In order to communicate brand concepts and values to the young generations, many brands are highly active on social media platforms such as Twitter, Facebook, and Instagram. Brand-Generated Content has already become the most common marketing strategy for fashion brands and plays a significant role in influencing consumers’ purchase intention. With increased competition on social media platforms, companies need to understand which posting features can bring in more consumer engagements such as number of likes and comments on social media platforms. In this paper, we develop (1) a model to predict the number of likes and (2) a methodology to detect anomaly of posts that have unusually high percentage of negative user comments based on Instagram design variables, including semantic text meanings, facial expressions, color scheme and background of photos, and post timing, among others. We collected a data set of brand-generated Instagram posts from ten fashion brands. The data covers the image, text, and user comments posted between 2019 and 2020. Image features were extracted using Convolutional Neural Network, and text topics were generated through Latent Dirichlet Allocation. Our results will help managers design Instagram posts to increase consumer engagement and to reduce negative consumer reactions.