How AI Is Changing the Way the Internet Speaks
Less than four years after ChatGPT went mainstream, AI isn't just generating text – it may be changing the way humans write
Illustration: EVG Kowakievska via pexels.com
Language has always evolved with technology. The telegraph rewarded brevity, email changed the conventions of correspondence, and social media introduced an entirely new vocabulary. Now, artificial intelligence may be changing something deeper: the way the internet itself sounds.
A new Pew Research Center study of nearly 500,000 English-language webpages published between 2021 and 2026 found that more than one-third of content published after ChatGPT's late-2022 launch shows signs of AI-generated text.
But the more interesting shift isn't simply how much AI-written content is online. It's how that content – and potentially human writing alongside it – is beginning to sound.
Since 2023, the use of em dashes in post-ChatGPT content has nearly doubled, while Oxford commas have increased by 63%. Words associated with AI-generated prose, including delve, testament, interplay and tapestry, have more than doubled in frequency. Meanwhile, constructions such as "it's not just X, it's Y" have nearly tripled.
These patterns may be more than useful clues for detecting AI. They could signal a feedback loop: LLMs learn from human language, humans increasingly use LLMs to write and edit, and that machine-shaped language is then published back onto the web.
The result could be a gradual homogenization of online prose: more polished, predictable and interchangeable.
The internet isn't simply filling up with machine-written words. It may be developing a machine-shaped dialect, and humans are beginning to speak it.