Unlocking Personalized Learning: The Potential of Large Language Models for Educational Feedback

The Transformative Power of Feedback in Higher Education

      Feedback is a cornerstone of effective learning in higher education, providing students with tailored support crucial for improving performance and achieving academic goals. It acts as a personalized guide, enabling learners to understand the quality of their work against expected standards and identify areas for future improvement. This process, by which students gain insights into their strengths and weaknesses, is vital for both academic and future professional growth. However, delivering high-quality, individualized feedback consistently can be a significant challenge for educators due to time constraints and scalability issues.

      In this evolving educational landscape, artificial intelligence (AI) is increasingly influencing feedback practices. Understanding the impact of AI, particularly Large Language Models (LLMs), on feedback generation is essential to harness its potential benefits and establish effective implementation strategies. A recent study, "Evaluation of Large Language Models’ Educational Feedback in Higher Education: Potential, Limitations, and Implications for Educational Practice," explored whether LLMs can provide valid support for teachers in producing educational feedback (Agostini & Picasso, 2026). The findings suggest a groundbreaking opportunity for generative AI in education, positioning it as a tool that could revolutionize how feedback is managed and delivered.

Defining Effective Feedback

      For feedback to truly foster learning, it must be well-structured and delivered effectively. Various theoretical frameworks underscore its multifaceted nature. Authors like Sadler (1989) and Dochy, Segers and Sluijsmans (1999) define feedback as a process where learners acquire information about their work, compare it to appropriate standards, and use this comparison to generate improved work. This perspective emphasizes corrective guidance and thought-provoking insights that enable students to self-assess and adjust their learning approaches.

      Hattie and Timperley (2007) categorize feedback into four distinct levels: task-oriented (correcting specific actions), process-oriented (focusing on strategies used), self-regulatory (fostering self-assessment abilities), and self-praise (which they argue can be counterproductive if generic). The most effective feedback bridges the gap between a student’s current performance and their desired goal, outlining the necessary steps to achieve it. Shute (2008) adds that feedback should reduce uncertainty, be supportive, timely, non-evaluative, and specific. Critically, Nicol and MacFarlane-Dick (2006) highlight principles such as clarifying performance standards, facilitating self-evaluation, offering high-quality information, encouraging dialogue, motivating students, reducing performance gaps, and informing teachers on adapting learning paths. These principles collectively underscore that effective feedback is prompt, frequent, detailed, outcome-focused, comprehensible, and actively utilized by students.

Evaluating LLMs in Providing Feedback

      To assess the capabilities of AI in this crucial area, the aforementioned study provided seven different Large Language Models (LLMs) with a structured rubric developed by a university instructor. This rubric defined specific criteria and performance levels for student-designed projects within a training course focused on inclusive teaching and learning. The LLMs were tasked with generating both quantitative assessments (scores) and qualitative feedback based on these guidelines. This approach allowed for a controlled environment to evaluate the AI's ability to interpret educational standards and formulate constructive responses.

      The generated AI feedback was then analyzed using Hughes, Smith, and Creese’s (2015) framework, which assesses feedback based on four key features: praise, critique, advice, and clarification. By examining these elements, researchers aimed to determine if LLMs could produce "well-formed" feedback capable of scaffolding student progress. The use of a platform connected to all models' APIs enabled simultaneous prompting of the LLMs in a generic, non-structured way, simulating a realistic scenario where an educator might request feedback without extensive AI-specific knowledge. This methodology provided critical insights into the practical application of LLMs in an educational context.

Key Findings: LLMs as Formative Feedback Tools

      The comprehensive evaluation revealed that Large Language Models possess a significant capability to generate well-structured feedback. The qualitative analysis and coding procedure, adhering to the Hughes, Smith, and Creese framework, indicated that LLMs could indeed provide elements of praise, critique, advice, and clarification. This suggests that, when properly guided, AI can produce feedback that is not only coherent but also aligned with established pedagogical principles for fostering formative learning experiences.

