Speaker
Description
Authors: Carine Mutatayi(French Monitoring Centre on Drugs and Addiction (OFDT)), Duong Vu (Alive & Thrive (A&T)), Roger Mathisen (Alive & Thrive (A&T))
Background: Since its legalization in France in 2010, online gambling has continuously increased among teenagers (14,7% of gamblers aged 17 in 2011 vs 27,9% in 2022), despite the legal ban on sale, free offers, and advertising to minors. This growth was particularly marked in sports betting (SB) and coincided with great advertising pressure. In 2025, an AI-powered system (AGATTA) was developed based on the VIVID solution by the French Monitoring Centre for Drugs and Addiction (OFDT) and Alive & Thrive (A&T) to identify digital SB advertising and potentially harmful content.
Methods: AGATTA explores SB advertisements and classifies content according to relevant criteria. These criteria are defined through human analysis, then detection and categorisation are implemented by machine learning, with final human validation. First, we reviewed the recurring features of gambling advertising that international research has shown to be attractive to minors and compulsive gamblers. Specific work was then conducted to translate certain abstract criteria (e.g., the world of minors) into factual or objective elements to build the AI algorithm.
Results: The selected criteria are evidence-based and help assess whether the content of advertisements is likely to breach French law. A functional AI tool was tested, classifying advertisements according to the defined criteria, particularly those relating to the legal ban, such as a strong appeal to minors or excessive gamblers, the staging of people resembling minors, or the omission of the government warning message. The first exploration episodes helped test the AI tool's performance.
Discussion: Using an AI-based solution to explore and analyse gambling advertising requires certain legal and methodological precautions, particularly if operators do not have jurisdictional powers. These considerations (e.g., respect for copyright) influence the public or confidential status of the AI tool and the labelling of results. Training an AI solution to encompass abstract notions/criteria presents methodological challenges.
| Conflict of interest | No disclosure |
|---|
