LLM Asks, You Write: Enhancing Human-AI Collaborative Writing Experience through Flipped Interaction

Abstract

Large Language Models (LLMs) have offered unprecedented writing assistance, significantly improving the quality of writing outcomes. However, this assistance often relegates users to passive reviewers rather than active creators, potentially compromising their creative engagement and subjective experience in the writing process. To enhance users' writing engagement and agency while preserving the benefits of AI assistance, we propose a novel flipped interaction framework called Guided-Writing for human-LLM collaborative writing. Unlike the traditional Prompt-Generate mode, where users prompt LLMs to generate content, the Guided-Writing mode features controlled questioning from the LLM, guiding users to stay focused on their writing while leveraging the LLM's strengths in creative inspiration and text editing. Through a within-subjects experiment comparing both modes, our findings demonstrate that the Guided-Writing mode significantly enhances users' independent writing engagement and strengthens their sense of agency, ownership, self-achievement, and self-expression, while maintaining comparable mental workload. Moreover, users in the Guided-Writing mode exhibited greater willingness to take responsibility for their writing outcomes, with two-day post-experiment assessments indicating higher perceptions of content authenticity and reproducibility. This study demonstrates the practical benefits of the flipped interaction framework in enhancing users' writing experience and offers valuable insights for the future design of user-centric LLM-assisted writing tools.

Keywords: Large Language Models (LLMs), Human-LLM Collaborative Writing, Writing Experience, User-centered design, Flipped Interaction

DOI: 10.54941/ahfe1006247