The Effect of the Process–Genre Approach on AI-assisted and Multimodal Writing

Synthesis: Günay and Özden (2026) tested whether a process–genre approach to writing instruction — planning, drafting, revising and genre analysis, taught over five weeks and thirteen lesson hours — improves seventh graders' informative writing, Multimodal AI text creation and AI-assisted writing, using a nonequivalent pretest–posttest control group design with 16 students per group in a Turkish middle school. All three effects were large and the pre-test differences were not significant: informative text writing reached a post-test mean of 13.56 against 10.38 for the control group (r = 0.66), multimodal text generation 12.812 against 7.1250 (r = 0.84), and AI-assisted writing 9.8750 against 3.8125 (r = 0.86). Interviews with the experimental group describe the approach as functional, motivating and supportive of multimodal thinking, and the authors note that students built multimodal texts with pen and paper as well as digital tools. The study's contribution is to treat AI-assisted writing as a writing skill to be taught rather than a temptation to be policed.

Key Findings

1. Two intact classes, 32 students, five weeks. Seventh-grade students at one Turkish middle school formed the experimental and control groups (16 each) in a nonequivalent pretest–posttest quasi-experiment spanning 13 lesson hours.
2. Pre-test equivalence held on all three rubrics. Baseline differences were statistically non-significant (informative text p = .359, multimodal p = .443, AI-assisted writing p = .396), which supports attributing post-test differences to the intervention.
3. Informative text writing improved substantially. Post-test means were 13.5625 for the experimental group against 10.3750 for the control group (MW-U = 30.00, Z = −3.73, p = 0.00, r = 0.66).
4. Multimodal text creation moved furthest among the writing outcomes. Post-test means were 12.812 against 7.1250 (MW-U = 2.50, Z = −4.77, p = 0.000, r = 0.84).
5. AI-assisted writing showed the largest effect in the study. Post-test means were 9.8750 against 3.8125 with no overlap between groups (MW-U = 0.00, Z = −4.885, p = 0.000, r = 0.86).
6. Three purpose-built rubrics were reliable. Cronbach's alpha was .780 for informative text, .934 for multimodal text generation and .930 for AI-assisted writing.
7. Students described the method as usable and motivating. Interviewees reported a positive attitude toward writing, better use of technological tools and AI, and gains in multimodal thinking, which the authors read as support for the product scores.

What was taught, and how the three outcomes were scored

The intervention is a process–genre sequence: students analyze model texts in a target genre, plan, draft, revise and publish, and the experimental group did this with AI support and a multimodal component, while the control group continued its usual writing instruction. Three researcher-developed rubrics scored the products, and the paper reports their internal consistency before reporting any comparison, which is the right order. Assessment was product-based: the pre-test and final test each took two lesson hours, and the difference between the two products was the outcome. The qualitative strand is deliberately secondary — a sequential embedded design in which interviews with experimental-group students were collected after the post-test to explain the quantitative result rather than to stand beside it.

Why the multimodal dimension is the interesting one

The authors present multimodal generation as the study's original contribution, and their argument is worth separating from the effect size. Multimodal literacy work in the Turkish literature has largely examined teacher competencies and textbook analyses, leaving open whether students can produce multimodal texts at all. This study found that students did so within a writing-plan framework using both digital tools and pen and paper, which undercuts the assumption that Multimodal AI composition depends on equipment. That matters for equity: a school without a device for every student can still teach multimodal composition if the plan, the genre analysis and the Scaffolding come first. The AI-assisted rubric is the more fragile of the two new measures, because it scores a practice that did not exist when earlier instruments were built.

Reading the effect sizes carefully

Three large effects from 16 students per group deserve the caution the design invites. The groups were intact classes rather than randomly assigned individuals, and matching on pre-test does not match on teacher, schedule or prior AI experience; the paper notes that the groups were not equalized and that initial differences were controlled by pre-testing only. All three rubrics were written for this study, so the outcome and the intervention were designed together, and no assessor blinding is reported. The AI-assisted writing result, where the experimental median rank sat far above every control student, is also the outcome most exposed to a scoring confound: a rubric that rewards competent prompting and iterative revision will favor students who were taught prompting and iteration, which is a real skill but not the same claim as improved writing. The five-week window with no delayed post-test leaves open whether the gains persist when the genre and the task change.

What this means for practice

• Instructors. Teach AI-assisted writing as a genre with its own conventions — prompt, evaluate, revise, attribute — rather than banning the tool or leaving students to discover it, because the group taught that sequence outperformed the group that did not receive it.
• Instructors. Put the writing plan before the tool. Students in this study produced multimodal texts with pen and paper as well as with digital tools once the plan and genre model were in place, so device access is not the binding constraint.
• Curriculum designers. Build multimodal composition into the writing curriculum rather than into a separate technology unit; the largest effects here came from combining genre instruction with multimodal production, not from adding a tool to an existing assignment.
• Assessment designers. Score AI-assisted writing with a rubric that separates prompting and revision from the quality of the final text, or the two get credited as one skill.
• Administrators. Fund the thirteen lesson hours, not the software license: the intervention that produced these effects was instruction, and the AI was one component of it.

Limitations

• The sample is 32 seventh graders in two intact classes at one Turkish middle school, with no random assignment, so selection and teacher effects cannot be separated from the intervention.
• All three outcomes were scored with rubrics developed by the researchers for this study and reported without assessor blinding, which couples the measure to the treatment.
• The intervention ran five weeks and 13 lesson hours with no delayed post-test, so durability and transfer to other genres are untested.
• Interviews were conducted only with the experimental group, so the qualitative themes describe satisfied participants and cannot compare experience across conditions.
• The control group continued its usual writing instruction, which is a weak comparator for a study claiming a specific method rather than time on task.

Citation

Günay, G., & Özden, M. (2026). The Effect of the Process–Genre Approach on AI-assisted and Multimodal Writing. Journal of Mother Tongue Education, 14(4), 801–822.