Co-Writing with AI, on Human Terms: Aligning Research with User Demands Across the Writing Process
Mohi Reza, Jeb Thomas-Mitchell, Peter Dushniku, Nathan Laundry, Joseph Jay Williams, Anastasia Kuzminykh · 2025
Read Co-Writing with AI, on Human Terms: Aligning Research with User Demands Across the Writing Process by Mohi Reza, Jeb Thomas-Mitchell, Peter Dushniku, Nathan Laundry, Joseph Jay Williams, Anastasia Kuzminykh (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-12488.txt.
Introduction
Reza and co-authors ask what writers actually want from AI writing help at each stage of their work. As chat tools enter daily writing, fears grow that they erode a writer's sense of ownership. Their 2025 study reviews 109 research papers on writing support and adds interviews with fifteen writers from varied fields.
Four styles of help emerge: structured guidance, guided exploration, active co-writing, and critical feedback. Writers wanted different amounts of help at different stages. Those focused on content guarded the planning stage most, while those focused on form guarded wording and review. Goals and values shifted the pattern further. The authors turn these findings into design advice for tools that support writers without taking over.
Catalog record
| CREATOR | Mohi Reza, Jeb Thomas-Mitchell, Peter Dushniku, Nathan Laundry, Joseph Jay Williams, Anastasia Kuzminykh |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 105 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-12488
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Co-Writing with AI, on Human Terms: Aligning Research with User Demands Across the Writing Process Mohi Reza, Jeb Thomas-Mitchell, Peter Dushniku, Nathan Laundry, Joseph Jay Williams, Anastasia Kuzminykh Abstract . As generative AI tools like ChatGPT become integral to everyday writing, critical questions arise about how to preserve writers’ sense of agency and ownership when using these tools. Yet, a systematic understanding of how AI assistance affects different aspects of the writing process—and how this shapes writers’ agency—remains underexplored. To address this gap, we conducted a systematic review of 109 HCI papers using the PRISMA approach. From this literature, we identify four overarching design strategies for AI writing support– structured guidance , guided exploration , active co-writing , and critical feedback –mapped across the four key cognitive processes in writing: planning , translating , reviewing , and monitoring . We complement this analysis with interviews of 15 writers across diverse domains. Our findings reveal that writers’ desired levels of AI intervention vary across the writing process: content-focused writers (e.g., academics) prioritize ownership during planning , while form-focused writers (e.g., creatives) value control over translating and reviewing . Writers’ preferences are also shaped by contextual goals, values, and notions of originality and authorship. By examining when ownership matters, what writers want to own, and how AI interactions shape agency, we surface both alignment and gaps between research and user needs. Our findings offer actionable design guidance for developing human-centered writing tools for co-writing with AI, on human terms. CCS: General and reference Surveys and overviews CCS: Human-centered computing Collaborative and social computing CCS: Human-centered computing Interaction design CCS: Computing methodologies Artificial intelligence Figure 1 . Research Overview: We present two interrelated qualitative studies exploring how to design for human agency in Human-AI Collaborative writing . (1) Study 1 : A systematic review and thematic analysis of 109 papers (2018–2024), filtered from over 1,600 records in the Human-AI collaborative writing literature using the PRISMA methodology; (2) Study 2 : A semi-structured interview study with 15 writers representing diverse writing genres, AI familiarity levels, and experience with generative AI tools. 1. Introduction As Large Language Models (LLMs) grow more powerful and pervasive, AI tools like ChatGPT have become integral to the modern writing process. Computers are evolving from mere tools to collaborative companions , raising concerns about the encroachment of writers’ co-creative boundaries ( Biermann et al., 2022 ) , eroding their agency and ownership over the writing process ( Lee et al., 2024 ; Lee et al., 2022 ) . At the same time, AI tools have proven to be remarkably helpful to writers, augmenting their capabilities across the composition gamut ranging from brainstorming and ideation ( Shaer et al., 2024 ; Si et al., 2024 ) , to editing and revision ( Reza et al., 2024 ; Laban et al., 2024 ) –and everything in-between ( Mollick and Mollick, 2024 ) –making it impractical to abandon this technology altogether. One