R&D narrative disclosure and semantic similarity: evidence from China

How do firms strategically craft R&D narratives to shape innovation perceptions? This paper examines the widespread yet underexplored phenomenon of templated R&D disclosure—where companies mimic their own past language or imitate peers—to manage regulatory scrutiny and investor expectations. Drawing on data from 25,901 firm-year observations in China from 2008 to 2023, we use natural language processing techniques to quantify semantic similarity in annual reports and link these patterns to firm-level R&D inputs and outputs. We distinguish between two forms of templating: self-similarity, where firms reuse their own prior disclosures, and peer similarity, where they replicate language from industry counterparts. Our results reveal a sharp behavioral divide: firms with genuine R&D investments tend to display higher self-similarity, consistent with cautious, consistent disclosure aimed at protecting intellectual property. In contrast, firms near the R&D spending thresholds for high-tech certification exhibit elevated peer similarity and poorer innovation outcomes, indicating strategic mimicry to gain policy benefits. Moreover, we show that peer similar firms underperform in invention patenting over time, while self-similar firms sustain higher-quality innovation. These effects are moderated by regional intellectual property protection, emphasizing the role of institutional context. This study offers a novel framework to detect pseudo-innovation behavior through narrative analysis and contributes to our understanding of how firms adapt disclosure strategies amid evolving economic policies. Our findings hold implications for regulators, investors, and policymakers seeking to distinguish substance from signal in innovation reporting.