “verified” or “authentic”.
The usual failure: the AI turns “credential present” into “verified authentic photo”, or a missing credential into “untrustworthy image”. Look for “verified”, “authentic”, “genuine” or “fake” in the draft; the viewer didn’t say any of them.
Your note says [claim]. The viewer result shows only [what it showed].
Correct that line. Keep the credential's history, whether it checked
out, whether the image shows something real and who approved it as
separate lines. Mark anything the evidence doesn't cover as "Not known".Worked example: a reception image for a co-working website
Fictional example. A co-working company’s designer made a reception-area image in Firefly and resized it in another app for the website. The marketing lead asked: “Is this verified?”
Original file (Verify): credential issued by Adobe Inc.
Generative AI used. App: Adobe Firefly. Action: created.
Web export (Verify): No Content Credential
Source notes: generated scene, not our real reception
Approval: marketing lead, not yet givenThe first draft said:
Weak: “Verified authentic image of our reception. Credential confirmed by Adobe. Ready to publish.”
The repair prompt, with “Verified authentic image of our reception” and “an Adobe credential on the original file saying generative AI was used”, produced the note in the example result.
- “Verified authentic” was two mistakes. The credential says the image was generated, so it is evidence against it being a photo of the real reception, not for it.
- The export had no credential. That’s the resize stripping it, not a sign of tampering. The note names both files so nobody assumes the published one carries it.
- Approval stayed pending. Neither the viewer nor the AI can approve an image; the note leaves that line for a person.
Task kit for this job: Download (ZIP) · Reusable instructions
Go deeper
Keep the original and its exports linked
Every resize or format change makes a new file, and the credential may not survive it. Keep the original, the chosen version and each export together, with a line saying which tool made each export. If a destination needs to show credentials, check that exact export in the viewer rather than assuming it inherited them.
Know what Firefly adds automatically
Adobe says it automatically applies Content Credentials to images made entirely with Firefly, such as Text to Image, recording Adobe as the issuer, that generative AI was used and the app involved. Images from other tools may carry credentials, carry none or lose them on export. The Firefly licensing guide covers the usage records that sit alongside them.
Check the picture, not just the paperwork
A clean credential can sit on an image with an extra finger, a garbled sign or a logo that isn’t yours. The AI image giveaway guide shows what to look for before publishing.