To verify an image’s authenticity, you need to answer four questions: who made it, when and where it was taken, whether it has been altered, and whether it shows what it claims to. Trace the image back to its original source, check its metadata and any Content Credentials, test the claimed time and place against things you can look up, look for other photos of the same moment, and run an AI check. An image is authentic when those answers line up with the story attached to it.
“Authentic” is broader than “not AI.” A real, unedited photo is inauthentic if it’s presented as showing something it doesn’t: the wrong city, the wrong year, the wrong event.
The four questions of authenticity #
| Question | Methods | Typical failure |
|---|---|---|
| Who made it? | Find the first poster, contact them, check Content Credentials | Aggregator accounts with no original source |
| When and where? | Reverse search, geolocation, weather and shadow checks | Old photos passed off as new |
| Has it been altered? | Compare versions, inspect edges and lighting, AI detection | Object removal, face swaps, composites |
| Does it show what’s claimed? | Cross-check with reporting and other images | Real photo, false caption |
Step 1: Trace the original source #
Every copy of an image is a step removed from the truth. Find the earliest version you can.
- Reverse search with Google Lens, TinEye or Bing. TinEye can sort results by oldest, which helps locate the first appearance.
- Look for the highest-resolution copy. Originals are usually larger than reposts.
- Identify the first person or outlet to post it, and check their history. A photographer or reporter with a track record in that place is a strong source.
- If you can, contact the source and ask for the original file and details of when and where it was taken.
Reverse image search vs AI detector explains what search can and can’t find.
Step 2: Check the metadata and credentials #
If you can get an original file, its metadata may record the camera, the capture time and the software that edited it. Content Credentials (C2PA) go further: a signed record of which device or tool created the image and how it was edited, which breaks visibly if the image is tampered with (C2PA).
Keep in mind that plain metadata can be edited, and that most shared images have had their metadata stripped. A valid credential from a known signer is strong evidence. Missing metadata is not evidence of anything. How to check image metadata for AI shows how to read it and what the signatures mean.
Step 3: Test the time and place #
This is where professional verification does most of its work, and you can do a lot of it yourself.
Where: geolocation #
- Look for landmarks, shop signs, street names, mountain outlines and distinctive buildings.
- Compare with street-level imagery and satellite views of the claimed location.
- Check that the language on signs, the road markings and the side of the road people drive on match the country.
When: chronolocation #
- Check historical weather for the claimed date. Snow, rain or clear skies should match.
- Look at shadows. Their direction and length should fit the time of day and season in that location.
- Look for dated details: storefronts that opened or closed, construction, vehicles, phone models, event posters.
Step 4: Look for alteration #
Even a real photo may have been changed. Look closely at edges where objects meet backgrounds, at shadows and reflections that should match the scene, and at texture differences between regions. If you find another version of the image, compare them side by side, since differences are the clearest evidence of editing. How to tell if a photo was edited with AI covers the details.
Step 5: Check whether it’s AI-generated #
A fully generated image has no real source, time or place, so the first steps often come up empty. That emptiness is itself a clue, but you’ll want a direct check too.
Expose AI handles this step on a phone. Share the image to it, crop out any interface if it’s a screenshot, and it reads Content Credentials, XMP and generator tags, then runs an on-device model trained on the fingerprints of AI generators. It returns Likely real, Uncertain or Likely AI-generated with a confidence meter, and a “How we decided” card showing which evidence made the call. Camera and editor metadata show up there as evidence, never as proof. The image isn’t uploaded, which matters if it’s a private photo or shows a real person.
A detector answers only the “is it generated?” part of authenticity. It won’t catch a real photo with a false caption. How accurate AI image detectors are explains the limits.
Step 6: Find corroboration #
Real events leave many traces. A protest, a fire or a celebrity sighting is usually captured by several people from different angles, and covered by local outlets. If an image claims to show a public event and you can’t find a single other photo, video or report of it, be very cautious.
Putting it together #
| Result | Conclusion |
|---|---|
| Source, time, place and credentials all check out | Authentic |
| Real photo, but older or from somewhere else | Authentic image, false context |
| Original found, current copy differs | Altered |
| No source, no corroboration, detector says Likely AI-generated | Likely AI-generated |
| Conflicting or missing evidence | Unverified; don’t present it as fact |
Frequently asked questions #
What is image authenticity? #
Image authenticity means an image is what it claims to be: made by the stated source, at the stated time and place, without alterations that change its meaning. An image can be a genuine photo and still be inauthentic in context if it’s presented as showing something else.
Can you verify a photo’s authenticity online? #
Yes, for many images. Reverse image search, metadata viewers, the Content Credentials Verify site, map and street-level imagery, historical weather records and AI detectors are all free or cheap. For high-stakes uses, like legal disputes or news publication, professional verification adds original files and source interviews.
Do Content Credentials prove an image is authentic? #
They prove where a file came from and how it was edited, as long as the signature is valid and the signer is trustworthy. C2PA itself says credentials don’t judge whether an image depicts the truth, so a credentialed photo can still be staged or miscaptioned.
How do journalists verify photos? #
They find the original source and contact the person who took the image, geolocate and date it using visual clues and public records, compare it with other images of the event, and check for signs of editing or AI generation. The steps above follow the same approach.