What Should You Look for in an AI Tutor for Higher Education?

AI tutors have moved from pilot projects into everyday course delivery, and the tools carrying the label vary widely. Some are general-purpose assistants with a study mode. Others are built around a specific course, sit inside the LMS and give institutions control over how they behave. For teaching and learning leaders, that difference shows up in what students actually learn.

This guide explains what AI tutors are and what recent research says about when they help students and when they hold them back, because that evidence shapes what institutions should look for. It then sets out eight evaluation criteria, each with a question to put to vendors and a short note on why it matters, and closes by assessing LearnWise AI Student Tutor against the same criteria.

What are AI tutors in higher education?

AI tutors are AI tools that help students learn course material through guided explanation, practice and feedback, whenever they study. In higher education, the strongest are grounded in the student's own course and designed to build understanding rather than hand over answers.

Does AI tutoring actually improve learning?

It can, but design decides the result. In recent controlled studies, an AI tutor built on proven teaching principles outperformed an active learning class, while unrestricted AI access left students with lower grades once it was removed.

In a randomized controlled trial at Harvard, published in Scientific Reports (Kestin et al., 2025), students using an AI tutor built on the same pedagogical principles as an active learning class achieved median learning gains more than double those of the in-class group, in less time. A PNAS study of nearly 1,000 high school mathematics students (Bastani et al., 2025) shows the other side: once access was removed, students who had practiced with an unrestricted GPT-4 interface saw a 17% reduction in grades compared with students who never had it, while a version with learning safeguards largely avoided that harm.

The Harvard tutor was deliberately built on research-based teaching principles, and the PNAS study shows what can happen when those safeguards are missing. For institutions, that makes design the right lens for evaluating any AI tutor.

What criteria should you use to evaluate an AI tutor?

Evaluate AI tutors on eight criteria: course alignment, LMS integration, pedagogical design, institutional and faculty control, academic integrity, access, data governance, and evidence and insight. Together they test whether a tutor supports learning, fits how your institution works and can be governed at scale.

The table below pairs each criterion with a question to put to vendors and why it matters. The sections that follow explain each one in more detail.

Is it grounded in your course content?

An AI tutor should answer from the course materials the institution and instructors connect, not from general web knowledge. Grounding keeps explanations consistent with how a course is taught, the terminology it uses and the way it is assessed.

Ask how content gets in, which content types are supported in your LMS, and how updates come through when a module changes: automatically, on a schedule, or on request. Check whether answers can link students back to the relevant material, so the tutor reinforces the course rather than replacing it. The most useful test is your own: put questions from a real assignment brief to the tutor and have the instructor review the answers.

Does it work inside the LMS students already use?

It should sit inside the LMS courses students already open, whether Canvas, D2L Brightspace, Moodle or Blackboard, launching from their existing session. A tutor that needs a separate app adds friction at exactly the moment a student is stuck.

The strongest integrations work alongside the LMS and extend what it already does well, respecting course access and roles so each student gets support for the courses they actually take. For online and distance programs this matters even more. Teams comparing the best AI tutors or agents for online courses should ask whether students can get help across several enrolled courses in one place, and in which LMSs that is supported.

Does it teach, or does it do the work for students?

A good AI tutor guides students toward answers through hints, questions and worked steps instead of handing over solutions. The PNAS findings show why: without that design, students lean on AI during practice and then perform worse on their own.

Ask the vendor to show what happens when a student pastes an assessment question and asks for the answer, and whether that behavior is a default or something your institution configures. Look for practice features that build retrieval, such as quizzes and flashcards drawn from the course, and for instructions you can align with your academic integrity policy. Integrity is easier to protect when the tutor is designed for learning from the start than when restrictions are added afterwards.

Can faculty control how it is used?

Institutions should control where the tutor is deployed, and instructors should be able to shape how it works in their own course. That control is what turns an AI tutor from a student-side tool into part of course design.

Opt-in matters for adoption. At one of our partner universities currently testing an AI study assistant in their LMS, staff choose whether to switch it on for their courses. Ask what instructors can customize, such as answering style and supporting materials, whether they can see the themes in student questions without opening individual conversations, and who sets those permissions.

Is it available when and where students study?

It should be available around the clock and in the languages students learn in. Independent study rarely fits office hours, and many students learn in a second language.

LearnWise's own data reflects this. In the cohort behind the 2026 State of AI-Powered Teaching & Learning Report, 52% of AI Student Tutor conversations took place outside standard business hours. The study covered 191,283 Tutor conversations across 56 institutions in 11 countries between September 2025 and April 2026. Also ask for current accessibility documentation, so the tutor works for every student who needs it.

What data and governance controls should it have?

At minimum, it should support FERPA and GDPR obligations, hold independent security assurance such as ISO 27001 certification and SOC 2 Type II audits, and commit that student data is not used to train AI models. Access to conversation data should be role-based and controlled by the institution.

Ask where data is hosted, what conversation-retention settings are available, and who at the institution can view transcripts. Course materials should stay under institutional control, with clear arrangements for returning or deleting data when a contract ends.

What evidence and insight should it give back?

Ask for two things: evidence that the tutor supports learning in settings like yours, and insight faculty can act on, such as which topics students ask about most. A tutor that shows where students struggle helps instructors adjust teaching during the term.

Usage counts alone say little. Look for depth of engagement, such as multi-message study sessions and practice activity, then course outcomes over time. LearnWise's 2026 report found that AI Student Tutor conversations averaged 8.9 messages, closer to a study session than a quick lookup.

How does LearnWise AI Student Tutor measure up?

The AI Student Tutor integrates into Canvas, D2L Brightspace, Moodle and Blackboard, grounding its help in connected course materials. With Study Mode enabled, students can practice with course-based quizzes and flashcards, and it offers around-the-clock support in over 100 languages, subject to institutional settings.

In D2L Brightspace, it is offered as Lumi Tutor, powered by LearnWise AI. Institutions control where the tutor is deployed and which content it uses, with automatic refreshes triggered on use and manual refreshes available. When enabled, instructors can personalize their course tutor's answering style and supporting materials within administrator-set controls, and institutions can configure it to guide students through concepts and reasoning, using hints and questions in line with their academic integrity policies. Where enabled and supported by the LMS integration, students can ask about multiple eligible courses in one conversation.

Authorized administrators, and instructors where enabled, can use Tutor insights to review engagement and the topics students ask about most, and instructors can receive course-level question trends without individual conversation access. LearnWise is ISO/IEC 27001 certified and undergoes independent SOC 2 Type II audits, and institutional data is not used to train AI models. Data is hosted in the institution's agreed region, with configurable conversation-retention settings.

For a deeper look at deployment and use cases, read our institutional guide to AI tutors in higher education, or see AI Student Tutor.

To see how these criteria apply in one of your own courses, book a demo.