AI detection tools like Pangram may be reshaping human writing habits

The author describes AI detection platforms, notably Pangram, which promise to verify that text is entirely human-produced, a claim leveraged by educators and others to guard against machine-generated content. While University of Chicago scholars and a journalist have lauded Pangram’s performance, the company’s blog asserts a “1 in 10,000 false positive rate,” a figure the author’s informal experiments do not support. He recounts running his own articles through Pangram, occasionally receiving modest AI-assistance percentages, which spurred him to edit passages to appear less AI-like.

This self-censorship has led to noticeable shifts in his prose, including a drop in em-dash usage and a rise in antithetical framing, reflecting the subtle influence of large-language models. Concerns extend to editors who might distrust flagged work, and the author fears that both AI-generated text and detection tools could erode genuine human writing. He concludes that the dilemma is as much a personal, perhaps therapeutic, issue as a broader industry challenge.