AI Content Detection

These tools are demonstrably unreliable: OpenAI withdrew its own classifier in 2023 because it was not accurate enough, and several studies found high false-positive rates, for example for authors who are not native speakers. Detection can also be undercut by simple rewording. Editorial teams should therefore never treat a detection score as the sole basis for a decision, such as penalising a writer, but at most as a rough signal alongside human quality checks. It makes more sense to judge every single text on factual accuracy, added value and editorial care, regardless of how it was produced. Detection is a separate question from disclosure: since August 2026 the transparency requirements of the EU AI Regulation have applied, requiring certain published AI-generated content to be disclosed – with exceptions, for instance where content has undergone editorial review and a responsible person stands behind it. Whether and how labelling is needed in a given case should be clarified individually.