Abstract

The proliferation of artificial intelligence (AI)-driven proofreading tools has fundamentally altered the landscape of written communication across academic, professional, and creative domains. This study investigates the dual challenge faced by such tools: delivering high-accuracy error detection whilst simultaneously preserving the original literary meaning, authorial voice, and stylistic intent embedded within texts. Employing a mixed-methods design, data were collected from 200 participants comprising undergraduate and postgraduate students, academic staff, and professional writers drawn from three Nigerian universities and two professional writing organisations. Quantitative data were gathered through structured questionnaires and systematic error-detection performance tests administered across six leading AI proofreading platforms—Grammarly, ProWritingAid, LanguageTool, Hemingway App, Quillbot, and GPT-4-based configurations. Qualitative data were obtained through semi-structured interviews and focus group discussions. Findings revealed that AI tools achieved a mean error detection rate of 84.6%, marginally below the human benchmark of 90.2%, whilst recording significantly higher false positive rates in stylistic and semantic categories. More critically, AI tools scored a mean literary preservation score of 6.1 out of 10, compared with 8.6 for human reviewers, indicating a substantive deficit in retaining figurative language, sentence rhythm, and authorial voice. These results support the proposition that AI proofreading is a powerful complement to, rather than a replacement for, expert human review, particularly in contexts where literary quality is paramount. The study advocates for user-centred AI design that integrates contextual literary awareness and proposes a pedagogical framework for responsible AI integration in writing instruction