By Shelly Gussis

Generative artificial intelligence (AI) can either support student learning or limit it by bypassing the cognitive work students should invest to enhance their learning. This article explores how customized, course-integrated AI assistants can promote meaningful learning as well as provide academic support. It considers the importance of system instructions in creating an AI assistant grounded in sound pedagogical methods, arguing that effective educational AI depends less on the technology than on how it is designed.

One of Mark Twain’s apocryphal sayings declares, “I have never let my schooling interfere with my education” (Seybold, 2017). This distinguishes between the act of completing educational tasks and genuine learning. Regarding AI, many educators fear it interferes with genuine learning. But AI is unavoidably part of education now. According to the Harvard Graduate School of Education (2024), 82% of college undergraduate students use AI for schoolwork. Smutny and Schreiberova (2026) warn, however, that AI assistance may improve engagement and support without improving academic performance. Students often use AI in ways that reduce learning like copying AI generated text, offloading cognitive work, asking AI to complete assignments, and accepting AI responses without evaluation (University of Minnesota AI Hub, 2025b). However, universities can use AI to help students learn by providing a well-designed, course-specific AI assistant. When done properly, an AI assistant integrated in the classroom prioritizes deep understanding over assignment completion, and it all starts with system instructions. 

Much Ado About System Instructions

Shakespeare wrote one Hamlet, yet audiences have seen hundreds of different Hamlets. The script remains essentially the same, but directors interpret the character in radically different ways. One Hamlet is brooding, another heroic, another tortured, another mad or seemingly mad, and another deeply philosophical. In all these cases, the script provides the foundation, yet the director’s vision shapes the performance. Similarly, a large language model–ChatGPT, Gemini, Claude, Llama 3–provides the foundation for an AI assistant, while its system instructions determine how it behaves. The model remains the same, but the “performance” can vary dramatically depending on the instructions that guide it. Users witness only the performance–the responses generated during a conversation. System instructions essentially tell the AI assistant its identity, purpose, behavior, rules, and process. If the user asks the AI assistant to do something that violates the system instructions, the AI will not complete the task. 

The Director’s Notes: Creating an Effective AI Assistant

The system instructions should direct the AI assistant to enhance learning rather than simply provide answers. Dr. De Liu argues that an AI assistant should produce “guided discovery” (University of Minnesota AI Hub, 2025b). He compared students who used a traditional search engine, a standard ChatGPT-style assistant, and a customized guided discovery AI assistant. Results showed students using Google search performed better on later examinations than students using a conventional AI assistant. The advantage came because students had to think during the Google search process by identifying relevant sources, comparing information, evaluating credibility, synthesizing ideas, and applying what they learned independently. In other words, students actively engaged in the material and search results, strengthening their long-term learning. However, the standard AI assistant often provided immediate answers, enabling students to complete assignments more efficiently yet reducing the amount of thinking and engagement necessary for long-term learning. But, when the AI system instructions were designed for guided discovery that withheld complete answers, broke complex problems into manageable steps, supplied background knowledge, and prompted students to reason through solutions, students achieved the highest learning outcomes of all three groups (University of Minnesota AI Hub, 2025b).

These findings suggest that an effective AI assistant should integrate several evidence-based instructional approaches, including the Socratic method, guided discovery learning, instructional scaffolding, active learning, and coaching. Rather than immediately solving problems, the assistant should first assess what the student already understands, identify misconceptions, provide only the background knowledge needed to move forward, and ask progressively challenging questions that guide students toward constructing their own understanding (University of Minnesota AI Hub, 2025a). Wong and Chan (2025) also argue that educational AI assistants can draw on students’ previous responses and learning histories to provide adaptive support tailored to their individual learning needs. Hints and feedback should be offered throughout the learning process, with complete solutions provided only after students have made meaningful attempts to solve the problem themselves.

In addition, personalization should be a defining characteristic of an effective AI assistant. It should adapt explanations to each learner’s level of understanding, generate individualized practice activities, provide immediate formative feedback, recommend effective study strategies, and encourage retrieval practice rather than passive review. To maximize accuracy and instructional relevance, the assistant should ground its responses in instructor-provided course materials so that guidance aligns closely with course objectives and expectations (Belsky, 2025).

