Argument-First Versus Language-First Generative AI Feedback in Chinese EFL Argumentative Writing: A Randomized Study

This eight-week randomized study examined how the order of generative artificial intelligence (GenAI) feedback shaped revision and near-term independent writing among 132 Chinese English-as-a-foreign-language (EFL) undergraduates. Students received argument-first (AF; n = 67) or language-first (LF; n = 65) feedback; 64 pairs underwent within-class, within-pair randomization, and four unmatched students were randomized within class. Both modules were independently generated from each initial draft (D0) and locked…

By the Truvace deskOctober 22, 20261 min read1 sourcePeer-reviewedTRV-2026-2158· cite & verify

In brief

An eight-week randomized study compared two orders of generative AI feedback for Chinese undergraduates learning English as a foreign language. One group got feedback on argument first, the other on language first, and students revised their drafts twice across three tasks.

The result matters for teachers deciding how to sequence AI feedback, since argument-first feedback helped argumentation both during revision and on a later AI-free task. It is less clear whether overall writing transfers, and the design cannot fully separate feedback content from the order in which it was released.

Main points

The study enrolled 132 Chinese EFL undergraduates: 67 in the argument-first group and 65 in the language-first group.
Argument-first feedback beat language-first feedback on overall quality and argumentation, but not on language quality.
On a new-topic, AI-free transfer task, argumentation favored the argument-first group.
Overall transfer was less consistent, as the complete-pair interval included zero.

The gain

Among 132 Chinese EFL undergraduates, receiving generative AI feedback on argument first produced larger gains in argumentation and overall essay quality than language-first feedback.

The rundown

Students were randomized within classes, with 64 matched pairs and four unmatched students randomized within class. Both feedback modules were generated independently from each initial draft and locked before release, so students received comparable material in different orders.

Students completed two revisions across three tasks, and the study then measured performance on a new-topic transfer task without AI support. Argumentation transfer favored the argument-first group by 1.057 points.

Language-quality contrasts did not survive primary Holm adjustment, although a complete-pair sensitivity analysis detected a crossover. Revision depth and implementation showed stage-specific crossovers. The authors support cautious use of staged feedback when argumentative development is the priority.