When AI Speaks on Radio: Synthetic Personas, Professional Boundaries and the Legitimacy of Substitution

Since 2022, artificial intelligence (AI) has moved from the back office of radio and audio production to the microphone, usually justified by efficiency or novelty, and its deployments have met uneven contestation within the profession. A German commercial stream presented by a voice cloned from a real presenter attracted no documented dispute and was later licensed nationwide, whereas a Polish public-service channel whose AI presenters worked from human-verified scripts triggered an open letter, a petition and ministerial criticism and closed after one week. Both disclosed the artificial nature of their voices and production-stage taxonomies classify them alike. This paper explains the divergence through a structured focused comparison of a provenance-classified corpus of ten deployments in five countries (2023–2026), read against two organisational boundary-setting acts and Article 50 of the EU AI Act. Contestation does not follow automation depth or disclosure in this case. It follows substitution visibility, that is, whether a synthetic voice is publicly perceived as taking the place of an identifiable person. Persona claim, in turn, shapes where organisations draw the line of acceptable use. Furthermore, contestation is initiated and framed chiefly by media professionals, consistent with boundary work (Gieryn, 1983), rather than by audiences.