What the detector found
The scrutiny began with a Semafor investigation published last month into AI use in guest columns at leading US newspapers. Pangram, a widely cited AI detector, rated one of Schnell’s Washington Post columns “100% AI-written.” The column itself addressed the use of AI to cheat in college.
The Dartmouth then compared Schnell’s published work before and after ChatGPT’s public launch in November 2022. Pangram classified all the work from before that date as entirely human-written. Across nine selected publications after the launch, the median score was 96% in favor of AI.
Those are detector results, not proof of who wrote the text. Still, the contrast between the two periods is striking, and Schnell acknowledged using chatbots in his writing.
Schnell’s account
In a statement to The Dartmouth, Schnell said he remained the author and that AI’s effect on the final versions was limited to language editing and proofreading. He said he develops the ideas and arguments, conducts research, evaluates data and sources, decides on structure, drafts the work and determines what makes the final text.
After being asked to explain the detector results, Schnell added that he first wrote a draft and then “used an AI tool to improve the language and clarity” of the text. He said he took full responsibility for anything published under his name.
There is an awkward coda: Pangram also rated Schnell’s explanation as 100% AI-written.
The standard students see
The day after The Dartmouth published its investigation, its opinion editor, Eli Moyse, called for Schnell to be fired. Moyse, a student studying government and creative writing, argued that undisclosed AI use was insulting—particularly when Schnell had written about AI-assisted cheating.
Moyse said a Dartmouth student who used AI to write an entire assignment, failed to disclose it and submitted it to an instructor would clearly violate the Academic Honor Principle. He also argued that using AI so extensively for a Washington Post column would insult both the university community and the paper’s readers.
The deeper issue is not whether a detector can settle authorship. It is whether the university’s senior academic officer is being held to a standard students would face themselves. I think Schnell’s account draws a meaningful line between using a tool to polish prose and using it to produce the substance. But the detector scores cannot establish where that line fell in these articles, and the source gives no account of what disclosure, if any, accompanied their publication. That gap leaves the dispute about responsibility unresolved, even as Schnell says he accepts it.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X