A few years ago, AI images were easy to spot: six fingers, melted faces, garbled text. Today’s image models are far better, and many AI pictures are impossible to identify at a glance. There is no single test that always works, but combining a few checks will catch most fakes and, just as important, stop you sharing them.
Step 1: Look at the context first
Before zooming in on pixels, ask simple questions:
- Who posted it, and are they a reliable source?
- Is any major news outlet reporting the same event?
- Does the image seem designed to make you angry, scared or amazed? Viral fakes usually are.
Most fake images are caught by context, not by spotting technical flaws.
Step 2: Check the details AI still gets wrong
- Text: signs, labels, licence plates and book covers can still contain gibberish or warped letters, even though newer models are much better at text.
- Hands, teeth and ears: extra or merged fingers, too many teeth, mismatched earrings.
- Backgrounds: people in crowds with blurred or repeated faces, buildings with impossible lines, objects that blend into each other.
- Lighting and shadows: shadows going in different directions, reflections that do not match.
- Too perfect: airbrushed skin, flawless symmetry and a glossy “stock photo” look.
None of these prove an image is AI-made; real photos have odd details too. Treat them as reasons to look closer.
Step 3: Run a reverse image search
Upload the image, or paste its link, into a reverse image search such as Google Lens or Bing Visual Search. You may find the original photo, an earlier version, the account that first posted it, or fact-checks that have already investigated it.
Step 4: Look for Content Credentials
A growing number of cameras, editing apps and AI image generators attach Content Credentials to files. They are based on an open standard from the C2PA, a group of technology and media companies, and record how an image was created and edited, including whether AI was used. You can upload an image to the verification tool at contentcredentials.org to see if it carries this information.
The limitation: credentials can be stripped when an image is screenshotted or re-uploaded, and many tools do not add them yet. If an image has credentials, they are useful evidence. If it has none, that tells you nothing either way.
Step 5: Be careful with “AI detector” websites
Online AI image detectors can be helpful, but they are not reliable enough to settle a question on their own. They produce both false alarms and misses, especially on compressed images from social media. Use them as one signal among several, never as proof.
What to do if you are not sure
- Do not share it, or share it with a clear note that it is unverified.
- Wait for reliable outlets to confirm the event.
- If the image shows a real person in a damaging or explicit situation, report it to the platform instead of spreading it.
The same goes for voices and video
AI voices and video are improving just as fast. New voice models, like the human-sounding Dessa model from Block, show how natural synthetic speech is becoming. Our guide to AI voice-clone scam calls covers how to protect yourself on the phone.
Sources
- Content Credentials: Verify Media Authenticity — Content Credentials
- Coalition for Content Provenance and Authenticity — C2PA