Explainer · Last reviewed 2026-10-26 · Updated 2026-10-26

How to tell if a photo was edited or AI-generated

Short answer: no single check proves it. Find the original, compare the two, look for mistakes in the details, read any provenance data, and treat AI detectors as a hint. If you cannot find the original, call it unverified.

1. Start with where the photo came from, not how it looks

Find who posted it first and when, and look for the original. A photo with no named source is weak evidence however convincing it looks. See the guide to finding the original source (guide).

2. Look closely at the details

Edits and AI images often show hands and fingers that do not add up, lettering and jewellery that does not read, shadows that disagree about where the light is, edges that are too sharp or too soft, and patterns that repeat. These are hints, not proof: real photos have compression and blur, and good edits and newer AI tools can avoid every one of these signs.

3. Compare it with the original, side by side

If you can find the original, put the two at the same size and compare faces, text, backgrounds and edges. In our case files the difference is almost always visible once the original is next to the copy.

4. Check provenance data, but do not rely on its absence

Content Credentials (the C2PA (opens in a new tab) standard) can record where a file came from and how it was edited, and can be read with the Content Credentials verify tool (opens in a new tab). EXIF metadata (opens in a new tab) can record the camera and time. Many sites and apps strip both when a photo is uploaded, and EXIF can be edited, so a missing record proves nothing.

5. Treat AI-image detectors as one hint

Some AI tools add invisible watermarks, such as Google’s SynthID (opens in a new tab), that only the maker’s own detector can read. General “is this AI?” websites give both false positives and false negatives. Use them as one clue next to the original source, never as the verdict.

6. Say what you cannot tell

If you cannot find an original and the image shows these signs, the honest label is “unverified”, not “fake” or “real”. We say so on a case file when we could not confirm something.

What this looked like in our case files

Case file	What was changed	How we showed it
FLR-002: “young Barack and Michelle”	Eyes and hair darkened, shadows added (Snopes)	Beside Barack Obama’s own post of 2019-01-17
FLR-003: Christmas-tree photo	Her face was doctored to look male (Snopes)	Beside the archived campaign photo, “Christmas in Hawaii”
FLR-006: the necklace	Letters changed from VOTE to MIKE	Beside a frame of her 2020 convention speech
FLR-010: face swap	Her face and her brother’s swapped (Lead Stories)	Beside the podcast promo of 2025-03-12
FLR-011: family photo	A striped building added (AFP, Lead Stories)	Beside the 2020 photo she posted; faces blurred by us
FLR-014: “1985” boat photo	An AI-edited copy of an earlier AI image (Snopes)	Beside a real 1985 yearbook portrait

Sources for this guide

Coalition for Content Provenance and Authenticity (C2PA) (opens in a new tab): the open standard behind Content Credentials
Content Credentials: verify a file (opens in a new tab)
ExifTool (opens in a new tab) documentation of EXIF and other photo metadata
Google DeepMind: SynthID (opens in a new tab) on invisible watermarks in AI-generated media

Change log

Date	Change
2026-10-26	First version.

This site is independent. It is not affiliated with, endorsed by, or connected to Michelle Obama, Barack Obama, the Obama Foundation, or any government office. To report an error, see the corrections page.