This dataset contains annotated and normalized promotional phrases collected from websites and mobile applications of illegal online lending (PINJOL) platforms explicitly blacklisted by the Indonesian Financial Services Authority (OJK) as of March 2025. Each promotional phrase was processed through a linguistic normalization and thematic synthesis pipeline to produce a list of 33 standardized risk pattern labels, each assigned a Risk Score (1–4) based on frequency-based quartile analysis. The dataset includes: Raw promotional phrases (in Indonesian) Normalized labels (standardized promotional patterns) Risk scores per label Frequency distribution table (as used in Figure 4 of the main study) This resource supports research on AI-assisted legal risk detection, consumer protection, and digital financial regulation. It was generated as part of a computational pipeline combining Indo-LegalBERT embeddings, semantic matching, and Gemini 2.5 Pro–based legal reasoning to evaluate statutory violations in digital financial promotions.