When AI Meets Institutional Reality: What Usage Data from 80+ Higher Education Institutions Actually Shows

What data from 280,000+ student-AI conversations across 80+ institutions reveals about AI implementation in higher education, and why the institutions seeing lasting results are treating AI as an operational decision, not a procurement one.

This piece draws on two years of LearnWise research, the 2025 State of AI Support in Higher Education Report and the 2026 State of AI-Powered Teaching & Learning Report, together spanning more than 280,000 real student-AI conversations across 80+ higher education institutions.

Most conversations about AI in higher education focus on technology. The more useful question is what happens after the contract is signed. Institutions are being asked to implement AI at scale during a period of constrained budgets, overstretched teams, and accelerating digital change, without the resources or infrastructure to absorb it well. The technology is moving faster than the conditions needed to use it effectively.

The gap between AI ambition and AI impact is not a technology problem. It is a context and capacity problem. The institutions closing that gap are not the ones with the largest budgets or the most sophisticated procurement processes. They are the ones that treated implementation as an operational question from the start.

Across the 2024-25 and 2025-26 academic years, LearnWise AI has worked with over 150 partner institutions globally. The findings in this article draw specifically on our two published research studies, together spanning 280,000+ conversations across 80+ institutions, HOGENT’s AI implementation journey, and the latest findings on AI readiness in higher education.

The strategy is there. The conditions to execute it are not.

The EDUCAUSE 2025 Landscape Study found that 57% of institutions now view AI as a strategic priority, up from 49% the year before. Yet only 2% are funding AI adoption through new budget lines. When asked what they were prioritising in their AI planning, institutions pointed consistently to faculty training (63%) and staff training (56%), not procurement or infrastructure. The human work of helping people understand and use new tools in their everyday context came first.

The January 2026 EDUCAUSE report on AI and work adds a sharper data point: 94% of higher education staff had used AI tools for work in the previous six months, but only 54% were aware of their institution’s policies on AI use. 56% had used tools not approved or provided by their institution. Staff are not waiting for top-down strategies. They are navigating AI already, with or without institutional support.

The bottleneck in AI adoption is not access to technology but the capacity to train people, maintain tools, align usage with institutional values, and build governance structures that make AI sustainable over time.

What a transition moment demands

In this context, educational institutions are considering if, or how, AI-powered solutions can be safely embedded in the student and faculty experience. HOGENT shows what AI implementation at scale can do for student and faculty support, meeting them where they are.

At HOGENT, the challenge didn’t start with technology, but with the sheer abundance of information available online. Both students and staff were increasingly struggling to find their way in this complex landscape.

The transition to Brightspace further amplified this challenge, acting as an accelerator for change and making the need for accessible guidance even more urgent.

HOGENT seized this moment to introduce LearnWise AI assistants: one supporting students and one helping staff in navigating both the new platform and the wider information ecosystem.

The results reflect sustained use rather than novelty traffic. “Ask HOGENT,” the staff-facing assistant, handled over 1,000 conversations from 400+ unique users between December 2025 and March 2026, with weekly volume still growing four months after launch. “Ask Wallie,” the student-facing assistant, handled 1,100+ conversations from 800+ unique visitors across the same period, with 79% of activity concentrated in January and February, tracking the academic calendar rather than the launch announcement.

The most common opening question on Ask Wallie appeared 33 times: “Waar kan ik mijn lesrooster zien?” Where can I see my timetable? A practical question, asked repeatedly, outside office hours, by students who needed an immediate answer and had nowhere else to turn. Across both assistants, more than 88% of conversations were grounded in cited HOGENT source material. That is what separates a tool people return to from one they abandon after a single session.

The assistants proved to be helpful beyond providing immediate support, serving as a tool for uncovering and filling knowledge gaps. Of the implementation, Emma Roose, Coordinator Educational Technology at HOGENT, shared: “As we introduced the AI assistants, it quickly became clear that their value goes beyond answering questions. They reveal patterns that point to missing, fragmented or difficult to find information. In this way, they not only support our users, but also help us to continuously refine and strengthen our information ecosystem.” 

The design principle the sector is still working out is straightforward: AI support tools are not drop-in solutions. Their value is a direct function of how well they reflect the institution’s actual context. A well-configured, institution-specific assistant absorbs a significant proportion of first-line queries. A generic one does not.

What 191,283 student conversations actually show

What does the wider landscape of AI adoption in education tell us? Our latest 2026 State of AI-Powered Teaching & Learning Report covers 191,283 anonymised student-AI tutor conversations and educator feedback actions logged across 56 global partner institutions in 11 countries between September 2025 and April 2026.

