Researchers at the University of Aberdeen and the Australian National University have shown that people can be trained to recognise AI‑generated faces, raising identification accuracy from about 40% to 80% after roughly an hour of training. The study used thousands of faces created with the StyleGAN3 image generator and tested participants before and after instruction on six visual cues: symmetry, proportionality, attractiveness, distinctiveness, expressiveness and memorability. Training highlighted subtle differences such as atypical eye shape, uneven facial proportions, overly attractive features, generic appearance and reduced emotional expression. Participants improved their accuracy, with most reaching around 80% and a few achieving close to 100%. Confidence in identifying deepfakes also increased after training. Detecting AI‑generated images is valuable for preventing fraud, such as the £25 million transfer scam that used a deepfake video of a Hong Kong executive, and for combating political misinformation. Ongoing practice is expected to enhance detection skills, though no single definitive cue guarantees reliable identification.