Writer loses book job after AI detector flags her human-written sample

Ghostwriter Sally McGraw lost a book proposal job after a client’s AI detection tools flagged her writing sample as machine-generated. She had not used artificial intelligence to write the text, and she believes the error cost her the assignment.

The client confirmed in an email to NPR that McGraw’s work was screened. The client stated that experience and fit were the deciding factors in hiring another writer, though McGraw feels the detector results influenced the outcome. She was paid for the samples but did not pursue the full proposal.

Submission volumes at many publications have risen sharply. Neil Clarke, publisher of Clarkesworld, reported that monthly submissions to his science fiction magazine doubled from roughly 1,100 in 2021 to more than 2,000 by February 2023. Clarke temporarily closed the submission portal and estimates his own workload increased by about 30 percent. He noted that the influx has introduced mistrust into the editing process.

Detection tools often produce conflicting results. A study published in June in the International Journal for Educational Integrity tested four detectors on 160 academic papers. Pangram was the most consistent, correctly identifying AI use in 37 of 40 papers containing combined text and 37 of 40 where a human rewrote AI output. However, Chris Callison-Burch, a professor at the University of Pennsylvania, explained that detectors analyze how predictable a passage is to a language model. This can cause false positives when the text resembles the model’s training data.

Writers are changing their habits to avoid flags. McGraw now avoids specific words and sentence structures. She also omits commas she would normally use, hoping the inconsistency will prove human authorship. She said she misses using em dashes, a punctuation mark she learned in journalism school and has used for decades.

The Authors Guild warned in May that AI detectors create an environment resembling witch hunts. The group stated that false positives can damage an author’s reputation and publishing contracts, while false negatives can erode reader trust. Clarke advised that submissions should end in rejection, not accusation, and urged editors to verify detector findings before acting on them.

McGraw described the process as an attempt to fool a machine into thinking she is not a machine. She remains braced for a future assignment where the stakes are higher and the outcome could determine her career trajectory.