Spotting AI images by eye is getting harder because the best generators now draw hands, faces and text well. The most reliable approach is to check where the image came from: look for Content Credentials or watermarks, run a reverse image search, and confirm the event with trusted news sources, rather than hunting only for visual glitches.
Why is it harder to spot AI images now?
Top models like OpenAI’s GPT Image 2.6 and Google’s Nano Banana 2.1 produce realistic photos in seconds, as shown in our image generator comparison. The old tells, like six fingers or melted text, are much rarer than a year or two ago.
What are the best checks to spot an AI image?
- 1. Find the source. Who posted it first? An anonymous account with no history is a red flag
- 2. Reverse image search. Use Google Lens or TinEye to find earlier copies, the original or fact-checks
- 3. Look for Content Credentials. Many tools attach C2PA “CR” data showing how an image was made; check with a Content Credentials viewer
- 4. Check for invisible watermarks. Google marks its AI images and video with SynthID
- 5. Read the text. Signs, labels and tiny print are still sometimes garbled
- 6. Check hands, teeth, ears and jewellery. Mismatched earrings or odd fingers still slip through
- 7. Look at the background. Warped lines, repeating faces in crowds and impossible architecture
- 8. Check light and shadows. Do shadows fall the same way for every object?
- 9. Ask if the story makes sense. Is any reliable outlet reporting the event?
Are AI image detector tools reliable?
Not on their own. Online detectors give probabilities and can be wrong both ways, labelling real photos as fake and missing AI ones, especially after an image is cropped, compressed or screenshotted. Use them as one clue among several, never as proof.
Do metadata and watermarks prove an image is real?
They help but have limits. Content Credentials can be stripped when an image is re-saved or screenshotted, so a missing label does not mean an image is real. A present label from a trusted tool is good evidence about how it was made. New EU transparency rules require AI-generated content to be marked, with the deadline for some existing systems falling in December 2026, which should make labels more common.
What should I do if I am not sure?
Do not share it. If it shows a real person doing something shocking, wait for confirmation from reputable news outlets. Our guides to fact-checking AI content and deepfakes cover video and audio too, and AI voice scams shows how fakes are used in fraud.




