A photograph of a dramatic storm appears on your social media feed. People are sharing it as breaking news, and the picture looks completely believable.
A photograph of a dramatic storm appears on your social media feed. People are sharing it as breaking news, and the picture looks completely believable. There is just one problem: the storm never happened. Artificial intelligence generated the image in seconds.
AI image tools can generate realistic faces, landscapes and events from simple written instructions. This opens creative opportunities for artists, teachers and advertisers. A children's author could illustrate an imaginary world without spending weeks drawing every scene. However, the same technology can create misleading content that appears to be authentic evidence.
How good are we at spotting the difference? In a 2025 Microsoft Research experiment, more than 12,500 people made approximately 287,000 judgements about real and AI-made pictures. They identified images correctly 62% of the time. Participants performed better with portraits than with some landscapes and city scenes. The result is not a universal score for all people and images, but it demonstrates how convincing the technology can be.
Sometimes people look for unusual fingers, impossible reflections or strange lettering. These clues can help, but improved tools may produce images without obvious errors. A picture that looks convincing is not necessarily trustworthy, and a real photograph may also be cropped or presented out of context.
Before reposting a surprising image, check who first published it, whether a trustworthy source confirms the story, and whether independent evidence exists. AI itself is not automatically harmful. The key question is how an image is made, labelled and used. A little scepticism is healthier than believing everything—or dismissing everything—as fake.
The stakes become higher when synthetic imagery appears during elections, emergencies or conflicts. A fabricated image may influence emotions even after a correction is published. Meanwhile, genuinely documented events can be dismissed with the convenient claim that the pictures were generated by AI. Both possibilities weaken public trust.
Watermarks and content credentials may provide useful information about an image's origins, although they are not universal solutions. Screenshots, editing and repeated uploads can remove identifying information. Reverse-image searches and reputable reporting are therefore complementary tools, not guarantees. The best response is to investigate a claim rather than simply hunt for visual imperfections.
Available levels: A1 · A2 · B1 · B2 · C1 · C2