Co-Writing with AI, on Human Terms: Aligning Research with User Demands Across the Writing Process

We see encouraging alignment between HCI and CSCW research and writers' concerns across key contextual factors-time, trust, task importance, and confidence-that shape preferences for AI assistance….

RQ1: What design strategies are used or suggested in existing AI-assisted writing research and how are they distributed across writing processes and contexts?

Conducted a systematic review of 109 HCI papers (2018-2024), filtered from 1,676 records using the PRISMA approach.

Classified papers across four core writing processes (planning, translating, reviewing, and monitoring) and five writing contexts (academic, creative, formal, personal, and general).

Synthesized findings into four overarching design strategies (structured guidance, guided exploration, active co-writing, critical feedback) that span across processes and contexts.

RQ2: Which writing processes do writers consider essential to control to maintain their sense of agency and ownership during AI-assisted writing? Conducted semi-structured interviews with 15 writers from diverse domains and varying levels of prior AI experience. Characterized key themes describing what writers want to own when ownership matters to them and how AI interactions shape ownership during writing. Analyzed the extent to which design strategies in the literature align with or deviate from what writers demand across different processes and contexts.

…but a critical distinction emerges: writers' ownership preferences vary by writing contextform-centric writers prioritize control over style and revision, while content-centric writers focus more on planning and ideation.

Monitoring is a crucial yet underexplored cognitive process in AI-assisted writing-future research should support writers in managing their own work while overseeing AI-driven decisions through explicit controls for modulating AI involvement.