Readers are pushing back against LLM-authored prose
An opinion essay cites survey findings that readers abandon writing they believe was produced by large language models and argues for human-authored public work.
Readers may be increasingly unwilling to overlook prose they believe was written by large language models, according to an opinion essay published on 5 September. The essay argues that detectable LLM involvement can damage trust in an author and cause readers to stop engaging with their work.
It cites a survey by Cynthia Dunlop of 668 responding developers, in which 78% reportedly said they stop reading immediately when they detect LLM involvement. A further 71% said they avoid the author in future, while 98% preferred an imperfectly written human piece to one polished by an LLM.
The survey methodology and respondent pool were not provided in the source, so the figures cannot independently establish how representative the findings are of readers generally. The essay presents them as evidence that authenticity matters more to some readers than linguistic perfection.
The author compares LLM-authored prose with email spam, arguing that reliable identification could reduce its effectiveness and increase the reputational cost of using it. The essay recommends keeping public-facing writing human-authored while using AI primarily as an editing tool.
Pangram Labs’ detection software is central to that argument. The author says Pangram 4, released about a month before publication, performed substantially better than earlier tools in personal testing, while acknowledging that such assessments are based on individual experience rather than independent evaluation.
Oxide has reportedly extended its public-writing policy to require material to be reported as human-authored by Pangram. Pangram Labs co-founder and chief executive Max Spero recently appeared on the Oxide and Friends programme, which the essay also recommends for discussion of the technology.
