PR Style
Detects promotional or marketing-style language in academic text, such as calling results 'groundbreaking' or 'revolutionary' — exaggerated claims that AI models frequently generate.
Technical description
Matches against dictionaries of promotional language patterns: superlatives ('groundbreaking', 'revolutionary', 'unprecedented'), impact inflation ('paradigm shift', 'game-changing'), self-promotion ('our novel approach', 'for the first time'), and hype phrases ('opens new avenues', 'paves the way'). Computes a PR density score based on the frequency and intensity of promotional language relative to text length. Sub-check 3 detects negative-parallelism contrast templates in English and Romanian (including the cross-sentence and about variants); sub-check 4 flags rule-of-three (tricolon) overuse by rate.
How it works
Layer 1 (deterministic): Matches text against a dictionary of ~100 promotional phrases and superlatives. Categorizes by intensity (mild, moderate, extreme promotion). Counts PR phrases per paragraph. Flags paragraphs with multiple high-intensity promotional terms. Computes overall PR density score.
Why this matters
AI-generated academic text tends to overstate the significance of findings because language models are trained on a mix of academic and promotional content. Human researchers typically understate their claims due to peer review norms and scientific caution. Excessive promotion in academic text is a red flag for AI generation.
Score thresholds
- 0-1
- Measured, appropriately modest claims
- 2-3
- Occasional promotional language
- 4-5
- Pervasive hype and promotional tone throughout
Limitations
Press releases and grant applications are inherently promotional. Some journals encourage emphasis on significance and impact. Emerging fields may legitimately use stronger language to describe novel findings.