Writers lose jobs over AI flags

Ghostwriter Sally McGraw submitted a writing sample for a book proposal in late 2025. The client ran the text through several artificial intelligence detection tools. The software flagged the work as machine-generated. McGraw wrote every word herself. She felt her ears burn with frustration. She assured the client the text was human-made. She did not press the issue because she considered the job minor. The client hired a different writer. McGraw said the detector results likely cost her the full proposal contract, even though the client cited experience and fit as the primary reasons for the decision.

Editors are using these tools to manage a surge in submissions. Neil Clarke, publisher of Clarkesworld, sees the pressure firsthand. In 2021, his science fiction magazine received about 1,100 submissions per month. By February 2023, that number doubled. The increase followed the wide release of AI agents like ChatGPT, Claude, and Gemini. Clarke temporarily closed his submission portal. He estimates his own workload has risen by about 30%. He said the influx introduced an element of mistrust into the editing process.

The detection tools disagree with each other. A study published in June in the International Journal for Educational Integrity tested four widely used detectors. The researchers analyzed 160 academic papers in four categories: entirely human, entirely AI, combined, and human-rewritten AI. Pangram was the most accurate tool. It correctly identified AI use in 37 of 40 combined text papers. It also correctly identified AI in 37 of 40 human-rewritten papers. Max Spero, CEO of Pangram, said in an email that the detector can make mistakes, especially when there is little writing to evaluate.

False positives are a known risk. Chris Callison-Burch, a professor at the University of Pennsylvania, explained how these tools work. They ask a large language model how predictable a passage is. Models often find text from their own training data unusually predictable. This can trigger a false flag on human writing. Science blogger Janelle Shane tested excerpts from her 2019 book in late 2022. One detector reported 98.7% confidence that the passage was machine-generated. Shane called it a clear false positive. She had written the text herself.

Detectors look for specific patterns. They scan word choice, punctuation, and sentence structure. The word delve, frequent em dashes, and lists of three items can trigger a positive result. Opening sentences with adverbs like notably or crucially can also flag a text. Positing a question immediately followed by an answer is another trigger. Callison-Burch noted that these are common habits for human writers. Detectors weigh patterns across a whole piece, so a single em dash does not prove a bot wrote the text.

The Authors Guild warned in May that AI detectors create an environment reminiscent of witch hunts. A false positive can cost an author a contract and reputation. A false negative can cause readers to lose trust in the industry. Clarke does not personally use detectors to screen manuscripts. He believes certain tools can help overwhelmed editors if they verify the findings. He cautioned against turning suspicions into accusations. He said submissions should end in rejection, not an accusation.

McGraw has changed her writing style out of fear. She now avoids certain words and sentence constructions. She occasionally leaves out commas to appear more human. She misses her em dash, a style she learned in journalism school and used for decades. She sees the irony in making her prose look less like human writing. Large language models learn from human text. She worries about facing another accusation on a high-stakes assignment. She said she is bracing for the time when the outcome makes or breaks her career.