way to address the tension between ensuring human control and increasing automation ( Shneiderman, 2022 ) is to examine how existing and proposed AI-assisted writing systems in the Human-AI collaborative writing literature support distinct cognitive processes in writing ( Flower and Hayes, 1981 ) , and whether that support encroaches on what writers consider to be central to preserving their sense of agency (i.e., their perception of control and autonomy over the writing process), ownership (i.e., their feeling of personal investment in and attribution of the final text), and task delegation (i.e., their choice about which writing subtasks to assign to AI, based on which cognitive processes they consider essential). Writers must maintain agency over the cognitive processes they value most through careful task delegation in order to preserve their sense of ownership over the final text. However, despite the unprecedented pace at which the AI-assisted writing research has grown over the past few years, there is currently no systematic understanding of what cognitive processes are being supported in numerous AI-assisted writing tools, what strategies are being used to offer that support, and how those strategies align with user perspectives across different forms of writing. Furthermore, these strategies have largely been evaluated from a usability and efficiency perspective ( Binns et al., 2018 ) , treating writing as a single task centered on optimization, with limited attention to preserving human agency through effective Human-AI collaboration ( Amershi et al., 2019 ) and across its distinct cognitive processes ( Flower and Hayes, 1981 ) . In this work, we explore whether recent research on Human-AI collaborative writing aligns with user needs, shifting the focus away from output-oriented concerns like productivity toward human-centered considerations–particularly how to preserve writers’ sense of agency and ownership when collaborating with AI. To investigate this, we ask two research questions: • RQ1: What design strategies are used or suggested in existing AI-assisted writing research, particularly in terms of interaction models and the intended use of AI outputs, and how are these strategies distributed across writing processes and writing contexts? • RQ2: Which cognitive processes do writers consider essential to control in order to maintain their sense of agency during AI-assisted writing, and how do user situations, writing contexts, and AI interaction types shape their perceptions of ownership? To answer RQ1, we conducted a systematic review and meta-analysis of 1,676 papers in the Human-AI collaborative writing space from 2018-2024. We first analyze the systems developed or proposed in recent Human-AI collaborative writing literature, then classify them according to distinct thinking processes involved during composition, as outlined in Flower and Hayes cognitive process theory of writing ( Flower and Hayes, 1981 ) . Then, to answer RQ2, we interviewed 15 writers across diverse domains, exploring how AI affects their sense of control and creative ownership across different writing processes. We then synthesized findings from both studies, revealing encouraging alignments as well as notable gaps between current AI writing support systems and writers’ expressed needs. Figure 1 provides an overview of our research approach. Our synthesis offers actionable guidance for designers, highlighting specific areas and methods of AI support that prioritize user agency. Rather than supporting all aspects of writing indiscriminately, our work helps focus design efforts on features that meaningfully preserve writers’ sense of control and ownership. Grounded in these findings, our contributions to the CSCW Human-Centered AI community include: • The first comprehensive study on designing for human agency in AI-assisted writing, combining a systematic review of post-generative AI research with a user-centered analysis of how writers seek to preserve ownership and originality. • A detailed characterization of four overarching design strategies for AI writing support, grounded in writers’ perspectives on when ownership matters, what they want to own, and how AI interactions shape that ownership. • Actionable design guidance for CSCW and HCI researchers developing AI writing tools, including concrete recommendations for supporting writer agency across the cognitive processes of writing. Our work informs future systems that foreground meaningful Human-AI collaboration, rather than automation alone. 2. Background & Related Work This section provides the theoretical and empirical foundation for our study. We begin by justifying our selection of the Flower and Hayes cognitive process model as our analytical lens. We then review the evolution of AI-assisted writing systems,