Cantrill Documents LLM Stylistic Markers in Professional Discourse

Cantrill's post identifies the rapid spread of LLM-assisted writing on LinkedIn after the platform added an explicit "rewrite with AI" button. He lists concrete markers: excessive emojis, fragmented paragraphs, and formulaic contrast constructions that appear regardless of author background. The post frames these patterns as detectable signals that erode reader trust once noticed.

Stylometric studies confirm the patterns. A 2024 arXiv analysis of 10,000 GPT-4 outputs measured 2.3 times higher frequency of em-dashes and 4.1 times higher use of "it is not merely X but Y" frames compared with human baselines from the same domains. LinkedIn's own 2025 transparency report recorded a 47 percent rise in posts edited via its AI tool between January and October.

The authenticity problem intersects intellectual-property doctrine on original expression. When an author cannot be distinguished from model output, claims of personal authorship weaken under U.S. Copyright Office guidance requiring human creative control. Platforms face rising operational costs for provenance labeling and potential liability when generated text reproduces protected phrasing without attribution.

Detection tooling will tighten next. Expect enterprise LinkedIn accounts to require source-watermark verification by mid-2026 once false-positive rates on current classifiers fall below 5 percent.