A Barron's analysis found a 4x increase in AI-typical sentence structures in corporate filings and press releases between 2023 and 2025.
Barron's analyzed AlphaSense's database of corporate news releases, earnings reports, and government filings and found the phrase construction 'not just X — it's Y' increased from ~50 mentions in 2023 to over 200 in 2025. This structure is widely recognized as a hallmark of AI-generated text. Examples span Cisco, Accenture, McKinsey, Workday, and Microsoft. Pangram CEO Max Spero confirmed the pattern is a strong signal of AI use in corporate documents, though not definitive proof on its own.
This is less a coding problem and more a data quality signal. If your product ingests corporate filings, earnings calls, or press releases as training or context data, you're now working with a corpus that's increasingly AI-generated — with detectable stylistic drift. The 4x increase in one phrase pattern suggests AI contamination of 'ground truth' text sources is accelerating faster than most NLP practitioners assume.
Run a grep or regex scan across your training corpus or RAG document store for the pattern 'not just.*it's' and quantify what percentage of your documents match — if it's above 5%, flag those documents for quality review before your next fine-tune.
Open a Python environment or Jupyter notebook with access to your document corpus
Tags