A global review of 271 studies from more than 45 countries has found Australia has little local evidence to guide decisions about generative artificial intelligence (GenAI) in schools, prompting researchers to call for a major expansion of Australian research into how GenAI affects student learning and development.

Generative AI (GenAI) tools such as ChatGPT and Claude can create new content, including text, images, audio, video and computer code, in response to a user’s instructions.

KEY FINDINGS

• Australia needs more research to understand GenAI’s impact on school students
• GenAI can support learning, but outcomes depend on how it is used
• Purposeful teaching and learning design remain central to helping students learn effectively with, from and about GenAI, and to shape and use it responsibly

Researchers from the University of Sydney and the Barker Institute analysed 271 studies examining the impact of GenAI on young people from pre-Kindergarten (age 3) to Year 12 (PreK-12), making it one of the largest evidence reviews of its kind worldwide. 

The review found growing evidence that GenAI can increase student engagement, motivation and confidence, help students complete learning tasks and improve the quality of their work. However, it also identified major gaps in understanding how GenAI affects deeper learning, reasoning, self-regulation and students’ longer-term learning and development. 

Of the 271 studies, only four were Australian, highlighting the urgent need for more local research to guide schools, educators and policymakers. According to the researchers, this makes investment in Australian school-based research an urgent priority.

Empirical studies on GenAI use in PreK-12 settings have so far focused on academic performance, engagement, and self-regulated learning, the researchers found.  
 
A broader set of social, civic and ethical capabilities that PreK-12 education seeks to foster remains underrepresented in the current research base. These include complex problem solving, ethical and intercultural understanding, personal and social capability, as well as broader qualities, such as curiosity, courage and compassion.

Professor Lina Markauskaite, a Learning Scientist from the School of Education and Social Work, said the findings highlighted an opportunity for Australia to lead internationally in research-informed GenAI education. 

“What we found is not a simple story of GenAI being good or bad for learning,” Professor Markauskaite said. “The evidence suggests outcomes depend on how these tools are used, what students are asked to do, and the support they receive from teachers.”  

“At the same time, there is a clear need for much more Australian research. Schools, educators and policymakers are making important decisions now, and we need a stronger local research base to understand what works best for Australia’s students.”

The review found that learning outcomes are shaped by five interconnected conditions: 

• the learner
• the GenAI tool itself
• the learning task
• the social arrangements around learning
• and the broader cultural and institutional context. 

For example, students with strong prior knowledge are often better able to question and evaluate AI-generated responses, while less experienced students can be more likely to accept answers at face value. Similarly, a GenAI tutor designed to scaffold students through questions and hints, may encourage deep thinking, while an unrestricted chatbot may encourage answer-seeking. 

The review also found that task design matters. GenAI may support learning when it helps students practise skills, explore ideas, engage in reasoning or reflect on their understanding. It is less likely to support learning when it performs the very thinking students are meant to develop themselves. 

Teacher involvement remains critical. Across a range of studies, some of the strongest outcomes occurred when GenAI support was integrated with teacher guidance, combining immediate feedback and personalised assistance with professional judgement, subject expertise and human relationships. 

Researchers also identified access to technology, digital literacy, community expectations and attitudes towards knowledge and expertise as factors that may influence how students engage with GenAI and whether it supports learning. 

Dr Natasha Arthars, researcher in the School of Education and Social Work, said public debate about GenAI in education often overlooked the complexity of how students actually use these tools.

“There is a tendency to assume students will use GenAI to bypass learning,” Dr Arthars said. “While the technology can certainly make that easier, that is only one possible outcome.”

“The evidence shows students use GenAI in very different ways. Some use it to generate answers, but others use it to test ideas, seek feedback, explore different perspectives and deepen their understanding. For some students, GenAI lowers the barrier to asking for help, particularly when they might feel uncomfortable approaching a teacher or speaking up in class.

Dr Hongzhi (Veronica) Yang, Senior Lecturer in the School of Education and Social Work, said the key question is whether learning experiences are designed so that students remain responsible for the thinking, reasoning, judging and reflecting that underpin meaningful learning.

“Rather than taking a one-size-fits-all approach, guidance on GenAI should recognise that students at different stages of development will benefit from different forms of support,” said Dr Yang, who is also a core member of the Centre for Artificial Intelligence, Trust and Governance.

Professor Danny Liu, Professor of Educational Technologies and co-chair of the University’s AI in Education Working Group, said schools should move beyond simplistic debates about banning or embracing GenAI. 

“Every major technological innovation in education creates uncertainty, but history shows us that technology alone does not determine learning outcomes,” he said.

“Young people are entering a world where GenAI will be commonplace in higher education, the workplace and everyday life. Our responsibility is to help them develop the judgement, expertise and confidence to learn and think, which will then help them use these tools critically, responsibly and effectively.

“The strongest message from this review is that GenAI works best when it enhances human learning rather than replaces it.”