AI detectors used to police agency and creator content are unreliable and biased against non-native English writers, with false-positive rates above 60%. As synthetic content floods the feed, trust comes from provenance, not a confidence score.
The Take: Marketers are reaching for AI detectors to police agency and creator work. The tools do not survive contact with reality. They collapse the moment text is lightly edited, and they flag non-native English writers most of all. As synthetic content floods the feed, trust will come from provenance, not from a confidence score.
The evidence: A 2026 study from Sultan Qaboos University found leading AI detectors scored overall accuracy of just 69% and 61%, with performance on hybrid human and AI text dropping to almost zero. Providers advertise 99%. The gap between the marketing claim and the measured result is the whole story. That failure is not an edge case either, it is the normal condition, because anyone who wants to pass a detector simply edits the text until it does.
Who pays: The cost lands unevenly. Research has repeatedly found detectors misclassify writing by non-native English speakers as AI-generated at far higher rates, with one 2026 follow-up reporting a 61% false-positive rate on essays by Chinese students against 5% for US students. Any brand using these tools to vet international or multicultural content is quietly building bias into its process.
False-positive rate when leading AI detectors were run on writing by non-native English speakers, against 5% for native writers
The counter: The instinct is understandable. Australian marketers are publishing more AI-assisted content than ever, and about 64% of Australians have already seen AI-generated summaries sitting above their search results, so the pressure to prove human authorship is real. The problem is that a broken detector proves nothing. It just moves the risk from the content to the false accusation.
For Australian operators: Do not make a detector score a gate. Treat it as one weak signal at most, and never as grounds to reject a supplier or a piece of work on its own. Build authenticity into the process instead: keep draft histories, brief sources and record how creative decisions were made, because process evidence survives scrutiny that a detector cannot. If your brand safety policy leans on detection, rewrite it before it wrongly burns a good writer or a good agency. The audiences you are trying to earn trust from care whether the work is good and honest, not whether a black box scored it as human.