Ghostwriter Loses Book Deal After AI Tools Flag Her Human Writing

Sally McGraw wrote a book proposal for a prospective client. She expected a review of her ideas. Instead, she received an email stating that multiple artificial intelligence detection tools flagged her text as machine-generated. The client hired a different writer for the full project. McGraw, a nonfiction writer with more than 20 years of experience, said the rejection stung because she had written every word herself.

"I was floored because I had not used AI to generate the text," McGraw said. "I'd written it all herself."

The client confirmed to NPR that she screened the work but said the detector results were not the sole reason for the decision. The client cited experience and fit as the primary factors. McGraw said the client paid for the writing samples, but she believes the detection results cost her the full proposal job. She described the experience as embarrassing and frustrating.

Editors are facing a surge in AI-generated submissions. Neil Clarke, publisher and editor of the science-fiction magazine Clarkesworld, said the flood of content has increased his staff's workload. In 2021, the magazine received about 1,100 submissions a month. By February 2023, that number had doubled. Clarke temporarily closed the submission portal to manage the volume. He estimates his own workload has increased by roughly 30%.

"It's introduced an element of mistrust into the process," Clarke said.

Detection tools struggle to distinguish between human and machine text. A study published in June in the International Journal for Educational Integrity compared four widely used detectors. Researchers tested 160 academic papers across four categories: entirely human-written, entirely AI-written, a combination, and AI material rewritten by a human. Pangram, one of the tested tools, was the most consistently accurate. It correctly identified AI use in 37 of 40 papers with combined text and 37 of 40 papers with human-rewritten AI text.

Max Spero, CEO of Pangram, said in an emailed statement that his company's detector can make mistakes, particularly when there is little writing to evaluate. Chris Callison-Burch, a professor of computer and information science at the University of Pennsylvania, noted that previously published writing can cause problems. Large language models find text from their training data unusually predictable. This can trigger false positives for human writers who use common structures.

Science blogger Janelle Shane encountered this issue in late 2022. She tested excerpts from her 2019 book, You Look Like a Thing and I Love You. One detector reported 98.7% confidence that a passage was machine-generated. Shane called it a "clear, glaring, blinking-light false positive." She said she had written the text herself. Callison-Burch said detectors look for clues in word choice, punctuation, and sentence structure. Patterns like frequent em dashes or lists of three can trigger a positive result, even in human writing.

The Authors Guild said in May that AI detectors create an environment reminiscent of witch hunts. The group warned that a false positive can cost an author their contract and reputation. Clarke said he does not personally use AI detectors to screen manuscripts. However, he thinks certain tools could help overwhelmed editors manage volume if they verify the findings. He cautioned against turning suspicions into accusations. "Submissions should end in rejection, not an accusation," he said.

McGraw now changes her writing habits to avoid future accusations. She avoids certain words and sentence constructions. She sometimes omits commas she would normally use. She hopes the inconsistency will prove a human wrote the text. "I miss my em dash. It was something that I learned in journalism school and I've used it for decades," she said. She worries about facing another accusation when a bigger opportunity is on the line. "I'm bracing for the time when it's an assignment that I'm really gunning for and the outcome makes or breaks me," she said.