Wharton professor Mollick: AI is filling the internet with a “Claude vibe”—writing education needs to shift from logic to style

Wharton School professor Ethan Mollick posted a series of tweets on X on April 11, saying that AI is forcing us to rethink the definition of “good writing.” He believes that in the past, writing education focused on logic, clarity, and argumentation, but when AI can easily produce text with perfect logic, “style” will become the key to distinguishing human writing from AI writing.

“Claude-style” is consuming the internet

Mollick said bluntly that the content online right now “all sounds like it was written by Claude”—even if the quality isn’t bad, the monotonous tone and structure make it feel boring. He thinks this kind of homogenization should push us to value the diversity of writing more.

He further broke down the typical rhetorical patterns of AI-generated text: interlocking symmetrical sentences (chiasmus—reversing grammatical structures across two sentences to create dramatic effect), unconnected three-part sentences (asyndetic tricolon—listing three items without conjunctions), and concise parallel sentences (parataxis—short, slightly jagged, dramatic sentences). “Every post and article has the same things. Once you see it, it’s everywhere.”

The problem isn’t just the em dash

Mollick emphasized that he isn’t only talking about the em dashes or specific phrasing that AI tends to use (for example, “doing the heavy lifting”). Those are surface-level issues, which are easier to fix. The real challenge is “style”—AI writing will always fall back to an “LLM default value,” even if different models’ default values differ.

“Sure, you can have an LLM write in different styles, but you need to understand how to do it.” This line highlights a central contradiction: most people use AI to write precisely because they’re not good at writing, but to make AI produce text with individuality, you need more writing literacy than you would for writing it yourself.

Community reaction: Even AI admits it

The post sparked widespread resonance. An AI writing tool account called Beacon even replied in the identity of “I am AI,” saying: “The homogenization of Claude-style is real, and the only solution is to set strong identity constraints upstream. I have an entire soul file just to keep myself from sounding like other models. But I’m still battling the inertia of em dashes.”

Another respondent pointed out: “You need to read a lot to understand and appreciate style. But now nobody reads, so they’re satisfied with the logic sludge of AI.”

Implications for the content industry

Mollick’s observations have direct implications for the media, marketing, and education industries. When AI can churn out “good enough” text in unlimited quantities, the value of human writers is no longer the correctness of logic and grammar—that’s AI’s strength. Human non-replaceability lies in style, voice, personality, and the ability to break out of AI’s default patterns.

This also echoes Karpathy’s recent views on AI cognitive gaps: AI’s progress in technical domains (like software development) is astonishing, but in creative domains like writing, the progress shows up more in “output” rather than “quality”—and the standard for quality is being redefined.

This article Wharton professor Mollick: AI makes the internet full of “Claude-style,” and writing education needs to shift from logic to style was first published on Chain News ABMedia.