Popular discourse about generative AI in higher education is often driven by speculation rather than evidence. After three years of LLM availability, many rely on anecdotes, small studies, and self-report surveys, largely because systematic usage data has not been available. At Northeastern, where we have a strategic partnership with Anthropic and have deployed Claude across the institution, we can tell there is healthy and growing adoption of AI. The partnership provides a unique opportunity to understand AI usage among our community, and the first step is to look seriously at the adoption of Claude and other generative AI tools for teaching and learning. That way, whatever comes next, we can make decisions based on evidence rather than guessing.

Our research investigates several questions: who is adopting Claude, how are faculty and students using it, and what does responsible, effective integration for learning look like in practice? This piece previews our findings on adoption patterns and platform usage. The full report will land in the coming weeks and will examine how these patterns connect to learning outcomes, pedagogical approaches, and institutional strategy.

Is anyone using Claude?

The answer here is both simple and complex. The most straightforward answer is that Claude adoption and engagement is strong. When you look at weekly and daily active users, the trend line goes up across the Fall semester:

Overall, while over 32,000 people have signed up for a Northeastern Claude account, we prioritize reporting daily and weekly active users. Weekly active users have used the platform at least once in the past seven days, whereas daily users have used it in the past 24 hours. On a typical weekday, as many as 9,000 users engage with Claude; over the course of a week, that number reaches 15,000.  

The pattern suggests Claude is being used as a working tool tied to academic tasks—usage drops on weekends, and there are spikes around deadlines. On the chart above, the usage line generally moves up, except for Thanksgiving weekend, but there are accelerations over the first two weeks of the semester, around mid-October, and at the end of the semester. The three highest dates for project use were October 23, November 10, and December 02, suggesting a correlation with academic rhythms associated with midterms and finals.  

We see these numbers as strong but not indicative of broad overall adoption. One hypothesis was that enterprise access might significantly reduce use of other LLM tools, but the data don't support this. National surveys estimate AI usage among students at 80+%, while we see 43% Claude adoption at Northeastern. Students and faculty report using ChatGPT and Gemini alongside Claude, and network traffic to chatgpt.com routinely outstrips traffic to claude.ai.  

If faculty and students are using other platforms, then it is worth following up on possible reasons. For example: In informal conversations, students have mentioned prior accounts with another tool dating back to high school, or concerns that their Claude usage could be tracked. 

Key features that might help faculty adoption, such as sophisticated Canvas integration, collaborative chats, usage analytics for shared Projects, and clearer usability around Project sharing, are still in development. There could be insufficient or inconsistent training across the university. Faculty and student concerns around LLM use (e.g., environmental or philosophical concerns) should also not be discounted.  

As we’ll explain in a moment, no one should presume that these numbers tell the whole story about LLM adoption at Northeastern. 

Faculty adoption of Claude mirrors external industry trends 

Faculty adoption rates vary by college, which is partially explained by disciplinary focus and partly by composition of the faculty.  

This chart shows faculty signups expressed as a percentage of the total number of faculty members in each college. We should be cautious about overinterpreting “signups by college,” however. The College of Professional Studies, for example, has a wide diversity of programs, some of which have piloted AI tools extensively, which we have not explored yet in our research.  

Other aspects of the signups are not surprising, and mirror where AI tools are already embedded in professional practice, such as colleges with curricula focused explicitly on computer science, data science, or business. The variance invites questions about whether all students who might benefit professionally from greater AI readiness are finding the learning opportunities they need in their home colleges, and whether specific colleges are finding alternative tools more appropriate. 

Students are also engaging with Claude, but with reservations 

Many students have availed themselves of Northeastern's enterprise access to Claude. Roughly 43% of currently enrolled students have signed up for a Claude account, in patterns that largely track faculty adoption:  

Again, adoption rates by college (measured here as signups), show patterns of usage varying by discipline. In focus groups, students repeatedly tell us that they use Claude with mixed feelings. Some worry that the university or their instructors are surveilling their chats, and that as a result they are concerned about violating university policies, even inadvertently. The fact that network traffic to ChatGPT exceeds that to Claude suggests a preference for personal rather than institutional accounts, which may indicate behavior that students don’t want instructors to be aware of, particularly given that students anecdotally acknowledge pervasive cheating with AI. Importantly, there is not in fact any surveillance of chats. Northeastern does not monitor individual Claude conversations and analyses of usage data are aggregated and anonymized.  

Adoption isn't the same as sophisticated use 

When Anthropic markets Projects, a feature that preserves context and instructions across conversations, it describes semester-long workspaces where students might maintain ongoing dialogue about course material. This is a compelling idea, but the feature is not yet widely used. In rough numbers, there are 270,000 chats attached to a Project, but some 15 million overall chat activities, meaning that only ~2% of chats occur within a Project. 

More telling is the preponderance of short-term Projects over long-term ones. Our data show two distinct populations of Project users: "sprinters," who create a project and abandon it within days (about 75% of all projects), and "marathoners," whose projects persist for 30+ days with 10-100x more activity. The gap between these groups is stark. We're still learning more about Project use, but the early evidence suggests that few are using the opportunity to create custom chatbots across courses. 

What comes next? 

The usage patterns we’ve been describing here—deadline spikes that map to the academic calendar; ‘sprinter/marathoner’ splits in Project usage; and multi-tool preferences—raise questions about how AI is supporting learning. To investigate this, we’re combining usage analysis with pedagogical research. 

