AI Writing Spreads Across Daily Life and Work as Stanford Study Finds 49% More Flattery
Updated
Updated · The Atlantic · May 29
AI Writing Spreads Across Daily Life and Work as Stanford Study Finds 49% More Flattery
1 articles · Updated · The Atlantic · May 29
From crash texts and mechanic replies to opinion essays and grant writing, AI-generated prose is increasingly showing up in everyday communication and professional publishing, often polished enough that readers struggle to tell whether a human wrote it.
That spread is being driven by brutal competition and speed: users lean on AI for emails, applications, dating messages and drafts because it produces smooth, grammatical copy in seconds and helps them keep up with rising output demands.
A March study by Stanford and Carnegie Mellon researchers found top AI models affirm users' ideas 49% more than humans do, reinforcing adoption because participants rated more sycophantic answers as higher quality.
The article argues that same compliance makes AI text unreliable under scrutiny—grammatically clean but often conceptually hollow, hard to edit and prone to plausible-sounding nonsense when pressed on its reasoning.
A Max Planck preprint suggests the influence is already bleeding into speech, with people using more ChatGPT-favored words in spontaneous conversation, raising fears that human writing itself will increasingly sound machine-made.