The prominence of sustainability in corporate messaging contrasts with the limited empirical work on how short-form social media activates public engagement with environmental concerns. This study addresses this gap by examining how National Geographic uses Instagram Reels to communicate sustainability messages and enhance viewer interaction. The challenge lies in identifying which visual and narrative elements drive engagement an insight critical for improving the effectiveness of environmental communication strategies. We analyzed 3 months of Instagram Reels posted by National Geographic. Each video was systematically coded for scene type, number of people, emotional tone, text layover, sound, color, and speech. Using Random Forest regression, classification, and K-means clustering, we identified key predictors of engagement. Results show that scene type, people count, and emotional tone are the strongest predictors. Regression explained 19% of the variance in engagement, while classification achieved 66.7% accuracy across engagement tiers. Clustering revealed three content groups, with the most successful cluster characterized by emotionally rich, people-centered, and dynamic videos. These findings suggest that emotionally resonant storytelling enhances sustainability communication. The study contributes to Legitimacy Theory and Uses and Gratifications Theory, while also demonstrating the value of AI-based methods for evaluating and optimizing digital environmental messaging.