LearnWise Education Report: The 2026 State of AI-Powered Teaching & Learning

Higher education has spent the last two years asking whether it should adopt AI. That question is largely settled. The one that matters now is different: now that AI tutors and AI-assisted feedback are running inside real courses, at real scale, is it actually working, and is it working the way institutions hoped it would?

Today we're releasing the 2026 State of AI-Powered Teaching & Learning report, LearnWise's second annual look at how AI is actually being used across our partner institutions. Where last year's report tracked the shift from experimentation to early implementation, this year's edition asks a more specific question: once AI tutoring and AI-assisted feedback are live inside the LMS and the rest of the campus edtech ecosystem, how do students and faculty actually use these tools?

The report is based on the largest dataset we've analyzed to date. Read on to discover a few of our key findings. Get the full report diving deeper into the methodology, our data set, and our strategic recommendations for learning designers, program leaders, LMS admins and technologists, here. 

How LearnWise Analyzed 191,283 Real AI Tutoring Conversations

To build this report, we analyzed an anonymized and aggregated dataset of 191,283 real AI-led study sessions with the LearnWise AI Tutor and 17,937 finalized feedback actions through the LearnWise AI Feedback & Grader, across 56 partner institutions that agreed to participate in this research, spanning 11 countries, between September 2025 and April 2026. In total, students exchanged over 1.7 million messages with the AI Tutor across 96 distinct assistants.

Rather than a survey of attitudes towards AI, the report is a direct analysis of what happened in real conversations and real feedback workflows, making our findings useful for institutions trying to separate what AI adoption actually looks like from what it's assumed to look like.

What the Data Shows: AI Resolves 99.4% of Student Questions, Faculty Stay in Control

AI is resolving nearly all student questions, in minutes rather than days. Across the dataset, the AI Tutor reached a 99.4% resolution rate, meaning only 0.6% of conversations ended with the AI explicitly stating it could not help. Resolution doesn't mean the AI keeps students away from people: in 15% of conversations, the Tutor referred the student to a human resource, an instructor, a campus service, or the library, alongside its own answer. That referral behavior is deliberate, routing judgment back to the people best positioned to make it rather than trying to answer everything autonomously. The median time from a student's question to a working answer was 3 minutes, compared to the hours or even days a student typically waits for an email reply from faculty. This wasn't limited to standard hours: 52% of conversations happened outside standard business hours (08:00–18:00), in the late evening, pre-dawn, or over the weekend, when human support is not available.

Students are using AI to test their knowledge, not just to search for answers. The single largest use case, at 35% of all conversations, was study practice: students asking the Tutor to generate quizzes, flashcards, or revision questions. That behavior shows up directly in what the AI produces: it generated a quiz or self-check question in 46% of conversations, a summary or key-points recap in 39%, and walked through a worked example in 18%. Sessions reflect this too, averaging 8.9 messages each. The longest of these, 16 messages or more, run for a median of 44 minutes - less like a quick chat and more like a full AI-tutoring session.

Faculty are treating AI as a co-author, not a replacement. On the feedback side, faculty across the 31 higher education institutions in this year's AI Feedback & Grader research group were involved in AI drafting for 32% of the 17,937 finalized feedback actions: 22% were sent to students exactly as AI drafted them, and 10% were AI-drafted then edited before sending, with edits typically changing 20 to 30% of the text. The remaining 68% were written entirely by faculty, with no AI involvement at all. Faculty are always the ones editing and approving what a student ultimately sees, and the pattern shows them picking their spots twice: once on whether to invite AI into the workflow at all, and again on which drafts to trust unchanged. AI-assisted feedback reclaimed an estimated 1,160 hours of faculty time across the research cohort during the study window.

AI tutoring is functioning as a global layer of academic support. While the 56 partner institutions are based in 11 countries, the students behind these conversations logged in from 97 different countries over the course of the academic year, extending academic support well beyond the physical footprint of any single institution.

Does AI Replace Faculty? What 56 Higher Education Institutions Show

Taken together, these findings point to a specific, measurable version of what "human-in-the-loop" AI adoption looks like in practice: AI absorbing the repetitive, time-sensitive work, so that faculty time goes toward the mentoring and judgment a machine can't replicate.

As Greg Marschall, CEO of LearnWise AI, puts it in this year's report: "The most successful applications of AI haven't been those that replaced human interaction, but those that amplified it."

This is a baseline, not a finish line. As we look toward the 2026-2027 academic year, we expect these patterns to keep evolving as more institutions move from early adoption to sustained, institution-wide use, and we'll continue tracking what that shift actually looks like in the data.

Download the Full 2026 State of AI-Powered Teaching & Learning Report

Access the full 2026 State of AI-Powered Teaching & Learning report for the complete usage data, institutional case studies, and strategic recommendations for learning designers, program leaders, LMS admins, and Teaching & Learning teams, including an AI evaluation checklist and guidance on integrating natively across the LMS and wider edtech ecosystem.

Download the full report