Artificial intelligence used to spot unhealthy child-targeted food marketing

A new study has shown that artificial intelligence (AI) and computer vision can accurately detect and monitor unhealthy food marketing aimed at children, offering a faster and more efficient way to track potentially harmful advertising practices.

Childhood obesity and diet-related diseases remain pressing global public health issues, fuelled in part by the continued exposure of children to persuasive marketing of unhealthy foods and drinks. Until now, identifying such marketing has relied on labour-intensive manual methods.

Researchers behind the study, titled “Using Artificial Intelligence and Computer Vision to Detect and Monitor Unhealthy Child-Directed Food and Beverage Marketing” have developed a deep learning system that can automatically recognise food products and packaging features designed to appeal to children. The approach could help governments monitor compliance with proposed restrictions on marketing unhealthy foods to young audiences.

The research team trained image classification algorithms on more than 8,000 food package images collected and annotated between 2021 and 2024. Each image was labelled according to a validated “child-appealing packaging” (CAP) coding tool, which identifies 22 different child-directed marketing techniques.

Three machine learning methods, k-nearest neighbours, support vector machines and convolutional neural networks, were tested for their ability to classify child-targeted food products. The convolutional neural network achieved the best results, reaching 90 per cent accuracy and an area under the curve (AUC) score of 0.96 when compared against manually coded data.

In addition, the latest version of the object detection model was fine-tuned to detect specific child-targeted design features such as cartoon characters, bright colours and playful fonts. Although performance varied across different marketing elements, the system proved capable of recognising a wide range of promotional tactics.

Applying this AI system to breakfast cereals, the researchers found that 39.2 per cent of products displayed child-directed marketing features. Of these, 89.5 per cent would be classified as “unhealthy” and therefore restricted under Canada’s proposed marketing-to-children regulations.

By automating the detection of child-targeted food marketing, the study provides evidence to support stronger, data-informed policies designed to protect children from exposure to unhealthy food promotions and, ultimately, to improve long-term health outcomes.