A Multimodal LLM Architecture and Triangulated Index for Detecting Rhetorical Drift in Corporate Sustainability Reports

• Abi cit

  Lautech

  Author

Corporate sustainability reporting has become a critical channel for stakeholder communication, yet the phenomenon of "rhetorical drift"—systematic shifts in language and imagery that obscure deteriorating environmental performance—remains inadequately addressed by existing detection frameworks. This study addresses this gap by developing a multimodal large language model (LLM) architecture integrated with a triangulated index for detecting rhetorical drift in corporate sustainability reports. Drawing on a corpus of 1,247 sustainability reports from 312 firms across seven sectors (2018–2025), we combined vision-language processing of document layouts and imagery with BERT-based linguistic analysis and verified ESG performance metrics. The architecture achieved 89.4% detection accuracy for regulator-sanctioned firms exhibiting rhetorical drift, outperforming text-only baselines by 23.7 percentage points. The triangulated index demonstrated strong convergent validity with stakeholder credibility assessments (r = .71). Our findings establish that rhetorical drift operates through coordinated multimodal mechanisms—lexical inconsistency, visual-narrative misalignment, and omission of negative performance indicators—rather than isolated textual signals. The framework provides practitioners with an auditable tool for sustainability governance and advances theoretical understanding of strategic environmental communication as a multimodal phenomenon.

• pdf

Copyright (c) 2026 Abi cit (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.