Low-level visual and auditory features predict preferences of commercials

Video content has become a dominant form of digital marketing, increasing the need to understand which features make videos more appealing and effective. Prior studies showed that low-level visual features in static images affect saliency and consumer preferences, while low-level auditory features influence preferences for music clips. However, it remains unclear how these features and their interaction shape preferences for dynamic stimuli such as video commercials. In this study, 165 participants watched 100 video commercials and rated how much they liked each commercial and their willingness to pay (WTP) for the advertised product. We extracted low-level visual and auditory features from the commercials (e.g., colorfulness, scene frequency, zero-crossing rate) and used them as inputs to linear and non-linear prediction models. Both visual and auditory low-level features significantly predicted Liking and WTP, with the combined model performing best. Importantly, these low-level features also predicted the aggregate engagement metrics at the population level as measured on YouTube, demonstrating the generalizability of our findings and their external validity. These findings show that low-level sensory information influences consumer preferences and ad effectiveness and demonstrate how computational tools can be used to evaluate video effectiveness.