Eye-Tracking in Infants and Young Children at Risk for Autism Spectrum Disorder: A Systematic Review of Visual Stimuli in Experimental Paradigms.

Higgins et al. (2021) looked at every eye-tracking paper that tried to spot autism in babies and toddlers. They wanted to see what pictures, videos, or objects each lab used.

The team found dozens of different setups. Some labs showed faces. Others showed toys or dots. No two used the same rules.

The review says the field is a mess. Stimuli, timing, and scoring change from lab to lab. Because of that, results rarely repeat.

Until labs agree on one standard kit, eye-tracking cannot be trusted as a screen for autism.

Ma et al. (2021) meta-analysis agrees that kids with autism look less at eyes. Their numbers show the signal is real, but small. M et al. explain why: every lab tweaks the task.

da Silva et al. (2025) moved the idea into the real world. They tracked toddlers in nursery rooms and still found a weak but usable effect. Their field data prove the signal survives even with the messy protocols M et al. warn about.

Vernetti et al. (2024) cut the task to 30 seconds of live play. Manual coding worked better than automatic tracking, echoing Falck-Ytter et al. (2015). Both studies back M et al.'s call for tighter, simpler methods.

You can keep using eye-tracking as a red-flag tool, but do not treat it like a blood test. Pick one short, face-based protocol and stick with it. Share your stimuli and code so the field can finally build the shared kit this review says we need.