YouTube algorithm recommends low-quality AI videos to children
A New York Times investigation found that YouTube's recommendation algorithm directs children toward low-quality AI-generated videos disguised as educational content, including on YouTube Kids. Over 40 per cent of recommended Shorts contained AI-generated visuals featuring unusual imagery and false information.
What happened
An investigation by The New York Times found that YouTube's algorithm recommends low-quality AI-generated videos to children, presenting them as educational content. The study analysed 1,000 Shorts recommended to young children and found that more than 40 per cent of the content featured AI-generated visuals.
The videos claim to teach basic concepts such as the alphabet, animals, colours and songs, but contain unusual facial features, excessive limbs, chaotic imagery and false information. The same pattern was identified on YouTube Kids, the platform's dedicated app for children under 12.
Reported impact
Children watching established content like CoComelon, Bluey and Ms Rachel are shown large volumes of low-quality AI-generated videos presented as educational material. The videos rely on familiar nursery-rhyme conventions and child-friendly descriptions despite containing poorly made, incoherent or misleading content.
Index analysis
YouTube Kids is marketed as a more controlled environment with parental controls and age-based viewing categories, but the platform's recommendation and classification systems treat low-quality synthetic content as suitable for children. The issue reflects a structural weakness in which parental controls do not address problems in the underlying recommendation engine.
Governance controls
Following the investigation, Google removed channels from the YouTube Partner Program, which enables creators to monetise their content. Various individual videos that violated platform rules were also removed.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.