Introduction
A UMD randomized trial, conducted in Fall 2025 and led by educational researchers in the College of Education and Division of Information Technology, found that student access to an integrated generative AI (GenAI) study assistant was associated with lower final grades and substantially lower learning management system (LMS) participation. Effects were estimated to be larger for first-generation students. These results do not establish that all AI tutoring is harmful, but they underscore the need for intentional instructional design and monitoring. The study does not provide a simple verdict on AI tutoring. It does show that access to a course-integrated AI tool is not, by itself, a learning intervention. How a tool is designed, configured, introduced, and embedded in course activities, and what it may displace, are likely as important as the tool itself.
Use the Virtual Study Assistant (VSA), and potentially any GenAI-based tool, such that it functions as a structured learning support, not as a substitute for engaging with primary course materials. When used with a clear learning purpose and guardrails, this and other AI-based tools have the potential to be beneficial to student learning, although systematic research in the higher educational context is still evolving. The practices below are intended to guide thoughtful use of GenAI technologies within teaching and learning spaces. It is important to note that the randomized trial referenced above did not establish that these practices will improve outcomes or eliminate the risks observed in the trial, but they are considered best practices within the evolving teaching and learning landscape.
This guidance also distinguishes recommendations for using UMD’s VSA from broader principles for AI-based tutors. Instructors considering other, non-UMD AI-based study and learning aids should also consider the study’s cautionary findings, without assuming that the same effects apply to every tool or context.
Using UMD’s Virtual Study Assistant
See below for some tips on how to use the VSA in a structured and thoughtful way.
Before introducing the assistant
Set expectations in the syllabus or course site. Tell students when and why to use the VSA, when not to use it, and that its output may be incomplete or incorrect. Preserve required engagement with readings, videos, problem-solving, discussion, office hours, formative quizzes, and instructor or TA support.
Prepare relevant course materials. Ensure that the course space contains sufficient, accurate, and well-organized materials for the VSA to draw on for the intended activities. Review the readings, explanations, and worked examples that students will need, and confirm which materials the VSA can access. Let students know that UMD’s VSA is specifically trained on the materials unique to the course and the benefits that this configuration should provide in terms of information accuracy.
Decide what it must not replace. Preserve required engagement with readings, videos, problem-solving, discussion, office hours, formative quizzes, and instructor or TA support.
Identify a specific learning purpose. Consider allowing the assistant for defined activities such as practicing a difficult concept, preparing for a problem set, or checking understanding after completing assigned material, rather than as a general “ask anything” course feature. Ensure that there are sufficient materials in the course space for the VSA to draw from for those activities.
Require an attempt before assistance. Prompt students to articulate their reasoning, show their work, or identify what they find confusing before consulting the assistant.
Set the VSA to tutor mode. In study mode, the agent provides direct answers to students' questions. In tutor mode, the agent engages the students through step-by-step explanations of the answers.
Suggested student-facing language
AI-based study tools, including UMD’s Virtual Study Assistant (VSA), are intended to support your learning, not replace your engagement with course materials, your own problem-solving, or conversations with your instructor and classmates. Use them only for activities permitted in this course and after making an initial attempt. Ask for help thinking through a next step, and verify responses using assigned course resources. These tools may make mistakes. For questions about grading, expectations, or if you are seeking additional clarity related to a course concept, your course instructor or TA serve as the best resources.
Monitor use and adjust
Review engagement indicators regularly. Watch for declines in LMS participation, page views, active days, assignment completion, discussion activity, office-hour use, or formative-assessment performance after introducing the tool. Course analytic data related to student engagement in the course space can be found on the left-hand side of your ELMS-Canvas page, where all available course integrations are also listed.
Explore the VSA admin console. Review the admin console for instructors to review recurring questions or topics that may need clarification for students. Use those patterns to inform follow-up explanations or instructional support, and check them against student feedback and assessments.
Solicit brief student feedback early. Ask what students are using the VSA for, how it is contributing to their learning behaviors, and where it might be helping or confusing them.
Be prepared to revise or pause. If course analytic data and feedback from your students related to the VSA appears to displace productive course engagement without evidence of improved learning, narrow the permitted use, revise the instructional approach, or pause its use.
General Advice on AI-Based Tutors
The following principles apply to UMD’s VSA and to other AI-based study tools. They are instructional recommendations, not claims that the UMD trial tested these approaches or established their effectiveness.
Design for learning
Favor guided inquiry over answer delivery. Configure or prompt the assistant to ask a question, identify a next step, offer a hint, or explain an error before providing a solution.
Connect use back to course materials. Ask students to locate the relevant reading, lecture segment, worked example, or course resource after interacting with the assistant.
Build in verification. For consequential or complex topics, direct students to validate AI-generated explanations against course materials or discuss unresolved questions with instructional staff.
Avoid positioning AI as an authority. The assistant should supplement, rather than replace, instructor judgment, disciplinary sources, and feedback from teaching staff.
Make learning visible
Create bounded, low-stakes activities. For example: “After attempting the first two problems, use the assistant for one hint; then submit a short note explaining what changed in your approach.”
Use reflection prompts. Ask students what they initially thought, what the tutor suggested, what evidence they used to assess the response, and what they still need to clarify.
Assess students’ own reasoning. Do not rely on AI interaction as evidence of learning. Students’ final work and assessments should continue to reveal their own reasoning and mastery.
Preserve traditional learning support structures. Encourage office hours and peer and instructor interaction. Frame AI tutoring as one option within a network of learning supports, not the quickest replacement for human help.
Consider equity and outcomes
Monitor and adapt. Use student feedback, engagement indicators, and assessments to determine whether the tool supports learning or replaces productive study behaviors. Revise or discontinue use of these tools when warranted.