In my research methods course, students are allowed—encouraged—to use generative AI in their project work, while also retaining the option to refuse it. There is one condition: they must document how they use it. Every prompt, revision, discarded output, refusal, and moment of uncertainty goes into a semester-long reflective blog.
I designed the course this way because I wanted students to slow down and critically engage with AI-generated outputs rather than treating them as neutral or immediately usable. Instead of simply accepting outputs, students revise prompts, compare iterations, verify claims, rewrite sections in their own voices, and examine the rhetorical and ethical consequences of those decisions. In this sense, the course responds to what Mehdi Mohammadi and Derek Mueller describe as the “warpy” and “elastic” nature of contemporary writing loops—loops that stretch, adapt, and shift across changing tools, contexts, and participants.
And the loop does stretch.
It slows down. It doubles back. It resists optimization.
This essay proposes care-in-the-loop writing pedagogy as a way to understand and teach the rhetorical labor that emerges within AI-assisted composing. Rather than treating care as an abstract ethical ideal, I operationalize it through three interconnected practices: attentiveness, alignment, and accountability. I cultivate attentiveness through AI friction logs, where students identify moments when outputs feel generic, decontextualized, or rhetorically unsettling. I develop alignment through prompt chain mapping, where students trace how prompts evolve in response to audience, context, voice, and rhetorical purpose. I foreground accountability through AI positionality statements, where students reflect on when and why they use, revise, limit, or refuse AI tools. Together, these assignments make visible the interpretive and ethical labor that often disappears within narratives of speed and optimization.
Attentiveness and the AI Friction Log
The first dimension of care is attentiveness.
Commercial discourse surrounding generative AI often frames writing as a problem of speed, efficiency, and optimization: faster drafting, quicker summaries, streamlined workflows, and automated revision. Within these narratives, friction appears undesirable because it interrupts production. Writing studies and feminist scholarship, however, have long understood writing differently. Revision, recursive composing, listening, and situated reflection are not obstacles to productivity; they are central to meaning-making and ethical engagement.
I became interested in care-in-the-loop writing through this tension between optimization and attentiveness. In my classroom, attentiveness takes shape through an assignment I call the AI friction log. Students document moments when AI-generated outputs feel rhetorically unsettled—when a response sounds persuasive yet oddly generic, when cultural specificity disappears into abstraction, or when confident explanations seem detached from recognizable sources or contexts. Rather than moving quickly past these moments, I ask students to pause and examine what produced the discomfort.
As I read across students’ friction logs, recurring patterns begin to surface. Discussions of feminist activism in Pakistan are reframed through sanitized language associated with Western corporate diversity discourse. Prompts about South Asian political histories drift toward Euro-American examples unless students deliberately redirect them. Other entries note how outputs flatten disagreement, smooth over scholarly tensions, or reproduce authoritative claims without evidence.
For me, the friction logs make these moments visible in ways that conventional AI disclosure statements often do not. Instead of treating friction as failure, the assignment reframes it as rhetorical attention: a way of noticing how AI systems organize visibility, legitimacy, and representation. In this sense, attentiveness becomes an act of critical noticing—an attempt to identify what feminist scholars such as Sara Ahmed describe as the work of attending to what systems normalize, erase, or conceal.
My goal is not to teach students to reject AI entirely. I want them to remain attentive long enough to ask what kinds of knowledge are being reproduced, erased, or privileged within AI-generated outputs.
Alignment and Prompt Chain Mapping
If attentiveness teaches students to notice friction, alignment asks them to respond to that friction rhetorically.
Writing studies scholars have long understood writing as recursive rather than linear. AI-assisted composing intensifies this recursive dimension because prompts themselves become rhetorical acts. Revising a prompt does not simply modify an output; it reshapes the interaction, redirects the workflow, and changes what kinds of knowledge and representation emerge within the loop.
To make this process visible, I use prompt chain mapping. In this assignment, students trace how prompts evolve across iterations and explain why those revisions became necessary. The emphasis is not simply on producing “better” outputs, but on examining how writers actively align AI-generated responses with rhetorical purpose, audience, context, and voice.
As students work through these iterations, they begin making deliberate rhetorical interventions. Broad prompts about digital activism often produce polished but generic summaries, leading students to redirect the interaction through local contexts, feminist rhetorical frameworks, or regionally situated examples. In other cases, students notice that AI-generated literature review summaries smooth over disagreement between scholars, flattening intellectual tensions in ways that make the prose appear coherent but rhetorically thin. Rather than accepting those outputs passively, students revise prompts, reintroduce complexity, and reshape the direction of the interaction itself.
