This presentation reports on empirical research examining how AI-assisted text revision affected first-year Japanese university students' confidence and language skills. In a six-week project, students progressed through a scaffolded cycle: writing responses to prompts, recording spoken versions of their written texts, revising their texts using AI tools, and finally recording the revised versions.

Data analysis from 232 participants revealed that 81% reported increased confidence in their English abilities after using AI for revision. Students identified grammar (153 mentions), vocabulary (107 mentions), and sentence structure (106 mentions) as the primary areas of improvement. While self-reported data was the primary assessment method, the research design allowed students to directly compare their original and AI-revised texts, creating heightened awareness of language patterns.

While ChatGPT was the dominant tool (67.7%), students independently selected diverse AI resources including DeepL (5.8%) and Microsoft Copilot (4.9%). The study demonstrates how structured AI integration creates a positive feedback loop between writing and speaking skills. The write → speak → AI-revise → speak revised text sequence provided students with immediate corrective feedback while building confidence through guided improvement.

This research contributes practical insights for instructors seeking to implement AI tools as confidence-building resources rather than mere shortcuts.