Summarize, elaborate, try again: exploring the effect of feedback literacy on AI-enhanced essay writing

Synthesis: Hawkins, Taylor-Griffiths & Lodge (2026) investigate how university students use generative AI for feedback during essay writing, using a novel simulated assessment task methodology. Thirty-two psychology students at The University of Queensland completed a screen-recorded, 25-minute essay writing task with unrestricted access to ChatGPT (and optionally other tools such as Quillbot, Co-pilot, and DeepL), then watched the recording back in a video-stimulated interview. A thematic analysis (Braun & Clarke 2006) of the transcribed interviews identified four distinct, temporally ordered themes: feedforward (initial content and idea requests), feedback (requesting assessments of one's own work, including criteria-based and line-level language improvements), feedback evaluation (making decisions based on AI output, enacted through 'expand', 'Summarize', 'elaborate', and 'try again' prompts, and providing feedback to AI), and AI avoidance (deliberately not using AI, often citing academic integrity or a desire that the essay sound 'in my own words'). A standard multiple regression found feedback literacy to be the only significant positive predictor of essay grade (β = 0.46, p = 0.017), supporting the study's working theory. The paper interprets feedback literacy (Carless & Boud 2018) within the broader scope of self-regulated learning (Pintrich 2000) and is one of the first to deploy Dawson et al.'s (2024) validated Feedback Literacy Behavior Scale.

Key Findings

• Feedback literacy predicts essay task performance. Feedback literacy was the only significant (positive) predictor of essay grade in a standard multiple regression (β = 0.46, t(25) = 2.56, p = 0.017, sr² = .17); feedback literacy and essay grade were positively and significantly correlated (p = 0.019). The overall model accounted for a non-significant 35% of variance (F(6, 25) = 2.26, p = .071), and the analysis was underpowered.
• Four themes of student AI use for feedback. Feedforward (initial content requests, e.g. idea generation, exemplar essays, comprehension support); feedback (requesting assessments of own work, mostly 'line-level' language improvements rather than overall feedback); feedback evaluation (evaluating AI output via prompts and comparing it to other sources — fewer than a third compared AI output to another internet source); and AI avoidance (expressed by half the participants, motivated by academic integrity or wanting the essay to be their own words).
• Simulated assessment task as a methodological bridge. The novel screen-recorded task plus video-stimulated interview bridged the gap between self-reported AI use and observed practice, gathering evidence of students' metacognitive awareness of their AI interactions.
• Feedback literacy within self-regulated learning. AI co-regulated cognition (outsourcing idea generation and comprehension), motivation (corroborating task understanding and Self-Efficacy), behavior (varying Help-Seeking sophistication), and context. Without SRL skills grounded in self-efficacy and a motivation to learn, the authors argue AI operates "more like a student than a student tool."
• Task-level feedback dominance. Students typically requested task-level feedback, which Hattie & Timperley (2007) argue is difficult to transfer to other tasks and limited in effect on long-term learning.

What this means for practice

• Instructors. Teach Feedback Literacy explicitly rather than assuming it: it was the only significant positive predictor of essay grade (β = 0.46, t(25) = 2.56, p = 0.017, sr² = .17), so developmental work on Judgment-making is the lever the study identifies.
• Instructors. Push students past task-level feedback toward transferable feedback: participants mostly requested line-level language improvements, which Hattie & Timperley (2007) argue transfers poorly, so require students to name what they will carry to another task.
• Instructors. Make evaluating AI output a required step rather than an afterthought — fewer than a third of participants compared AI output against another internet source, yet feedback evaluation (deciding what to expand, Summarize, or retry) was one of the four observed themes.
• Instructors. Build tasks that keep learning goals in view: half the participants avoided AI at some point to keep the essay "in my own words," and the authors argue that without self-regulated learning skills grounded in Self-Efficacy and a motivation to learn, AI behaves "more like a student than a student tool."
• Researchers. Use the simulated assessment task plus video-stimulated interview to study AI use as observed practice rather than self-report; all 32 participants engaged with AI at least once and half also described active avoidance within the same task.

Limitations

• The regression was underpowered: with 32 participants the overall model explained 35% of variance and was not significant, F(6, 25) = 2.26, p = .071, as the authors state.
• Single site and single discipline: 32 psychology students from The University of Queensland.
• Task fidelity was compromised: the 25-minute essay task was simulated, performance did not affect grades, and AI use was not penalized — conditions the authors say likely shifted motivation away from learning goals.
• Feedback literacy was measured by a self-report behavior scale (Dawson et al., 2024), so the predictor is students' account of their own behavior rather than an observed measure.

Citation

Hawkins, B., Taylor-Griffiths, D., & Lodge, J. M. (2026). Summarise, elaborate, try again: exploring the effect of feedback literacy on AI-enhanced essay writing. Assessment & Evaluation in Higher Education, 51(5), 879–891.