In composition classrooms, generative AI is still often framed in terms of misuse: will students cheat, become dependent, or make grading harder? A student sits down to write an essay and opens a generative AI tool. What happens next depends less on the tool itself than on the role the student gives it.
In one version, the student asks for a full draft, copies what appears, changes a few words, and submits the result as if the writing emerged from their own composing process. The AI’s labor disappears, and the student’s decision-making becomes difficult to trace. In another version, the student names the assignment’s purpose, audience, genre, and constraints before asking AI for several thesis directions. They compare, reject, revise, and disclose the tool’s participation. Here, AI is still in the loop, but it supports a visible and accountable writing process rather than quietly replacing student judgment.
I call the first version the ghostwriter model and the second version the co-author model. This distinction builds on Alan Knowles’s account of rhetorical load sharing, which describes AI-assisted writing along a spectrum rather than as a simple human/tool binary (Knowles, 2024). The difference between these models is not technical; it is rhetorical, not because AI shares human authorship, but because co-author-in-the-loop names the rhetorical labor in which the student remains responsible for judgment. The question is not simply whether AI was used, but whether the writer can account for the decisions that shaped the text.
The Ghostwriter Problem
The ghostwriter model is familiar because it seems to solve the most immediate problem students face: getting words on the page. AI can quickly produce polished, academic prose. For students uncertain about academic writing, that fluency can feel like relief: the blank page disappears, the awkward first sentence is solved, and the essay begins to look like an essay. But that relief comes with a cost. When AI functions as a ghostwriter, students often skip the parts of writing most central to learning: invention, audience awareness, revision, and judgment. They may still be “in the loop” physically, but their role shrinks to accepting, smoothing, or lightly personalizing language produced elsewhere.
That matters because writing is not only the production of text. Flower and Hayes argue that writing is “best understood as a set of distinctive thinking processes which writers orchestrate or organize during the act of composing” (Flower & Hayes, 1981, p. 366). The danger of the ghostwriter model is that it compresses planning, translating, reviewing, and monitoring into a finished-looking product before students practice the decisions that make writing transferable (Flower & Hayes, 1981). When AI supplies the draft, and the student cannot explain why a claim, sentence, structure, quote, or example exists, the work has drifted from collaboration to outsourcing. This is where rhetorical load sharing becomes useful. The problem is not that writing labor is distributed; writing has always moved across people, tools, genres, institutions, and feedback systems. The problem is when that distribution becomes invisible, making it difficult to tell where judgment happened and whether the student practiced the decisions the assignment was designed to teach. In other words, the ghostwriter model raises questions not only about academic integrity, but about learning.
Co-Authoring, But Not in the Usual Sense
Calling AI a “co-author” can sound uncomfortable, and for good reason. Composition studies has long challenged the assumption that writing is always solitary, asking instead how collaborative labor is organized, recognized, and valued (Ede & Lunsford, 1990). AI is not a co-author in the full human sense: it has no consent, lived experience, accountability, or responsibility for the text’s consequences. By co-author-in-the-loop, I mean something classroom-specific: AI can function as a visible participant in distributed rhetorical labor while the student remains responsible for purpose, judgment, revision, and attribution. In this model, AI does not replace the writer; it generates possibilities the writer must evaluate. It might offer thesis versions, counterarguments, transition feedback, or alternative structures. But those outputs do not become “the writing” until the student decides what fits their audience, purpose, evidence, and stance. The point is to make AI’s participation visible enough that students and teachers can ask better questions, not to elevate AI to the status of the human collaborators Ede and Lunsford describe
This distinction helps clarify the separation between different kinds of AI participation. When AI functions as a tutor or editor, it primarily gives feedback that supports the writer’s decision-making; as a co-author, it contributes generative material that the student must evaluate and integrate. However, when AI functions as a ghostwriter, it supplies the draft that makes the student’s role nonexistent and places them in a role of acceptance. The difference is not whether AI is involved in the process, but whether students remain able to account for their decisions and actions.
Co-author-in-the-loop writing is not about whether the student typed every word, but whether they remain accountable for the rhetorical choices that make the text work.
