January 15, 2025
blog
How can AI provide formative feedback on students’ writing assignments?
Ask any student what really helps them learn, and chances are you’ll hear 'feedback.' It’s no coincidence that John Hattie (2009), in his large-scale research, concluded that feedback is one of the most powerful learning interventions.
But what do we actually mean by feedback? In their well-known paper, Hattie & Timperley (2007) define it as: 'information provided to a student about aspects of their performance or understanding.'
That information can take different forms. Sometimes it’s a grade on an assignment, also known as summative feedback. Useful for administration, but with one big drawback: the student only hears what could have been improved after submitting, with no chance to apply those improvements. The learning process essentially ends too soon.
That’s why more and more teachers are choosing formative feedback: feedback given before the final evaluation. This allows students to actively work on improvements, refine their assignments, and at the same time deepen their understanding. For learning, this is invaluable.
Providing formative feedback takes a lot of time. And not just a little, it often means hours spent reviewing long reports, formulating concrete suggestions, and repeatedly explaining learning objectives. For teachers or teaching assistants, this can be downright exhausting.
And then the question naturally arises:
Shouldn’t modern technology, such as AI, play a bigger role here?
We believe it should. In this blog, we’ll explore that very question: how can AI provide formative feedback on students’ writing assignments?
What makes good formative feedback?
Okay, so we now know: formative feedback is worth its weight in gold for learning. But… there’s a catch. Feedback can also completely miss its purpose. Or, as Valerie Shute (2008) aptly put it:
Valerie Shute (2008)
“Feedback can be extremely powerful if done well and extremely ineffective, or even harmful if done poorly.”
In other words: not all feedback is automatically good.
Simple or elaborate?
A good example comes from the study by Elder & Brooks (2008) in a nursing program. Students there received two types of feedback:
simple (“correct” or “incorrect”),
elaborate, with an explanation of what a better answer would be and why.
The outcome? Students who received elaborate feedback performed better on follow-up assignments and reported having a deeper understanding of the material. Makes sense: explanation + context = more learning.
The three big questions of Hattie & Timperley
But elaborate feedback alone is not enough. Hattie & Timperley (2007) formulated three key questions that good feedback should answer:
Feed up – Where am I going?
Feedback – How am I doing now?
Feed forward – What is the next step?
And to make it even more concrete, they also distinguish four levels of feedback:
Task level – Is the answer correct?
“This answer is correct because you used the right formula.”Process level – How well does the student apply strategies?
“Good approach, but next time also think about X.”
Self-regulation level – How can the student improve themselves?
“You check your work thoroughly—that helps prevent mistakes.”
Person level – Feedback on the student as a person.
“Well done, you must have studied hard.” (not effective for learning).
The seven rules of Nicol & Macfarlane-Dick
As if that weren’t enough, Nicol & Macfarlane-Dick (2006) formulated seven “rules” for good feedback. Think of it as a checklist or mini-rubric:
Clarifies learning goals and performance standards.
Encourages self-evaluation and reflection.
Provides informative, usable feedback.
Promotes dialogue between student and teacher.
Strengthens motivation and self-confidence.
Offers concrete guidance for improvement.
Gives teachers insight into learning problems.
Their main message: feedback is not a one-time event, but a continuous process. Students need multiple rounds of feedback to truly steer and improve their learning (see also Shute, 2008; Hattie & Timperley, 2007; Juwah et al., 2004).
But…
Sounds great, right? The only issue: all this takes an enormous amount of time. How often can teachers realistically provide such elaborate, repeated, and well-thought-out feedback to every student?
How AI can support the feedback process 🤖✍️
When I (Jeroen van Gessel) started Eduface in 2023, I asked myself a simple question: can AI help provide formative feedback? Because let’s be honest: even the most dedicated teacher finds it incredibly challenging to always phrase feedback in a way that perfectly aligns with the guidelines we just discussed.
And that’s not even mentioning the time it takes to give all that feedback.
How we approach this
With Eduface, we are building an AI feedback model that supports teachers in reviewing writing assignments. The model is didactically grounded, trained on key research sources (such as Hattie & Timperley, 2007 and Nicol & Macfarlane-Dick, 2006), and delivers feedback that is immediately actionable.
One important choice we had to make: how do we present that feedback?
Many teachers provide feedback at the end of an assignment or in a separate document. But research by O’Donovan, Rust & Price (2015) shows this is often ineffective: students understand such feedback less well, don’t see exactly where the issue lies, and rarely apply it.
That’s why we place feedback directly in the text itself, right where improvement is needed. Short, concrete, with explanation, and always with a feedforward: what can the student do now to move forward? This way, the feedback loop stays as small as possible.
Safety and transparency
We know teachers are rightfully critical of AI. That’s why from day one we have made clear promises:
🔒 Our own AI model – no OpenAI, 100% developed for education.
📚 Built in Europe – fully compliant with the AI Act.
🧾 GDPR-proof – your files belong to you, we don’t train on them, and we delete them immediately if you want.
🎓 Didactically trained – tailored to writing assignments and educational insights.
With these guarantees, we want to make sure AI is not something scary or distant, but a tool that helps teachers save time and helps students learn better.
👉 One thing is certain: AI cannot replace formative feedback, but it can strengthen it—even when it comes to creativity. It lightens the heavy workload for teachers and gives students the opportunity to take their assignments to the next level.
Table of contents
What makes good formative feedback?
The three big questions of Hattie & Timperley
The seven rules of Nicol & Macfarlane-Dick
How AI can support the feedback process 🤖✍️
How we approach this
Safety and transparency