      The study concluded that LLMs hold immense potential as a sustainable and meaningful feedback tool. This is particularly true in higher education, where personalized feedback is often constrained by the sheer volume of student work and limited instructor time. By automating aspects of feedback generation, AI could free up educators to focus on more complex, high-impact teaching activities and deeper student engagement. However, the efficacy of LLM-generated feedback is highly dependent on clear contextual information and well-defined instructions provided to the models. This highlights the crucial role of human oversight and strategic prompting in leveraging AI effectively. For instance, incorporating advanced natural language processing capabilities, similar to those found in ARSA AI API suites, could further refine the nuance and specificity of AI-generated feedback.

Practical Implications for Educators and Institutions

      The emergence of LLMs as viable feedback tools has profound implications for educational practice. For universities and colleges, deploying such technology could:

• Enhance Scalability: Automate initial rounds of feedback for large classes, allowing instructors to manage more students effectively without compromising quality. This can address the "scalability issues" often faced in higher education.
• Improve Efficiency: Reduce the time educators spend on routine feedback tasks, enabling them to allocate more time to complex assessment, curriculum development, or one-on-one student support. This directly tackles the "time constraints" noted in traditional feedback processes.
• Boost Learning Outcomes: Provide students with prompt and frequent feedback, a critical factor for improved knowledge retention and performance. Timely feedback ensures students can act upon suggestions while the material is still fresh.
• Support Personalized Learning: While not fully replacing human interaction, AI can offer a foundational layer of personalized guidance, making learning pathways more adaptive to individual student needs.
• Standardize Feedback Quality: Ensure a consistent standard of feedback across different assignments and even different instructors, particularly if LLMs are guided by shared rubrics and guidelines.

      However, the findings also underscore the need for "clear contextual information and well-defined instructions." This means educators must be adept at crafting effective prompts and rubrics, serving as intelligent orchestrators of the AI’s capabilities. Institutions might also consider custom AI development for tailored solutions that integrate seamlessly with their existing learning management systems, a service that experienced since 2018, ARSA Technology is adept at providing.

Navigating Limitations and Ethical Considerations

      While the potential is significant, it is important to acknowledge the limitations. LLMs, despite their sophistication, require precise guidance. A generic, non-structured prompt might yield less impactful feedback compared to a prompt that meticulously outlines the desired tone, focus, and evaluative criteria. The quality of AI feedback is directly proportional to the clarity and detail of the input instructions.

      Furthermore, ethical considerations surrounding AI in education remain paramount. Institutions must develop guidelines that promote responsible use, ensuring data privacy, fairness, and transparency. The role of AI should be seen as augmentative, supporting human educators rather than replacing them entirely. It is crucial to maintain the human element in feedback, especially for nuanced interpretation, empathetic communication, and fostering mentor-student relationships. As AI technology continues to advance, the focus will increasingly shift towards designing hybrid feedback models where AI handles the quantitative and routine qualitative aspects, leaving educators to provide the deeper, more personalized, and critically human insights.

Conclusion

      The evaluation of Large Language Models' capacity to generate educational feedback in higher education reveals a promising avenue for innovation. With proper guidance through clear rubrics and contextual information, these AI tools can deliver well-structured, formative feedback that significantly enhances student learning experiences. This capability can alleviate the burden on educators, offering a scalable and efficient solution to deliver timely, high-quality feedback. As educational institutions continue their digital transformation journey, integrating AI for feedback generation, potentially through advanced video analytics, smart retail counters, or even customized AI solutions, presents a compelling opportunity to improve learning outcomes and operational efficiency.

      To explore how AI and IoT solutions can transform your educational or operational processes, we invite you to discuss your specific needs with our experts. Learn more about how ARSA Technology can deliver measurable and impactful AI-powered solutions.

contact ARSA

      Source: Agostini, D., & Picasso, F. (2026). Evaluation of Large Language Models’ Educational Feedback in Higher Education: Potential, Limitations, and Implications for Educational Practice. arXiv preprint arXiv:2602.02519. https://arxiv.org/abs/2602.02519