Plural Voices: A Cultural Contestability Framework for Evaluating AI-Mediated Service Work

Workplace AI increasingly mediates professional voice in service work. Rewriters, templates, coaching assistants, and quality-assurance systems shape which ways of speaking are treated as professional before workers’ language reaches customers or supervisors. This paper argues that such deployments should be evaluated not only for productivity, bias, or toxicity, but also for whether they preserve voice plurality: multiple legitimate, intelligible, customer-serviceable ways of being professional. We introduce cultural contestability as a deployment-level criterion for this value. A culturally contestable system supports plural professional registers, makes rewriting decisions legible, lets workers challenge judgments that conflate professionalism with linguistic conformity, and preserves worker control over voice defaults. We motivate the framework with a descriptive March 2026 survey of 313 Philippine BPO/IT-BPM and shared-services workers. The survey is used as a directional design probe, not as a causal or representative study: it indicates heavy everyday AI use, limited formal training, concentrated accent-and-neutrality pressure in voice work, and weak awareness of challenge pathways. We then specify an interpretive evaluation protocol with four modules—paired draft–rewrite close reading, neutralisation-burden testing, benchmark-conversion analysis, and contestability audit—and four metrics: standardisation delta, worker voice loss, QA preference gap, and contestability gap. The contribution is a conceptual and methodological framework for evaluating workplace AI as a cultural technology, with cultural contestability as an auditable design target.