<p>Prolonged exposure to fine airborne particles has a significant impact on health, natural ecosystems, climate, and the economy. This study examines how citizens perceive air quality in a Latin American megacity, such as Bogotá, through natural language processing techniques and correlates their perceptions with data recorded by the Bogotá Air Quality Monitoring Network, focusing on three air quality categories: good, moderate, and low. Initially, a thorough preprocessing of Instagram posts and online news was executed. Next, Bidirectional Encoder Representations from Transformers models were fine-tuned using synthetic data from Claude 3.5 Sonnet for sentiment classification. Latent topics within the texts were identified using Latent Dirichlet Allocation, and the correlation between PM_2.5 levels and social media posts and news articles was calculated. The findings reveal that (1) public sentiment on Instagram and in online news does not consistently reflect particulate matter _2.5 pollution levels. (2) The Pearson correlation between air quality and social media mentions was 0.20 for low quality, 0.68 for moderate quality, and 0.10 for good quality. For news articles, the correlations were 0.27, 0.25, and 0.13, respectively. This research demonstrates the capabilities of natural language processing in enhancing our understanding of online environmental discussions, providing insight into how citizens perceive specific issues. Such insights can help government entities make more informed decisions.</p>