How well does AI feedback support academic performance?

Teachers are increasingly turning to AI to give real-time feedback to their students, as up to 75 percent claim they don’t have sufficient time to support those who are struggling.

The tools have proven popular with learners who get access to tailored advice, based on their learning history. But how well do they support academic performance, long term?

Dr Jill McGuire of Education Perfect says the evidence base underpinning AI tools is growing daily, but that questions around their efficacy are more nuanced than some think.

“None of the gains we are seeing happen solely because the technology is there. They happen because a teacher built the task and set the expectation. Take that away and you get faster progress through the lesson material, but no learning growth,” she said.

“But yes, when you couple good teaching with AI support, the answer is clear cut. We are seeing strong immediate, short, and long-term learning benefits amongst students who receive AI feedback.”

Correcting misconceptions before they harden

One of the best ways AI feedback tools can help is by correcting misconceptions before they become consolidated in a student’s long-term memory.

“Immediate and personalised matters because reasoning and knowledge retention is perishable. Our working memory can only hold approximately four items at once, and if a misconception isn't caught early, it gets built into the schema underneath everything that follows,” Dr McGuire said.

“A student who gets targeted feedback seconds after submitting can still connect it to the decision they made. A student who gets generic feedback a week later, is reading a comment about something they might have forgotten.”

Real time feedback can also help teachers spot patterns in the students’ understanding.

“They can walk into the next lesson already knowing which four students share the same misconception,” Jill said.

Shifting learning attitudes

Good feedback can also help students reflect on their work differently.

In an independent study of Education Perfect technology users, one teacher said her students did not reflect on their work unless specifically asked to do so, claiming: ‘they [students] don't look at the feedback, they look at the marks’.”

Mc Guire adds, “AI is great because instead of "what did I get?" students might start to ask, "where did my thinking break down, and what would I do differently?" That shift is much more likely when the feedback is separated from a mark.”

Willingness to try

There is emerging evidence that AI feedback could also increase a student’s willingness to try. When EP users were surveyed, all respondents said making mistakes on the platform felt safe, with one reporting: ‘it's not permanent, you just go back and do it again’.

Across 20,000 students over ten weeks, 87 percent went back and reworked a response after getting feedback on it. In a larger set of 200,000 responses, 69 percent of the low-scoring ones showed stronger understanding by the student's final attempt.

While these are preliminary findings, Dr McGuire says they are “not soft”.

“We will get harder evidence from our independent study, but this is a positive signal. If being wrong is low-stakes and instantly fixable, students attempt harder things, and the barrier drops most for the ones with the least confidence,” she said.

Better visibility

Another advantage of AI feedback is that progress becomes visible to the student. McGuire says this can help students develop metacognition, but only when it is coupled with coaching from a human teacher.

“AI can help students see exactly what they did to achieve their learning gain and a trajectory of what comes next, which is something most students never get to see.

“But I want to be careful here, because this is where the sector overclaims. Visibility is necessary but is not sufficient on its own. Showing a student their data does nothing with a context of what to pay attention to next and why.”

Through its independent study, EP has found metacognition emerging in classrooms where teachers made progress visible and then modelled how to read it.

“A teacher and platform that shows a student their data, and teaches them how to read it, is a motivating force for the learner and changes what the student does next.

“A Languages teacher in the study showed a struggling student his own data and described the moment he grasped the link between effort and outcome as a big eye opener for him.

“In the same school, where students were left with reflection prompts and no coaching, they clicked five stars and moved on. Same platform, different teaching,” McGuire said.

Opening up new possibilities

As well as the immediate learning advantages, McGuire points out that teachers who use AI feedback have more freedom to expand their lesson plans, highlighting that the technology was born out of the need to relieve an overburdened workforce.

“The evidence behind feedback as an intervention strategy was never the constraint,” she said. “The constraint was that a teacher with 120 students a day cannot possibly give high-quality feedback at the regularity required for meaningful learning gain.

“So, the gain isn't just "AI marked the work". The gain is what becomes possible once the mechanical part of generating feedback stops being the bottleneck for the teacher.”

Sharing more on what these possibilities might look like, Dr McGuire will join an expert panel at the upcoming AI in Education Brisbane Conference.

This year’s event will be held 8 October at the Stamford Plaza Brisbane.

Learn more and register your tickets here.