AI writing detectors can wrongly flag human-written work, and those results can influence who gets hired or published. An NPR report describes a professional writer whose sample was flagged, alongside evidence that different detectors can disagree about the same text.
Watch Desk analysis
What happened
Ghostwriter Sally McGraw told NPR that multiple AI-detection tools flagged a writing sample she says she wrote herself. The prospective client confirmed screening the sample, but told NPR the results were not the deciding factor in hiring someone else. McGraw believes the flags cost her the larger assignment.
A study cited by NPR compared four detectors on 160 academic papers. Pangram performed best in two tested categories, correctly identifying AI use in 37 of 40 papers with mixed human-and-AI text, and 37 of 40 where a person had rewritten AI-generated material. That is useful evidence of performance on those test sets, not a guarantee that a detector can reliably judge any individual writer.
Why it matters
A detector result can look satisfyingly precise while still being a shaky basis for an accusation. Researchers cited by NPR say tools can disagree, and that writing patterns associated with AI can also be ordinary habits of human writers. A flag may invite scrutiny; it does not establish who wrote the text.
The consequences are not just theoretical. McGraw says the experience has led her to change her writing style, avoiding words and constructions she fears could trigger another flag. That is a grim little inversion: writers adapting their prose to satisfy software that learned from human prose.
Our read
Treat detection scores as a prompt to ask questions, not as proof. Editors and employers should seek context and give writers a fair chance to respond before turning a machine’s suspicion into a verdict. The tools may help manage a flood of submissions, but the human review cannot be optional when someone’s work and reputation are at stake.
What to watch
- Whether publishers and employers set clear rules for reviewing detector results.
- Whether further studies test tools across more kinds of writing and real-world conditions.
- Whether writers are increasingly pushed to change their style to avoid false flags.
Discussion spark: Should editors and employers be allowed to use AI-detector scores in hiring or submission decisions, or should a score never count without independent evidence?
Sources and evidence
- Source update (31 October 2026, 09:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.