Generative artificial intelligence (AI) is increasingly being used to provide agricultural advice, but the recommendations it generates may not always support sustainable crop protection. Published in the Journal of Environmental Management in 2026, this short communication by Kris A.G. Wyckhuys and colleagues, funded and carried out by the Food and Agriculture Organization of the United Nations (FAO), assesses five AI advisory systems: ChatGPT, DeepSeek, NormAI, KissanAI and VirtualAgronomist. The researchers evaluated advice for 20 crop–pest systems, using structured prompts submitted across five locations in Europe, Asia-Pacific and the Americas between September and November 2025.

The assessment found that, although most systems recommended preventive and non-chemical measures consistent with Integrated Pest Management (IPM), important differences emerged in how individual practices were prioritized. NormAI, in particular, tended to rank chemical control more highly and gave limited attention to biological control and soil health. The evaluation also identified potentially harmful recommendations, including broad-spectrum insecticides suggested for pest prevention without adequate supporting evidence. Such advice could increase pesticide dependence, undermine beneficial organisms and ecosystem services, and impose additional costs on farmers. The findings are exploratory and reflect a limited selection of crops, locations and AI versions; they do not establish the prevalence of these problems across all agricultural AI systems.

The authors call for greater transparency in AI training data, independent evaluation, effective public governance and appropriate regulatory safeguards. They emphasize that AI-powered agricultural advice should complement, rather than replace, scientific evidence, human expertise and farmers’ place-based knowledge, while giving greater weight to biodiversity, ecological resilience and agroecological approaches.