Algorithmically Mediated Audience Design: Voice, Pragmatic Force, and Linguistic Normativity in Generative-AI Rewriting

Generative AI is increasingly inserted between multilingual writers and their audiences, yet its linguistic consequences are often discussed as matters of accuracy, fluency, or efficiency. This study reconceptualizes AI-assisted rewriting as algorithmically mediated audience design: a process in which a generative system can redistribute pragmatic force, epistemic commitment, register, authorial visibility, and the indexical meanings through which writers position themselves. A controlled qualitative elicitation corpus was constructed from 12 researcher-designed English utterances representing recurrent applied-linguistic phenomena, including directness, stance, culturally situated formulaicity, pluricentric lexis, code-meshing, self-mention, and relational deference. Each item was contrasted across generic “improvement” and audience/identity-aware revision conditions. Analysis combined functional linguistic coding with deviant-case analysis and an explicit audit trail across lexicogrammatical, pragmatic, discourse, and sociolinguistic levels. Five recurrent processes were identified: sociolinguistic normalization, pragmatic mitigation, epistemic recalibration, redistribution of authorial voice, and selective retention of multilingual indexicality. Audience-explicit prompting constrained some homogenizing tendencies but did not eliminate normative mediation. The findings support a model of algorithmically mediated audience design in which prompting, model output, and human uptake jointly shape communicative meaning. The study argues that applied-linguistic evaluation of GenAI should move beyond error reduction toward consequential changes in voice, identity, normativity, and communicative agency.