Semafor reported six Pangram AI-detection scores for opinion pieces opposing Waymo, ranging from 68% to 95%. Pangram defines its percentage as the share of text, measured by character, assigned an AI-related classification. It is not the probability that a named writer used AI.
Semafor associated scores of 95% and 71% with two Dallas Morning News opinion pieces by Amy Witherite; 88% with a San Francisco Chronicle piece by Michael Redmond; 68% with a New York Daily News piece; 95% with a San Diego Union-Tribune piece; and 87% with a Times of San Diego piece. Semafor also reported that the cited pieces did not disclose AI use.
The pieces address opposition to Waymo amid a broader debate over robotaxi rules. NeoTeo previously covered Waymo and Uber’s dispute over those rules.
A detector percentage describes a text classification; a draft history describes how a piece was prepared.
What a Pangram percentage measures
Pangram’s guidance on interpreting AI-detection results defines the percentage as the character-measured share of text assigned an AI-related label, rather than a confidence level or a probability about an author. Its documentation describes segment labels such as Human, AI-assisted, and AI-generated, with a separate confidence level for each segment. Those are distinct outputs: a label classifies a segment, while its confidence is reported separately. (Pangram API documentation)
That distinction matters when discussing the six figures. They are scores Semafor attributed to Pangram, not records of a writer’s intent or a complete account of how a particular piece was drafted. Pangram’s guidance treats a detector result as a reason for further review.
Writers and editors described different uses of AI
Ira Goldstein, author of a New York Daily News opinion piece and former head of the New York Black Car Fund, told Semafor that a public-relations firm prepared its first draft, which he reviewed and edited. He said he did not know AI had been used in that draft.
David Margulies, a spokesperson for Amy Witherite, said her communications operation sometimes uses AI to prepare initial drafts. He described staff review and fact-checking, followed by review with Witherite or other attorneys before her approval. Michael Redmond separately told Semafor that he uses AI for research, testing arguments, organizing ideas, and clarifying his writing.
Editors also described differing standards. San Francisco Chronicle editor Matt Fleischer said the paper asks opinion writers to disclose AI use. Dallas Morning News commentary editor Ryan Sanders said limited AI use for research or polishing does not automatically disqualify a submission, while the paper rejects pieces generated entirely by AI. New York Daily News executive editor Andrew Julien said the paper planned stricter AI safeguards.
