AI provides quick feedback, the teacher retains the judgement
AI can provide feedback at lightning speed. But that does not automatically make it a good assessor. That, in a nutshell, is the lesson from a new systematic review on AI in formative and summative assessment, published in the journal Educational Technology & Society.
Researcher Ani Asryan brought together 26 scientific publications from 2020 to 2026. She looked at both feedback during learning and assessment at the end of a learning process.
For us at CivAI, this is familiar territory. It started in 2023 with one idea: the Feedback Companion. An AI companion that helps pupils and students to truly engage with formative feedback. EduGPT has grown out of that.
What AI does well
The studies are strikingly unanimous about the opportunities. Personalised feedback is mentioned in 17 of the 26 studies, as is the automation of repetitive tasks. Thirteen studies point to better insight into pupils’ progress.
Eight studies observe a shift from end-point assessment to formative evaluation during learning. Eight studies also report reduced workload for teachers. In one experiment, the marking time per assignment dropped by 43 percent.
And the pupil? Seven studies see increased self-regulation. Pupils take initiative with their work when feedback is prompt and matches clear criteria.
Where the friction lies
At the same time, the limitations are clear. AI can miss context, reproduce biases, and struggles to assess creativity or the quality of reasoning. A swiftly formulated judgement is therefore not automatically a reliable one.
The most frequently mentioned concern is privacy, in 12 of the 26 studies. This is followed by ethical questions (11), insufficient AI literacy among teachers (10), and the risk of biased algorithms (9). Seven studies warn against placing too much trust in what the machine says.
Deliberately divide the work
The review therefore advocates a hybrid approach. Let AI handle well-defined tasks, such as language checking, marking closed questions, and collecting information on performance. The teacher remains essential for interpretation and the final judgement, especially with complex assignments where creativity, argumentation and context matter.
For open assignments, the researcher suggests double marking as an option. Both human and AI assess, and the teacher weighs the results. The review also calls for clear agreements on data protection and for training that goes beyond mere button-pressing skills.
This is how the Feedback Companion works in EduGPT
This division of labour is precisely what we designed the Feedback Companion for. In practice, it works like this.
The teacher sets up the companion with their own assessment criteria or rubric for the assignment.
The pupil or student shares a draft version and receives feedback suggestions per criterion, with a concrete next step.
The companion does not write the work for the pupil. It asks questions in return, so the thinking remains with the pupil. This fits with the entire EduGPT approach: guiding rather than providing.
The pupil processes the feedback and reflects on what they did with it. The teacher monitors, discusses it, and gives the final judgement.
In this knowledge clip, you can see how the Feedback Companion works in practice.
This way, the pupil receives useful feedback more quickly, and the teacher has more time for what AI cannot do: the conversation, the nuance, and the judgement.
We also take privacy concerns seriously. EduGPT runs on European servers and the data remains under the control of the school or institution.
What we see in practice
Practice shows that this can work very well, for both pupils and teachers. But under one condition: AI feedback works best as a fixed part of a routine in the lesson.
Think of feedback from the companion, alternated with feedback from peers and a brief conversation with the teacher. Then, AI becomes an additional source of feedback alongside the people in the classroom, not a replacement for them.
Training in feedback with AI
Ten of the 26 studies mention insufficient AI literacy among teachers as a barrier. Especially in an era of AI, feedback literacy is just as important as AI literacy.
That is why we at CivAI also provide training. In these, teachers learn how to set up a feedback companion based on their own criteria, critically review AI feedback, and give it a permanent place in their lessons.
Curious what this could mean for your school or course? View our training and support or contact us. We are happy to think along with you.
Source
Asryan, A. (2027). The role of artificial intelligence in enhancing formative and summative educational assessment. A systematic review. Educational Technology & Society, 30(3), 211-227. doi.org/10.30191/ETS.202707_30(3).RP12
Image for illustration, created with AI