A snapWONDERS report answers two different questions about a file, and gives each its own grade. Keeping them apart matters: a photo can be completely genuine and still give away a lot about you, or give away nothing and still have been edited.
The two grades
Authenticity: has this photo been edited or AI-generated? It looks at where the file came from (camera or editing software), signs it was changed, and signs it was generated by AI.
Exposure: what does this file give away? It looks at what the file reveals about you, such as location, time, device and names, and at anything hidden or unexpected inside it.
Each runs from A (nothing found) to F (serious findings).
Here are both grades for one of our own photos, a footbridge in Melbourne's Docklands, copied straight off an Android phone with location on, with "What does this mean?" opened on each:
Authenticity A: nothing significant suggests editing or AI, though we're less sure about this one (low confidence). Exposure D: the photo gives away where it was taken.
See it as a before and after: two genuine photos, both Authenticity A, one giving away more
Why one bad area decides the grade
Each headline grade is set by the weakest area underneath it, never an average. Think of a car check: good lights and a good accelerator don't make up for failed brakes. If a photo's location data is a serious concern, its Exposure grade says so, however clean everything else is.
What "confidence" changes
Each area of the report comes with a confidence: how sure the checks behind it are.
- High confidence counts in full.
- Medium confidence is softened by one grade.
- Low confidence is softened by two grades. If that area's own grade was D or F, the report also shows a warning so you don't miss it.
- Couldn't tell: some checks can't reach an answer for some files. That's shown, but it doesn't count against the photo.
Softening never turns a finding into an A. An A means nothing significant was found. A few checks are listed at very low weight because they rarely tell a real photo from an edited one; they can appear under an A without changing it.
So a "B with a warning" means: probably fine, but there's a possible serious issue the checks couldn't confirm. Read the warning.
The detail underneath
Below the headline grades, the report breaks the findings into areas: Origin, Manipulation and AI generation (which make up Authenticity), and Privacy and Security (which make up Exposure). Each finding says what was checked, what was found, and why it matters.
The AI-generation card
The AI generation area has its own card: a letter and a short label. Read it as a ladder, from signed proof of a camera at the top to signed proof of AI at the bottom.
- A: Verified Camera Capture. The file carries Content Credentials from a trusted signer, and the last step they record is the camera capture itself. An A needs that signed proof. Nothing else earns one.
- B: Real Capture, Edited. The signed history starts with a real camera capture, and the edits after it were made without AI tools.
- C: No AI-Generation Evidence Found. Nothing pointed either way. This is not a finding that the photo is real. Most genuine photos land here, because most carry no signed proof.
- C: Inconclusive. Something turned up that can't settle the question by itself, such as a pixel analysis flag with nothing to back it up, or credentials that claim a camera capture the history can't confirm.
- D: AI evidence. The label names what was found. "Processed by an AI Image Generator" means a known image-generator watermark was found: the file may be generated, or a real photo edited with an AI tool. "Declared AI-Generated (Unsigned Metadata)" means the file's own details say an AI made it, but nobody signed that statement. "AI-Edited" means signed credentials record a real capture that was later edited with AI tools.
- F: AI-Generated (Certified). Content Credentials from a trusted signer declare the file AI-generated.
A C on this card counts as "couldn't tell", so it doesn't pull the Authenticity grade down. A D or F does.
A signed AI declaration is graded by how far it can be trusted. From a trusted signer it is an F. If the signature checks out but the signer isn't one we can verify, the card reads D "AI-Generated (Signer Unverified)". If the file was changed after it was signed, the card reads D "AI-Generated (Changed Since Signing)" and the report highlights a warning: the AI declaration may have been copied from another file.
The pixel analysis line
Under the card, a pixel analysis line says what a trained check found in the image itself. It doesn't depend on any label, so it can still catch many AI images after one has been stripped: most from older AI tools, fewer from the newest. It reads one of:
- Signs of AI generation detected.
- No signs found. The line adds that this check can miss images from newer AI tools.
- Not yet calibrated for that file type or size. The check says so rather than guess.
The pixel analysis can't grade the card on its own. A flag rules out an A or a B, and with no other AI evidence the card reads C "Inconclusive". It is most reliable on original files. It catches most images from older AI tools but can miss many from the newest ones, and wrongly flags fewer than one in a hundred ordinary camera photos. Very plain scenes, like a bare wall or an empty sky, are more often misread. A low reading, or "no signs found", is never proof that a photo is real.
The evidence table
Below the card, AI evidence, check by check lists every AI check and what it did for this file:
- Content Credentials
- AI declared in file metadata
- Known image-generator watermark
- Pixel analysis
- Statistical estimate (a weak reading)
- Video motion check (videos only, and never part of the score)
Each row gives the result and how strongly it counted:
- Decisive, Strong, Moderate, Weak: how much that check moved the AI score, from most to least.
- Decides the card: the check set the card's letter without moving the score.
- Not counted: the check added nothing for this file, and the result says why, for example "None found" or "None in this file".
So the table tells you which check graded the card, and which ones only nudged the score or didn't count at all.
Here is that part of the report for one of our own phone photos (a different one from the footbridge above), checked on 15 October 2026:
C: No AI-Generation Evidence Found. Nothing pointed to AI, and nothing proved a real capture either. Only the statistical estimate counted at all, and it counted as Weak.
Content Credentials history
When a file carries signed Content Credentials, the C2PA Content Credentials section shows its history. The headline is the last step: what was done to make this file, in which app, when, and who signed it. A line below says where the history began, such as "Started as a camera capture, then edited", and the steps follow, oldest first.
The last step leads because it describes the file in your hands. A photo that started as a real capture and was then edited with AI is an AI-edited photo, and the card says so.
An image we made with an AI generator, checked on 15 October 2026. The last step leads, and the line under it says where the history began.
The originality score (0 to 100)
Inside the report you'll also see an originality score out of 100. It works differently from the grade. It starts in the middle and moves as each check reports: a check that finds nothing wrong can add points, and a check that finds signs of editing, AI or re-saving takes points away. The Authenticity grade, by contrast, goes by your weakest area.
So the two can disagree. If most checks look clean but one area shows clear signs of editing, the score can still be high while the grade is lower. It can go the other way too: a photo sent through a messaging app can lose points for several small reasons while its grade stays good. When they disagree, go by the Authenticity grade. The score is extra detail.
Here's that same photo's originality score:
The same untouched photo that earned Authenticity A scores 47.9 out of 100 here, just under the middle. That's the two measures disagreeing, as described above: go by the grade.
Reading a result sensibly
- A good grade isn't proof. It means the checks found nothing. A screenshot or heavily re-shared copy may simply have nothing left to find.
- A poor grade isn't a verdict. It's a reason to look closer and ask questions.
- Genuine photos often have a poor Exposure grade. That's normal: phones record a lot. It tells you what to remove before sharing, not that anything is wrong with the photo.
Related guides
More guides on
