From Planning to Revision: How AI Writing Support at Different Stages Alters Ownership

Synthesis: Gero, Long, Schnitzler & Dhillon (2026, DIS '26) ran a between-subjects essay study (n = 253) showing that where AI support enters the writing process determines how much students feel they own the work: any AI assistance decreased ownership, but planning support cost the least while drafting support cost the most. The mechanism is AI-contributed text and ideas — and there is a genuine quality–ownership trade-off.

The experiment

• Short-essay writing, n = 253, between-subjects
• AI support offered at one of three stages: planning, drafting, or revising
• Measured: felt ownership, AI-contributed text/ideas, essay quality

Findings

• Any AI assistance decreased ownership
• Planning support: minimal decrease (outline-level help preserves authorship)
• Drafting support: largest decrease — and an AI draft built from the participant's own outline still contributed far more ideas than planning support
• More AI-contributed text/ideas → less ownership, but also better essay quality
• Recommendation: writers, educators, and designers should consider writing stage when introducing AI assistance

What this means for practice

• Instructors. Put AI support at the planning stage when authorship is the learning objective. Prompt-adjusted ownership ran 6.74 with no AI, 6.30 for planning support, 5.57 for revision support, and 4.29 for drafting support on the 1–7 scale.
• Instructors. Do not let the AI produce the draft if you want students to feel the essay is theirs. Drafting assistance lowered ownership by 1.65 Likert points relative to planning and revision combined, and participants attributed 56.9% of their final text to the AI, against 9.6% after planning support.
• Instructors. Decide which outcome the assignment is for before choosing where AI enters: drafting support produced the biggest quality gains and the biggest ownership loss, so one grade cannot reward both.
• Designers. Keep AI output editable and iterative. The authors argue that shifting contributions from finished deliverables to malleable drafts reduces the psychological distance that stops writers feeling like the author.
• Instructors. Have writers split credit between themselves and the tool for ideas and for wording; those attribution shares tracked the ownership gradient in this study and can be collected in an ordinary classroom assignment.

Limitations

• The study ran remotely on Prolific with 253 paid participants (mean age 38; 99% native English speakers), and 18% of submissions were rejected before analysis for using unapproved external AI tools; the authors state that controlled studies cannot replicate authentic in-the-wild engagement.
• Participants wrote a 200–300 word argumentative essay with no real stakes, in a genre AI handles well, and the authors note the grading rubric may be biased toward the stylistic fluency their human and LLM graders share.
• Writing was forced into a linear outline → draft → revise sequence that the authors call "somewhat artificial," with AI drafting allowed only once and no multi-turn conversation, unlike the looping real writing process.
• Ownership was measured post-task with a single 7-point Likert item ("I feel this piece of writing is truly mine"), and the sample consists of amateur writers, whom the authors expect to respond differently from professionals.

Citation

Gero, K. I., Long, T., Schnitzler, C., & Dhillon, P. (2026). From Planning to Revision: How AI Writing Support at Different Stages Alters Ownership.