The 'Not X, But Y' Trap: Why AI Writing Sounds So Formulaic
Key Takeaways
- ▸"Not X, but Y" constructions appear three times as often in AI-generated text as human writing
- ▸Usage in corporate communications increased more than fourfold from 2023 to 2025
- ▸All major AI models (ChatGPT, Claude, Gemini, open-source) exhibit this pattern, but only OpenAI is publicly working on a fix
Summary
Across the AI industry, a distinctive rhetorical device has become so prevalent that it's become a hallmark of artificial writing—the "not X, but Y" construction. Once obscure, this pattern now appears in everything from corporate communications to AI-generated fiction, with researchers at Pangram finding it occurs three times more often in AI writing than human writing. Barron's reported that its appearance in corporate communications more than quadrupled between 2023 and 2025, earning it multiple names including "negative parallelism" and "contrastive phrasing."
OpenAI acknowledges the issue and is working on solutions, with product manager Laurentia Romaniuk confirming that ChatGPT relies too heavily on this construction, making outputs feel formulaic. The pattern affects all major chatbots including Claude, Gemini, and various open-source models, though Anthropic and Google declined to comment. Users have begun trading workarounds on social media, including custom instructions and feeding ChatGPT outputs through other AI tools with strict "negative pairing" bans, but no one—including the companies themselves—fully understands why AI models gravitate toward this particular construction in the first place.
- No AI company has identified the root cause of why their models develop this particular linguistic tic
- While other AI tells like obsession with goblins have been corrected, negative parallelism has proven stubbornly persistent despite being technically valid rhetoric
Editorial Opinion
This piece highlights a fascinating paradox in AI development: as models become more capable and convincing, their stylistic quirks become impossible to ignore. The "not X, but Y" construction is particularly insidious because it's a perfectly valid literary device—the problem is that AI's overreliance on it exposes the mechanistic nature of its training. Until companies understand why their models gravitate toward these patterns, they'll remain locked in a game of linguistic whack-a-mole, patching obvious tells while subtler ones multiply beneath the surface.
