Does homogenization at the text level cause homogenization of perceived authors?
This explores whether sameness in the text itself — AI flattening word choice and phrasing — translates into sameness in the *person* readers think they're hearing, i.e. whether homogenized prose produces homogenized authorial identities.
This explores whether sameness in the text itself produces sameness in the perceived author — and the corpus suggests the answer is yes, but through a more specific mechanism than 'bland writing makes bland people.' The link runs through *directional distortion*, not just flattening. A large study of 2,939 writers and 11,091 readers found that AI writing assistance shifted readers' perception of the author across all 29 measured dimensions — toward more extremism, more confidence, more agreeableness, more perceived privilege Does AI writing assistance change how readers perceive the writer?. The key word is *directional*: every author gets nudged the same way, so distinct writers converge on a common AI-inflected persona. That's homogenization of perceived authors, and it originates in the text.
What makes this more than a one-off finding is the channel that lets it spread unchecked. Writers edited the AI-generated paragraphs only 23% of the time, and when they did, their edits stayed 96% similar to the original Do writers actually edit AI-generated text before publishing?. So the model's opinionated voice reaches readers almost untouched — the text-level homogenization isn't filtered back out before it lands on the page as 'the author.' The perceived-author effect is essentially the text effect, propagated.
The corpus also shows the squeeze happening on the *input* side, before a word is even generated. Adam's Law describes how distinct prompts get flattened at comprehension time as users rephrase toward the higher-frequency forms the model handles best Does high-frequency text homogenize user input before generation?. So writers are funneled toward common phrasings going in, and emerge with a common persona coming out — pressure from both ends. Zoom out and this looks less like a quirk of one tool and more like a culture-industry pattern: AI mass-generates similar outputs disguised as personalization, and the contextual customization is exactly what makes the underlying sameness invisible to each individual user Does AI homogenize culture the way mass media did?. You feel like you're reading a distinct voice precisely because the homogenization is hidden behind surface personalization.
Here's the twist worth taking away, and it complicates a naive 'yes': homogenization and identity may live at different layers of the text. Work on AI fiction found that machine-written stories are distinguishable not by surface style — which is easily mimicked or humanized — but by discourse-level choices like character agency and chronological structure Can AI stories be detected without analyzing writing style?. If authorial signature actually resides in those deeper structural choices, then surface-level text homogenization wouldn't fully erase a distinct author — and conversely, a writer could polish the prose and still leak an AI-shaped 'self' at the structural level. So the honest synthesis is layered: AI homogenizes the *socially perceived* persona strongly and directionally because readers judge on surface cues that go unedited, even while the deeper structural fingerprint of authorship resists the same flattening.
Sources 5 notes
A study of 2,939 writers and 11,091 readers found AI assistance shifted every tested dimension—29 total—toward extremism, confidence, quality, agreeableness, and perceived privilege. Distortions were statistically significant and directional, not random noise.
Writers edited AI-generated paragraphs only 23% of the time, with edits averaging 96% similarity to the original. This means AI's opinionated and distorted voice propagates with minimal human filtering before publication.
Adam's Law shows LLMs flatten distinct prompts at comprehension time as users rephrase toward higher-frequency forms the model handles best. The same distributional property that creates accuracy on common tasks filters out distinctiveness on the input side.
AI mass-generates similar flows disguised as personalized outputs, suppressing novelty more deeply than pre-stamped commodities because contextual customization makes homogeneity invisible to individual users. Evidence: independent LLMs converge on similar outputs despite nominal competition.
StoryScope achieved 93.2% accuracy separating AI from human fiction using only discourse-level features like character agency and chronological structure, retaining 97% of performance while eliminating stylistic cues. These structural choices resist humanization because they require rewrites, not surface edits.