Lexplore — AI Eye-Tracking Reading-Risk Screening, Tested Against Independent Scrutiny

Lexplore grew out of the Kronoberg Project, a 1980s Swedish longitudinal study of children's reading development. Karolinska Institutet researchers Gustaf Öqvist Seimyr and Mattias Nilsson Benfatto rediscovered the archived eye-movement recordings and, with KI Innovations support and Vinnova funding, reanalysed them with machine learning; a 2015–2016 scaling study of 3,444 Swedish pupils in grades 1–3 reported 86% accuracy in flagging reading-risk profiles from eye-tracking data alone.

The service now screens classrooms in Sweden (including a citywide rollout in Lidingö and use by Stockholm's municipal education board), the UK and the US: Karolinska Institutet and the education nonprofit HundrED independently confirm more than 40,000 children screened, with roughly 10% flagged for specialist reading support.

Evidence is mixed. Lexplore's own site cites 95–97% accuracy from an unpublished 2018 University of Memphis study with no public peer-reviewed citation. The one fully independent academic test — University of Central Florida and University of Kansas researchers, Frontiers in Education, 2026 — examined just 29 children at a single US charter school. It found statistically significant, large differences in fixation duration between struggling and typical readers, but explicitly stated it did not validate Lexplore's AI classification model, and warned that eye-tracking data alone should not be used to diagnose a reading disability, citing calibration drift and the opacity of the underlying proprietary algorithm.

Read the full analysis: https://lexplore.com/en-gb/research-reading-ai