Abstract: The rapid development of generative AI has markedly enhanced marketing professionals’ capacity for content creation. Yet influencer marketing—widely regarded as a key engine for expanding domestic demand and stimulating consumption—has encountered a practical bottleneck: sales conversion rates have remained sluggish as brand and influencer content becomes increasingly homogenized. A review of prior research shows that, although scholars have examined how to improve the effectiveness of influencer marketing from multiple angles, existing work has not sufficiently extended its scope to the problem of optimal distinctiveness in environments where multi-source information coexists simultaneously—namely, where multiple brands, multiple influencers, and multiple pieces of content compete for consumers’ attention within the same social media ecosystem. Put differently, the literature has yet to offer a systematic answer to a central managerial question: when consumers are exposed to a crowded and repetitive stream of social media posts, how similar should a piece of sponsored content be?
Our research reasons that a pathway to overcoming the conversion dilemma is to (1) identify the optimal distinctiveness threshold for multimodal sponsored content and (2) leverage generative AI to produce content that meets this optimal level of distinctiveness. Optimal distinctiveness refers to the point at which content is sufficiently distinct to capture attention and improve consumer decision quality, but not so dissimilar that it undermines comprehension, credibility, or persuasion. Because contemporary influencer marketing content is intrinsically multimodal—often integrating text, audio, images and video—the proposed threshold must be defined and measured at the multimodal level.
To accomplish this, our research first builds on social media affordance theory and uses a focus group interview approach to extract the feature dimensions of sponsored content across key modalities, including textual, audio, and visual elements. These qualitative insights serve as a foundation for constructing a structured, theory-informed representation of what constitutes sponsored content in influencer seeding contexts. The study then applies multimodal deep learning techniques to decompose and model sponsored content, enabling the measurement and characterization of content similarity from multiple perspectives: (a) similarity across modalities (e.g., whether the visual storyline aligns with or repeats the textual proposition), (b) similarity across multiple content dimensions, and (c) similarity with respect to multiple reference entities. By integrating these layers, the study aims to move beyond coarse, one-dimensional similarity metrics and instead provide a fine-grained and operationalizable framework for diagnosing content homogenization in real-world influencer marketing.
Second, the study examines the underlying dual psychological mechanisms through which sponsored content similarity influences consumer engagement, thereby locating the optimal differentiation threshold that can guide generative-AI-enabled content creation. Specifically, the study employs a combination of causal inference methods, behavioral experiments, and eye-tracking experiments to investigate content similarity under two common but theoretically distinct scenarios, organized as two parallel studies. The first study focuses on similarity between campaigns of competing brands, where legitimacy and differentiation are expected to be critical for standing out in a cluttered category environment. The second study examines similarity between different influencers within campaigns of the same brand, where managers often face a trade-off between information processing fluency and advertising intrusiveness. Through an empirical analysis of these dual mediators, the study identifies the threshold at which the net effect on consumer decision outcomes is optimized. We contribute influencer marketing literature by providing a new lens and a set of novel antecedents for influencer marketing effectiveness to refine an integrated analytical framework of influencer marketing.
Finally, the study conducts both laboratory experiments and field experiments to test whether generative AI tools can, in practice, produce optimal distinctive sponsored content. Rather than treating generative AI as a generic productivity tool, the study evaluates it as a controllable content generation mechanism whose outputs can be steered toward an empirically grounded optimal distinctiveness target. The expected outcome is a set of actionable principles and evidence-based procedures for using generative AI to improve influencer marketing performance under real operational constraints.
Overall, our research will enhance the theoretical framework for explaining influencer marketing effectiveness and offer a scientifically grounded, practically feasible solution to a pressing managerial problem. The research aligns with the core needs of the “AI + consumption” business ecosystem in the digital economy era and carries substantial practical value for expanding domestic demand and boosting household consumption.

Key words: influencer marketing, consumer psychology, multi-modal analysis, generative AI, optimal distinctiveness theory