The resolution rate across those conversations was 99.4%, with a median time to a working answer of 3 minutes, compared to over 24 hours via email. But the more revealing finding is what students are actually doing in those conversations. The single largest category of student intent, accounting for 35% of all interactions, is practice: students asking the AI to generate quizzes, flashcards, and revision questions. A further 14% of conversations involve conceptual explanation, and 18% exam preparation. Students are using AI as a study partner, and the session data reflects that. Conversations average 8.9 messages, typically resolve in 3 minutes, and 15% of all sessions exceed 16 messages, lasting a median of 44 minutes.

52% of all engagement happens outside standard working hours, with 25% occurring in the late evening and pre-dawn window and a further 21.7% at weekends. For institutions serving working adults, online learners, and internationally distributed student populations, that figure represents something no staffing model can replicate: consistent, academically grounded support available outside of class and support hours, from students in 97 different countries across the academic year.

The content of those conversations also addresses one of the most common faculty concerns. In 58% of conversations, it names a specific institutional source in the response (a chapter, a module, a course video); in the rest, it is drawing from the same institutional scope through quiz turns, clarifying questions, or paraphrased explanations. When the AI could not provide an answer, it either asked a clarifying question or referred them to a human faculty member. The system is designed to route judgment back to people, not replace them.

On the faculty side, 31 of the 56 institutions used the LearnWise AI Feedback and Grader across 4,050 unique assignments and 17,937 finalised feedback actions. AI was involved in drafting 32% of those feedback actions: 22% were sent to students exactly as AI drafted them, and 10% were AI-drafted then edited before sending. The remaining 68% were written entirely by faculty, with no AI involvement at all. AI absorbs the cold-start writing burden where faculty choose to use it; faculty retain full control over what students receive. The narrative that AI replaces faculty judgment is not supported by how institutions are actually using it.

What the data suggests for institutions evaluating or expanding AI

Four conditions distinguish institutions where AI is delivering durable impact from those where it has stalled.

The first is deployment specificity. Institutions that ground their AI tools in actual institutional content, real policies, course catalogues, and LMS configuration, see substantially better outcomes. Generic deployments produce generic results.

The second is integration depth. Institutions with deep LMS integration and unified student-facing environments achieve higher engagement and completion rates. Standalone tools accessible only through a separate portal consistently underperform. AI needs to be where students and staff already are.

The third is governance that goes beyond the policy document. The EDUCAUSE 2025 Landscape Study tracked a sharp rise in institutions developing AI acceptable use policies, from 23% in 2024 to 39% the following year. Yet the January 2026 report found more than half of staff using AI for work had not encountered their institution’s guidance on it. Governance is not complete at the point of policy publication. It requires ongoing communication, clear boundaries, and meaningful institutional control over what the tool says and to whom.

The fourth is treating AI as a living system. The highest-performing institutions are not those that deployed AI and moved on. They maintain their knowledge bases, monitor usage patterns, update content as policies change, and use conversation data diagnostically to identify where students are struggling, where information is missing, and where human support still needs to step in.

Governance, integrity, and long-term vision

Academic integrity remains one of the most cited concerns when institutions evaluate AI tools. Data from the LearnWise 2025 State of AI Support report, covering 100,000 conversations across 24 institutions, puts the risk in proportion: exam cheating attempts across the dataset remained at 0.12%, and plagiarism incidents at 0.06%. Institutions with clear, transparent AI guidelines saw the lowest rates of academic dishonesty. Implementing institutionally approved tools and enforcing AI literacy are stronger safeguards than blanket technical restrictions.

Questions about data privacy, model transparency, and institutional control over the knowledge base are not obstacles to implementation. They are part of what makes implementation sustainable. A tool that creates compliance risk a year after deployment is not a solution. The institutions in the strongest position are those that ask those questions at the outset, and work with partners who engage them directly. A tool that creates compliance risk a year after deployment is not a solution, and no implementation timeline should treat governance as something to revisit later.

Join the conversation at OEB Berlin, 2–4 December

LearnWise AI and HOGENT will be presenting at OEB Global this December, focusing on how the AI assistants supported the Brightspace migration and helped reshape the institution’s information landscape. We’ll also be at the exhibition. Come find us if you want to talk through your own transition challenges.

The full data behind this article is available in two reports: the 2025 State of AI Support in Higher Education Report, analyzing nearly 100,000 conversations across 24 institutions, and the 2026 State of AI-Powered Teaching & Learning Report, drawing on 191,283 student-AI tutor conversations across 56 global partner institutions. Visit learnwise.ai to learn more.

Written for OEB Global 2026 by LearnWise AI, an AI platform purpose-built for higher education, embedded in the learning environments institutions already use.