Media Engagement in Health and Climate Change

This indicator analyses online news coverage of climate change and health using articles collected from 306 news sources across 75 countries, covering 40 languages and the period 1 January 2021 to 31 December 2025. Articles were collected mainly through NewsAPI, with additional web scraping through ScrapAI where sources were not indexed or collection costs were high. This indicator uses a multilingual large language model approach to classify news articles on climate change and health. Each article is assigned labels covering climate change, health, health effects of climate change, and health effects of extreme weather. A large annotated sample was first classified using a frontier language model, then used to fine-tune a smaller multilingual model for classification across the full corpus. The model was validated against held-out model annotations and a human-coded sample, showing strong agreement. A Bayesian hierarchical model was then used to estimate the prevalence of climate-health news coverage across countries, regions, climate zones, HDI groups, and time.

The media engagement estimates should be interpreted with caution because country and source coverage is incomplete. The sample includes 75 countries and up to five online news sources per country, which may underrepresent lower-income countries, regional outlets, and non-major-language media. Some countries also have fewer than five sources, affecting comparability. On the model side, validation was based on a relatively small set of annotated articles, with very few positive examples for some key labels, so some classification scores are uncertain. Finally, results by HDI group, especially for Low and Medium HDI countries, should be interpreted carefully because these groups include relatively few countries in the sample.