Currently reading Nazaretsky et al. (2026) on “Who Gives Feedback Matters: Student Biases Towards Human and AI-Generated Formative Feedback”

Most teachers I talk with recognize that asking GenAI to grade student exams is a terrible idea, but when it comes to formative feedback, there are a lot of teachers that think that that is maybe not such a terrible idea. Students can get feedback much more promptly than if a teacher has to read student work, think, and then write the feedback. And it is easily scalable even for very large classes. Yeah, it might not be 100% correct all the time, and we are also missing out on all the relational aspects. But maybe…?

In their study, Nazaretsky et al. (2026) investigate whether it matters to students if feedback is given by a teacher or GenAI (spoiler alert: it does!). They use 472 students from different disciplines and academic levels and gave them two feedbacks on one of their own, authentic writings, one teacher- and one GenAI generated, and asked them to evaluate them. They then did that again, but this time disclosing which one was which. They find that there is a bias against GenAI as feedback provider: in the blind condition, most students attribute the better feedback to the teacher rather than GenAI. Once it is revealed which feedback is which, the teacher-generated one is evaluated more positively and the GenAI-generated one more negatively.

What I find especially interesting is what makes a feedback good in the students’ perception. The “positive pedagogical features (e.g., being ‘precise’, ‘personalised’, ‘accurate’, and ‘useful’)” are typically associated with feedback from teachers rather than GenAI. When students were informed that a feedback was generated by GenAI, they significantly reduced as how genuine (measured as authentic and sincere) they perceived the feedback.

Another really interesting finding is that only very few students mentioned meta-cognitive aspects of feedback, for example whether there was advice on learning strategies (even though that was there in many of the feedbacks, but students did not seem to recognize or appreciate it), whereas many more, 1 in 5, commented on task-specific feedback like identifying and correcting mistakes. This shows a lack of students’ feedback literacy.

Nazaretsky et al. (2026) then discuss that despite students evaluating GenAI feedback better when they don’t know that it isn’t coming from their teacher, teachers need to be transparent about their use of GenAI. But, Nazaretsky et al. (2026) suggest, teachers could consider a human-in-the-loop approach, where teachers quality-control and edit GenAI generated feedback to make it more their own but still benefit from the saved time, at least if they are sufficient feedback- and GenAI-literate themselves. Nazaretsky et al. (2026) conclude (in their abstract) that “[t]hese insights highlight the need for targeted interventions, such as improving AI literacy and building human-­in-­the-loop systems, to mitigate biases and enhance the effectiveness of AI in educational feedback systems.” Which sounds sensible enough, at least if we agree with the premise that we should use GenAI in producing feedback at all…

Nazaretsky, T., P. Mejia-Domenzain, V. Swamy, J. Frej, and T. Käser. 2026. “ Who Gives Feedback Matters: Student Biases Towards Human and AI-Generated Formative Feedback.” Journal of Computer Assisted Learning 42, no. 1: e70153. https://doi.org/10.1111/jcal.70153.

Featured image: Real or not, who can tell…

And some more pictures from my ferry trip back to Sweden! Leaving Travemünde…

Then I was working on my laptop and when I looked up, I saw this wake right next to us! Of course I had to go investigate!

Only few minutes later it looked like this: Crossing wakes!

Very nice and calm trip, I was the only foot passenger!

And Öresund bridge in the mist!

That looked pretty cool, too!