Making the Loop Visible
For teachers, that means asking students to document not only that AI was used, but how it was used. “I used ChatGPT” tells us little about the tool’s role, where it entered the process, what the student accepted or rejected, or how student judgment shaped the final draft. A more useful disclosure identifies the tool, role, location, and student authority over the final choice.
This shifts attribution from a policing mechanism to a learning mechanism. Johnson-Eilola and Selber’s work on assemblage helps here because it reframes writing as selection, arrangement, transformation, and attribution rather than pure originality; remix and assemblage practices can “aid invention, leverage intellectual and physical resources, and dramatize the social dimensions of composing” (Johnson-Eilola & Selber, 2007, p. 375). AI’s fluency can be seductive: a polished paragraph can feel correct simply because it sounds correct. Co-author-in-the-loop writing slows students down at the exact moment AI makes speed tempting.
Anchor, Field, Filter
Because of this, I created a concept titled: The Rhetorically-Governed Co-Authoring (RGC) Framework. I think of this structure as a three-part loop: Anchor, Field, Filter. This loop translates rhetorical load sharing into a classroom routine that keeps students in control of purpose-setting and evaluation while using AI for bounded option generation.
The Anchor comes before AI. Students name the rhetorical situation: audience, purpose, genre, constraints, and criteria for success. Before asking AI for anything, they decide what the writing needs to do. Without an Anchor, almost any polished AI output can look usable.
The Field is where AI enters in a bounded way. Instead of asking for one final answer, students ask for multiple options: thesis directions, counterarguments, structures, or introductions for different audiences. The goal is to create choices the student must compare, not accept the first fluent response.
The Filter is where students reassert judgment. They compare the AI-generated options against their Anchor, select what fits, revise what needs work, reject what does not serve the purpose, verify factual claims, and explain how they exercised authority over the final text.
Put simply: Anchor before AI, Field with AI, Filter after AI.
Why This Distinction Matters
The debate about AI and writing often gets stuck in a yes/no frame: Should students use it? Is it cheating? Is it helpful or harmful? Those questions matter, but a better question is: What relationship to writing is this AI use training? The ghostwriter model trains students to treat writing as something the tool produces for them. The co-author-in-the-loop model trains students to treat writing as a series of decisions they continue to make, even when tools participate. One hides rhetorical labor; the other makes it visible. One asks students to trust fluency; the other asks them to practice judgment. In Bandura’s terms, self-efficacy grows when students can connect outcomes to their own actions (1977).
This is the real pedagogical issue: AI does not determine whether students become stronger writers; pedagogy does. If AI operates as an invisible ghostwriter, students may learn that academic writing is mainly polished prose. But if AI use is designed around visible roles, criteria-based comparison, revision, and attribution, it can strengthen writerly agency rather than replace it. The goal is not to pretend AI is human or that writing has ever been fully solitary. The goal is to design writing loops where students can name, judge, revise, and account for the choices that shape their work.
That is what makes AI a co-author-in-the-loop rather than a ghostwriter. When AI is in the loop, writing teachers have to decide what else stays there: purpose, audience, judgment, accountability, and the student’s sense that they are still the writer.
References
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84(2), 191–215.
Ede, L., & Lunsford, A. (1990). Singular texts/plural authors: Perspectives on collaborative writing. Southern Illinois University Press.
Flower, L., & Hayes, J. R. (1981). A cognitive process theory of writing. College Composition and Communication, 32(4), 365–387.
Johnson-Eilola, J., & Selber, S. A. (2007). Plagiarism, originality, assemblage. Computers and Composition, 24(3), 375–403. https://doi.org/10.1016/j.compcom.2007.08.003
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
Generative AI Disclosure
I used the Grammarly extension in Google Chrome to review my Google Doc and help with APA formatting, including in-text citations and the reference page. As well as using it to catch grammatical errors, ensure punctuation is correct, and check coherence, making sure everything flows smoothly and makes sense.
Biographical Blurb
https://clasprofiles.wayne.edu/profile/ia2155
https://www.youtube.com/@TheRhetoricDiaries
