You're Trying to Spot AI-Generated Faces Wrong

Just a couple of years ago, spotting an AI-generated face was easy; you just had to look for an obvious deformity. But in 2026, the machines have gotten better.

The sad reality is that some fake images have fooled even the sharpest observers. But a report yesterday from Scientific American encourages people to search for these facsimiles differently.

Instead of looking for an obvious defect, such as six fingers or a floating earring, people should instead focus on six perceptual qualities: distinctiveness, memorability, proportionality, symmetry, attractiveness, and expressiveness.

“Our training directs people’s attention to global qualities that differ between AI and human faces,” says Amy Dawel from the Australian National University (ANU). “AI faces tend to be more symmetrical, proportional, and attractive, but without training, we often think these are markers of being human.”

Dawel, who led the research at the ANU’s Emotions and Faces Lab, says in a press release that by training people to look for these broader patterns led to remarkable results — with some participants getting 100% accuracy.

“It was amazing to see the dramatic improvement in people’s ability to detect AI faces,” Associate Professor Dawel says.

“We’ve shown our training is effective for some of the most convincing fakes available, StyleGAN faces. Now we need to find out whether that training generalises to other AI-generated faces.”

The thing about AI faces is that they are created from a massive database of photos that were mostly stolen from the web — without authorization. What the AI does is use all of this data to construct a statistically average human face.

An AI face tends to be unmemorable and less expressive, while also being in perfect proportion and symmetrical. As AI models improve, this is what you must look out for, rather than a glaring error.

“We found that even relatively short training sessions helped participants improve their accuracy in detecting AI-generated faces, highlighting the potential for practical education tools in this area,’’ says ANU Honors student Tanya Georg, who trained the participants.

“AI image generation technology is improving extremely quickly, and many people underestimate how convincing these faces can be. Research like this can help people navigate increasingly complex online environments.”

Image credits: Header photo licensed via Depositphotos.