Dartmouth Provost's Writings Flagged 96% AI by Detector

An investigation by The Dartmouth has highlighted a significant shift in the writing patterns of Provost Santiago Schnell. Using a text analysis tool known for its high accuracy, the review found that a median 96% of his recent publications display characteristics typical of artificial intelligence generation.

The findings contrast sharply with his pre-2022 work, which was consistently identified as human-written. Schnell maintains that he uses AI only for editing and clarity, asserting full responsibility for the ideas and final content of every piece he signs.

Detection results show sharp divergence

The analysis utilized Pangram, a detection tool that was audited by the University of Chicago in late 2025 and found to have a near-zero error rate. The system works by breaking text into smaller segments and classifying them as human, AI-assisted, or fully generated. This method allows for a granular view of how much of a document aligns with machine-generated patterns.

When applied to Schnell’s writings from before the widespread public release of ChatGPT in November 2022, the tool labeled all works as 100% human-written. However, nine pieces published in 2026 returned a median score of 96% for AI-written content. This suggests a substantial change in the stylistic and structural signatures of his writing over the past few years.

One specific example is a column published in The Washington Post, where Schnell argued that university degrees should distinguish between independent student work and tasks accomplished with AI assistance. That particular piece was identified by another outlet using the same detection method as 100% AI-written, a characterization Schnell disputes by claiming he only used the technology for minor refinements.

Schnell defends his use of tools

In response to the findings, Schnell issued a statement explaining his workflow. He stated that he develops all core ideas, conducts the research, and determines the structure of his arguments. He described AI as an assistive tool used for language refinement, copyediting, and improving clarity, emphasizing that he reviews the final text and takes responsibility for its content.

However, the credibility of this explanation was further complicated when the email statement itself was tested. The detection tool flagged the response to The Dartmouth as 100% AI-written. Schnell subsequently clarified that he drafted the substance and then used an AI tool to assist with language and clarity, accepting several suggestions before making minor changes and sending it.

Institutional policies face scrutiny

The incident raises questions about the enforcement of editorial standards in major publications. A spokesperson for The Washington Post noted that its policies require guest writers to confirm that their submissions are not created or manipulated with artificial intelligence or editing software. This creates a potential conflict between the publisher's stated rules and the practices of high-profile contributors who view AI as a standard editing aid.

The case highlights a growing tension in academia and journalism regarding the definition of authorship. While Schnell insists that his intellectual judgment and research remain central to his work, the high detection scores suggest that the resulting text bears a strong resemblance to machine-generated output. This distinction is increasingly difficult for readers and institutions to verify without specialized tools.