AI-written guest columns are appearing in major opinion sections. Human authors still dominate, but detection tools reveal a growing AI presence. Editorial policies and detection accuracy are now under scrutiny.
AI-generated guest columns are now appearing in the opinion sections of leading US newspapers. According to a recent Semafor analysis, while most guest submissions to The Wall Street Journal, The Washington Post, and The New York Times remain human-written, AI-generated content is increasingly present, raising questions about disclosure, quality, and editorial standards.
Over the past month, Semafor reviewed 310 guest opinion pieces published by these three outlets. Using the Pangram AI detector, the analysis found that 10 articles were labeled as at least 80% AI-generated, while another 40 were flagged as partially AI-written. Pangram, which claims a 0.5% false positive rate, marks content as AI if more than 80% of the text is machine-generated. Informal testing by Semafor suggested Pangram is highly accurate, though some users have reported occasional false positives.
The debate over AI authorship intensified after billionaire Stanley Druckenmiller and Wall Street Journal opinion editor Paul Gigot defended Druckenmiller’s use of AI to draft a recent op-ed. The piece, which criticized Treasury Secretary Scott Bessent’s bond market interventions, was confirmed by Pangram to be 100% AI-generated. Druckenmiller stated he now relies on AI for writing, likening it to using a calculator for math. Gigot responded that AI is now a fact of modern life, emphasizing that the key editorial concern is whether a submission reflects the author’s original argument and credibility.
The New York Times maintains a policy prohibiting AI use in developing and drafting guest essays. However, Pangram flagged an August 2 guest essay by Jen Easterly, former director of the Cybersecurity and Infrastructure Security Agency, as AI-generated. The Times did not comment on this finding, and Easterly did not respond to requests for comment. In total, Pangram identified 11 other Times guest articles in the past 30 days as partially AI-written.
At The Washington Post, an August 9 guest column by Dartmouth provost Santiago Schnell was also flagged as 100% AI-generated. Schnell explained that he wrote the essay himself but used ChatGPT to refine arguments and check grammar before submitting. He also noted that AI detection tools may be biased against non-native English speakers, a concern supported by Stanford research. The Post’s policy requires guest writers to confirm their work was not created or manipulated by AI, but the paper did not comment on the 17 guest articles Pangram found to be partially or fully AI-written.
Other publishers are also grappling with AI’s role in opinion content. The Financial Times recently added a note to a guest column after readers questioned its authorship, later confirming AI was used to shorten the article. The FT prohibits AI in the writing process. These developments echo findings from a recent study showing that nearly 10% of web content is now written or edited by AI, highlighting the challenge of measuring meaningful human contribution.
Semafor’s Reed Bergman noted that, despite the controversy, the majority of guest opinion content remains human-written. He suggested that quality concerns and fear of embarrassment may be limiting AI adoption among high-profile contributors. Bergman argued that large language models are not yet capable of producing high-quality, distinctive writing in a single attempt, and that AI labs are focusing on more marketable skills rather than improving writing style. He predicted that future AI models capable of learning and adapting to individual writing styles could eventually produce content indistinguishable from human authors, but for now, most contributors still rely on their own skills or those of their teams.
Research from the University of Michigan found that general audiences sometimes prefer AI-generated content, but MFA-trained experts favored human writing unless the AI was fine-tuned on high-quality human examples. The so-called “Claudish” style, associated with Anthropic’s chatbot, has become shorthand for low-quality AI writing, reinforcing the value of well-crafted human prose in opinion journalism.