More Than Pedagogy, Horatio

Research such as Dr. Liu’s demonstrates how an AI assistant can improve learning through guided discovery, but a course-integrated assistant can also support course management and student success. Unlike a public chatbot, it would be available directly within the learning management system (LMS). An application programming interface (API) could allow a Gemini-based assistant to exchange approved information with the LMS, while an embedded interface or Learning Tools Interoperability (LTI) connection could authenticate users and provide access. The institution and LMS administrator would determine what the assistant could access, potentially including course materials, assignments, student work, grades, and instructor feedback. System instructions would dictate how it uses this information. The system’s logging, privacy, and access settings would determine whether instructors could view student data or conversations; these practices should be clearly disclosed to students. With appropriate access and safeguards, the assistant could:

To AI or Not to AI Is Not the Question

Students are using AI for schoolwork. The question is how higher education institutions can use it to help them learn effectively. With the right construction, an AI assistant integrated into the classroom can become a powerful, 24/7 learning partner that supports students, complements instructors, and strengthens both learning and retention. While the possibilities abound, this discussion has not addressed other concerns: FERPA compliance, data privacy, cybersecurity, algorithmic bias, accessibility, intellectual property, and institutional governance. Institutions must address these concerns through thoughtful policies, careful implementation, and ongoing oversight (The Chronicle of Higher Education, 2026). However, these challenges are not incompatible with effective educational AI. Institutions can design AI assistants that both protect students and foster meaningful learning. To become a lasting part of education, success will depend not only on choosing the right technology but also on designing it responsibly. In a world where instructors cannot always know whether students are using AI to benefit their learning, perhaps it would be beneficial to create a course-integrated AI assistant grounded in the science of learning and the goals of the course that helps students bridge Twain’s divide between schooling and education.

References

Blesky, L. (2025, July 30). How AI is transforming education. [Video]. The Open IA Podcast. YouTube. https://www.youtube.com/watch?v=QCLkJra0PjY

The Chronicle of Higher Education. (2026, April 28). Teaching critical thinking skills in the age of AI [Video]. Zoom. https://chronicle.zoom.us/rec/play/Nlp7kGu3FT3WBmNbtq4OMHgYxG0omIEzvOq386BkGEUDCB6OP-sAi2O3CDZwrspeiQRhiqeVfXh5CDh7.5jJRljQqYkmWWUDt?accessLevel=meeting&canPlayFromShare=true&from=share_recording_detail&startTime=1777399203000&oldStyle=true&componentName=rec-play&originRequestUrl=https%3A%2F%2Fchronicle.zoom.us%2Frec%2Fshare%2FMaGZQLuGKnjbVL4QOF9UNM4JIcWV4Omrr1WUxAYfLOizjAa_9eK0rsqlwK3eSy5H._ZpBodZvwjw9uB-r%3FstartTime%3D1777399203000

Harvard Graduate School of Education. (2024, September 12). How AI is shaping the future of education | Askwith Education Forum [Video]. YouTube. https://www.youtube.com/watch?v=KT-B2wZoY6g

Seybold, M. (2017, November 16). The apocryphal Twain: “I have never let schooling interfere with my education.” Center for Mark Twain Studies. https://marktwainstudies.com/the-apocryphal-twain/i-have-never-let-schooling-interfere-with-my-education/

Smutny, P. & Schreiberova, P. (2026). From rules to language models: A comparative study of chatbot learning assistants. Frontiers in Education, 11, https://doi.org/10.3389/feduc.2026.1794807

University of Minnesota AI Hub. (2025a, October 6). Scaffolding AI use in educational settings: An AI in higher education webinar with Danny Oppenheimer [Video]. YouTube. https://www.youtube.com/watch?v=QIEOmGOCGsY&list=PLjr-UljHeQ4xAbss6fwaB5oPTEmZHSicy&index=6

University of Minnesota AI Hub. (2025b, November 12). Harnessing generative AI for effective learning: An AI in higher education webinar with Dr. De Liu [Video]. YouTube. https://www.youtube.com/watch?v=RhIfsBhgQDM

Wong, A. K. L., & Chan, L. L. (2025). Integrating chatbots with learning management systems for personalized learning: A comprehensive review and framework proposal—the CLIF. Discover Education, 4, Article 523. https://doi.org/10.1007/s44217-025-00958-w

---

About the Author

Shelly Gussis is a full-time instructor at Purdue University Global and has been teaching college composition and writing for over 25 years.