Prompt chain mapping also helps me think differently about Alan Knowles’ framework of Rhetorical Load Sharing, which describes how composing tasks move across human and nonhuman agents within accessorized workflows. Care-in-the-loop writing suggests that these workflows distribute more than tasks. They also redistribute attentiveness, judgment, responsibility, and rhetorical authority. Even within AI-assisted workflows, students continue performing interpretive labor that cannot be delegated. They decide what aligns with their goals, what sounds credible, what forms of representation deserve amplification, and what outputs require revision, limitation, or refusal.
For me, alignment is not merely a technical adjustment. It is a rhetorical negotiation.
Accountability and the AI Positionality Statement
The third dimension of care is accountability.
If attentiveness asks students to slow down and alignment asks them to reshape the interaction, accountability asks them to remain answerable for the consequences of their rhetorical choices. I operationalize this dimension through AI positionality statements. Unlike standard AI disclosure statements, positionality statements ask students not only to document what tools they used, but also to reflect on why they made those choices and what ethical concerns shaped their decisions.
In designing this assignment, I draw from feminist and cultural rhetorics scholarship on situated knowledge and relational accountability. Rather than treating writing as detached or universal, I want students to recognize themselves as rhetorically situated participants within broader technological, institutional, and cultural systems.
These positionality statements often become some of the most reflective writings students produce. Some students describe moments of refusal. One student chose not to use AI for a literacy narrative assignment because she felt the project required forms of emotional and cultural situatedness she did not want algorithmically mediated. Others describe strategically limiting AI use after recognizing how certain outputs flattened cultural specificity or reproduced generalized explanations that did not align with their rhetorical goals. Several reflections return to questions of credibility and authority, especially when AI-generated responses sound polished while remaining disconnected from verifiable sources.
What becomes visible in these reflections is not simply whether students use AI, but how they begin thinking about the relationships those uses produce: Who benefits from these representations? Whose perspectives become erased or normalized? Who becomes accountable when biased or decontextualized outputs circulate publicly?
As Maria Puig de la Bellacasa argues, care is not simply a feeling or moral orientation; it is an ongoing relational practice. For me, the AI positionality statement operationalizes this relational dimension by asking students to articulate how their rhetorical choices affect audiences, communities, and systems of knowledge production. It also complicates conventional ideas of authorship by encouraging students to think about writing not as an isolated individual production, but as a negotiated continuum shaped through interactions among writers, AI systems, prompts, institutions, and infrastructures.
Care-in-the-loop writing, then, reframes AI-assisted composing as more than an efficiency problem. Writing loops are not simply systems for producing text faster. They are spaces where writers negotiate representation, authority, and responsibility in relation to technologies that increasingly shape public discourse and knowledge production.
My goal is neither technological refusal for its own sake nor uncritical adoption. Instead, I want students to develop forms of critical AI literacy grounded in rhetorical attentiveness, recursive revision, situated knowledge, and relational accountability. If dominant AI discourse often asks how humans and machines can share rhetorical work more efficiently, care-in-the-loop writing asks a different question: what forms of attention, responsibility, and ethical negotiation remain necessary within those shared workflows?
For me, those moments—however small they may seem—are what ultimately shape the work.
References
Ahmed, S. (2017). Living a feminist life. Duke University Press.
CCCC Executive Committee. (2026). CCCC resolutions and position statements on generative AI, consent, and transparency. Conference on College Composition and Communication. https://cccc.ncte.org/cccc/2026-resolutions/
Knowles, A. M. (2024). Machine-in-the-loop writing: Optimizing the rhetorical load. Computers and Composition, 71, 102826. https://doi.org/10.1016/j.compcom.2024.102826
Mohammadi, M., & Mueller, D. (2026). [Blank]-in-the-loop writing: Rethinking processes, tools, and workflows [Blog carnival call]. Sweetland Digital Rhetoric Collaborative.
Puig de la Bellacasa, M. (2017). Matters of care: Speculative ethics in more than human worlds. University of Minnesota Press.
Royster, J. J., & Kirsch, G. (2012). Feminist rhetorical practices: New horizons for rhetoric, composition, and literacy studies. Southern Illinois University Press.
Sano-Franchini, J., McIntyre, M., & Fernandes, M. (2024, November). Refusing GenAI in writing studies: A quickstart guide. Refusing Generative AI in Writing Studies. https://refusinggenai.wordpress.com/
Generative AI Disclosure
Generative AI tools were used in limited ways during the development of this essay for brainstorming, editing, and proofreading support. All conceptual framing, pedagogical examples, analysis, and final writing decisions were developed and reviewed by the author.
