Diagnosing Image-Ad Success: An Interpretable and Scalable Framework for Liking, Sharing, and Memorability
Description
Successful image ads are liked, shared, and remembered. Although the success of an image ad can often be understood with the benefit of hindsight, predicting image ad success ex ante remains challenging. We integrate established ideas from advertising and visual-evaluation research into an ad-level framework that relates five constructs—typicality, complexity, ease of understanding, enjoyment, and interest—to liking, sharing, and memorability. The framework provides an interpretable representation of an image’s aggregate evaluative profile. We assess the proposed structure using human judgment data for 300 Facebook image ads posted by three fast-food chains and estimate a measurement-error corrected path model at the image level. To enable scalable implementation, we operationalize the five constructs using computational measures derived from pixel-level, object-level, and design-level image features. These measures approximate human image-level evaluations and provide automated image-level scores associated with ad success. As a supplementary robustness check, we examine whether a persona-conditioned synthetic consumer panel recovers similar image-level ratings. Agreement is strongest for typicality and complexity and weaker for enjoyment and interest, indicating that synthetic panels can complement, but not replace, direct computational measures and observed human judgments. By combining conceptual integration with interpretable computer-vision operationalization, the proposed framework provides a scalable and practically implementable approach to image ad diagnosis, pretesting, and design iteration.