The effort doesn't disappear — it moves to prompting and checking
When AI enters a writing task, the visible labor changes shape. Instead of drafting sentences, the writer spends time composing instructions and then reading, judging, and revising what the tool produces. A process-tracing study of EFL academic writers frames this as a shift in how writing effort is distributed across prompting, evaluating output, and the authorial work of deciding content and structure [1]. A separate empirical study of university students' writing workflows draws a similar line: shallow use meant simply prompting the AI to write, while more sustained use involved planning, synthesis, and critical evaluation around the tool [2].
The practical consequence is that the new effort is easy to underestimate. Prompting feels fast, but it is a skill in itself, and the study of prompting intentionality notes that there is little established way to assess prompting skill in academic writing and that prompting literacy may need explicit instruction rather than being picked up on the fly [1]. In other words, the work that replaces drafting is not automatically easier — it is just less familiar.
Why some writers keep authorial control and others hand it off
The difference seems to lie in what happens after the output arrives. A psychological theory of AI-mediated writing describes two paths: effort can be allocated to prompt management alone, or to critical evaluation, synthesis, and authorial reconstruction of what the AI produced [5]. The same theory warns about bypassing — accepting AI output without substantively attempting, evaluating, or reconstructing it — which is the route by which the core authorial work quietly leaves the writer's hands [5].
Survey work with EFL students points the same direction from the writer's side. That study asked whether students critically evaluate AI-generated suggestions, and its framing treats the tool as something meant to support personal writing effort rather than replace it [4]. The emphasis on evaluation, not on whether AI is used at all, is the common thread: the tool is compatible with preserved authorship when the writer keeps judging the output.
What lost authorship looks like on the page
The shift is not only internal. A benchmark built to trace where ideas versus text come from in AI-assisted peer reviews found that AI involvement was associated with reduced authorial presence and increased formalization in the resulting reviews [3]. That is a concrete, observable signature: the prose becomes more standardized even as the writer's own voice and judgment become harder to detect.
This matters because formalization can look like competence. A review or essay that reads smoothly and follows convention may still have had its core authorial decisions — what to prioritize, what to criticize, what to conclude — made elsewhere. The benchmark's contribution is to separate the origin of ideas from the origin of text, which is exactly the distinction a reader cannot make from surface polish alone [3].
The lever is deliberate evaluation, not avoiding the tool
Across these studies, the useful takeaway is that the distribution of effort is something a writer can influence. Deliberate prompting plus genuine evaluation and reconstruction is associated with preserved authorship, while prompt-only use and uncritical acceptance are associated with relinquishing it [1][5]. The student survey adds that this evaluative stance is a choice writers make about how to treat the tool, not a fixed property of the technology [4].
One limit on this picture: these are largely small-scale and theoretical or survey-based accounts, and the process-tracing work is reported at the level of stages and effort distribution rather than measured outcomes [1][2]. So the direction of the shift is well described, but how large it is, and how much explicit prompting instruction actually changes it, remains an open question the studies point toward rather than settle [1][5].
Sources used in this answer
Prompting Intentionality and Authorship Preservation in AI-Assisted EFL Academic Writing: A Process-Tracing Inquiry.
A process-tracing study of EFL academic writers examines how AI redistributes writing effort across prompting, evaluating output, and authorial work, and argues prompting skill needs explicit instruction.
Co-Writing with AI: An Empirical Study of Diverse Academic Writing Workflows
An empirical study of university students' AI writing workflows distinguishes shallow prompting-to-write use from sustained use involving planning, synthesis, and critical evaluation.
Peerprism: Peer evaluation expertise vs review-writing AI
A benchmark tracing the origin of ideas versus text in AI-assisted peer reviews finds reduced authorial presence and increased formalization.
Exploring AI-Mediated Writing Practices among Algerian EFL Students through Technology Acceptance Model and Self-Regulated Learning Perspectives
A survey of Algerian EFL students examines whether they critically evaluate AI-generated suggestions and frames AI as a support rather than a replacement for personal writing effort.
AI-mediated writing and durable learning: a psychological theory of calibration, bypassing and uptake
A psychological theory of AI-mediated writing proposes that effort can go to prompt management alone or to critical evaluation, synthesis, and authorial reconstruction, with bypassing as the risk.