Work

Generative feedback: causal effects of LLM’s on writing quality and evaluative equity in higher education

Year: 2026
Type: book-chapter
Abstract: This chapter examines, within the framework of SDG 4, how large-scale generative feedback using large language models (LLMs) impacts the quality of scholarly writing and evaluative equity in higher education.The aim is to estimate improvements in coherence, argumentation, use of evidence, and clarity, and to determine whether gaps between subgroups (L1/L2, first generation, and baseline performance) are reduced.An experimental or quasi-experimental design (stepped-wedge cluster trial) with blinded assessment based on analytical rubrics and secondary metrics such as time to feedback, revision iterations, editing distance, self-efficacy, and teaching load is recommended.A synthesis of recent evidence suggests consistent gains when LLM feedback is effectively integrated into explicit rubrics and revision microtasks; furthermore, it increases engagement and can decrease inter-rater variability.However, risks of dependency, stylistic homogenization, and algorithmic bias are noted.Therefore, this chapter proposes a framework for didactic governance with traceability (registration of prompts, versions, and logs), open science principles (pre-registration, repositories), and a dual human-AI evaluation scheme with bias audits and human-in-the-loop thresholds.Among the limitations acknowledged are the heterogeneity of contexts, sensitivity to model versions, and external validity.The conclusion is that, under ethical safeguards and with teacher training, generative feedback can raise writing standards and broaden inclusion; further research is suggested on differential effects, disciplinary transferability, and institutional sustainability of scaling up.
Language: English
Subfield: Education
Open Access status: gold
OpenAlex ID: W7127048369
FWCI:
Cites: 17
Cited by:
Related to: 9