A randomized controlled study by Sichterman and colleagues examined whether AI feedback could support students’ presentation performance without teacher interpretation. Sixty students in Dutch higher education were block-randomized to AI-only feedback (n = 30) or teacher-supported AI feedback (n = 30) after one five-minute rehearsal in an immersive virtual reality (IVR) classroom.
Both groups prepared the same five-minute talk and practised it before five virtual audience members. The system generated feedback on speaking pace, volume, intonation, and eye contact. The AI-only group reviewed the automated graphs and messages independently in IVR; the comparison group received the same AI data interpreted verbally by a trained teacher in a five-minute face-to-face session. Students then repeated the talk before a live audience. Two blinded experts rated six presentation skills, while questionnaires measured presentation perceptions, feedback utility, and anxiety.
No statistically significant between-group differences were detected in rated presentation skills (M = 7.06 vs. 7.50, p = .11), perceptions of the presentation activity (M = 3.57 vs. 3.40, p = .46), presentation anxiety (condition p = .67; condition-by-time p = .21), or perceived feedback utility (p = .12). Anxiety fell across time in both groups (p < .001), from 2.86 to 2.43 in the AI-only group and from 2.83 to 2.63 in the teacher-supported group. Under these conditions, teacher interpretation produced no statistically detectable short-term advantage.
The authors propose using structured AI feedback for formative practice of measurable non-verbal behaviours, leaving teachers to address content, structure, and argumentation. The study did not test statistical equivalence, so the absence of significant differences does not show that teacher support is unnecessary. It also lacked a no-feedback or practice-only group, measured most outcomes only after the intervention, involved one session and a small, concentrated sample, and did not blind participants. Larger longitudinal and cross-context studies are needed before drawing causal or general conclusions.
Source (Open Access): Sichterman, B., Banihashem, S. K., Verstappen, M., Schipper, M., Waisvisz, P., Noroozi, O., & van Ginkel, S. (2026). Can AI feedback stand alone for fostering students’ presentation performance? A comparison of AI-only versus teacher-supported AI feedback. Interactive Learning Environments, 34(3), 1280-1297.