Detecting freebooted content in social media ads: multimodal provenance and e-commerce implications

Abstract

This study examines the phenomenon of content freebooting on social media and its exploitation for marketing counterfeit and “dupe” products. Using a four-week dataset of TikTok ads linked to 32 distinct e-commerce domains, we develop and evaluate a multimodal provenance pipeline—combining perceptual hashing, audio fingerprinting, vision embeddings, and natural-language clustering—applied to 54 ads, 180 landing pages, and over 3,000 extracted video frames. The primary contribution is methodological: multimodal late-fusion substantially outperforms single-modality detectors in identifying copyright-infringing reuse of creator content under adversarial transformations. Empirically, we document systematic asset theft from legitimate fashion creators, with several videos and review images reappearing across more than 10 separate domains. Purchases from three advertised shops, alongside control items, reveal systematic misrepresentation of product quality and unreliable fulfillment, situating freebooted ads at the intersection of copyright infringement, trademark-like “dupe” positioning, deceptive advertising, and consumer fraud. Network analysis of ad handles and domains indicates a coordinated cluster of shell actors, with a median time-to-reupload of 18 h. As a secondary contribution, the study uses this provenance pipeline to illuminate how freebooted cultural assets are rapidly converted into counterfeit-linked sales, and to surface gaps in platform integrity and consumer protection. By integrating computer vision, audio analysis, and NLP techniques with network and fulfillment audits, the paper offers both a methodological framework for analyzing freebooting pipelines and socio-technical insights for platform governance in digital commerce.