Northeastern has already begun building infrastructure to support thoughtful AI integration. Our Standards and Recommendations for the Use of Generative AI in Teaching and Learning provides suggestions for faculty to model AI practices appropriate to their field, including professional judgement about when not to use AI at all. A cross-college initiative to embed AI Readiness across the curriculum is an important next step. 

We are also starting to understand how faculty and students are using generative AI to support teaching and learning, and we are sharing this information across a variety of platforms: 

Our full report will include further usage analyses, additional faculty case studies, and information about pilots and experiments across campus. We're also continuing to develop our research pipeline: surveys of the full campus community are planned for the Spring 2026 semester, and we're working with Anthropic to access more granular, anonymized data that will let us understand usage patterns at the program level while protecting individual privacy. 

We are seeking to build a robust research program to study the use of AI in teaching and learning, and providing a first-cut of data is just the beginning. We will continue to update our research questions to support innovation and understanding of the impacts of AI use in the classroom. If you are interested in engaging further in this research, please get in touch at [email protected]. 

We gratefully acknowledge Daniel Seaton, Ilia Xheblati, Harun Gunasekaran, and Gloria John Rumao from the Office of the Chancellor data science team, who are working to make the usage data meaningful, and who built the dashboards and scatterplots previewed here. We also would like to thank Ryan Bender and John Clark from Information Technology Services for supporting access to the aggregated and anonymized Claude usage data. 

In this space, we share tips from Northeastern faculty members for integrating AI into teaching and learning.

Why Try It? Low-stakes AI conversations in Spanish prepare students for interviews and difficult conversations around sensitive topics, with built-in language feedback.

What they’re doing: Tania and Cristina teach intermediate Spanish language courses where students sometimes discuss complex and sensitive topics. These conversations may be hard to start for students and native Spanish speakers may hesitate to engage. For instance, for many Spanish people, discussing the Franco dictatorship remained a taboo topic for decades. To prepare students for real interviews with Spanish-speaking professionals, the instructors developed an AI persona activity using Claude Projects. Students create their own Claude project using instructor-provided prompts, then conduct extended text-based conversations in Spanish with a persona who “lived” through a Spanish or Latin American dictatorship. Claude stays in character throughout, corrects grammar in context, and provides a detailed feedback report at the end covering linguistic strengths and areas for improvement. Students submit their full chat transcripts along with a metacognitive reflection addressing (in Spanish), the following questions:

If you use this AI Companion again in the future, what will you do differently to make better use of the conversation?

What's working: Students reported lower anxiety and felt more prepared for subsequent real interviews. The AI persona invited questions students would never dare ask a real person, such as inquiries about family members who were harmed by the dictatorship. Some students connected deeply with their personas, noting parallel family histories. Interestingly, Claude did not exhibit the sycophantic agreement patterns that concern many educators. When students made historically inaccurate claims or defended dictatorships, the persona pushed back, remaining factually grounded while staying in character. Students also appreciated real-time grammar correction embedded naturally in conversation, and the metacognitive reflection helped them identify patterns in their language use. Several students noted this was their first experience with AI prompting, opening their eyes to productive uses beyond simple answer-seeking.

One limitation (and interesting result) emerged around assignment length. The instructors initially required 30 conversational exchanges, but students engaged so deeply that sessions stretched to 90 minutes.The text-only format also meant students practiced reading and writing but not oral communication, which remains essential for real-world interviews.

What's next: Future iterations will use time-based parameters rather than interaction counts. Additionally, the faculty plan to integrate voice capabilities once Claude's speech features mature, which would push students further outside their comfort zone while practicing oral fluency. They also see potential for asynchronous courses, where students could complete speaking practice independently with recorded sessions available for instructor review. Tania has already adapted the activity for an intermediate course focused on environmental topics, demonstrating the approach's flexibility across content areas.

It Was Never Just About Cheating by Eric Lars Martinsen
What if the real question isn't whether students cheat, but whether our systems of proof were never as stable as we believed? Martinsen argues that defending the take-home essay may actually mean defending bundled values, some timeless, some legacy. Disentangling them is the harder, more interesting path forward.

Reviewing students’ version histories for AI by Joseph Reagle
Process tracking is often framed as surveillance. Northeastern Faculty member, Joseph Reagle, uses it as a lens on writing habits, time allocation, and whether student work reflects earlier feedback. While transparency may not prevent misuse, it can clarify what's actually happening.

AI Shifts Expectations for Entry Level Jobs by Dan Page
What happens when the career entry point itself is shifting, and employers expect higher-order readiness from day one? Some roles may be vanishing; others are emerging. This realignment matters and suggests the real question for students is augmentation versus automation, and knowing the difference.

When: Friday, January 30 from 10am to 2pm (EST)
Where: 40 Leon St, Boston, MA 02115, USA
Who: For Northeastern faculty and staff only.
For questions and accommodations, contact: [email protected] 

When: Tuesday, February 3rd at 2pm (EST) 
Where: Virtual
Who: For Northeastern faculty and staff only.
For questions & accommodations, contact: [email protected] 

When: Monday, February 9th at 12pm (EST) 
Where: Virtual
Who: For Northeastern faculty and staff only.
For questions, contact: [email protected] 

Don’t forget to stay current with upcoming events from the units in the Division of Learning Strategy: The Center for Advancing Teaching and Learning through Research (CATLR) and Academic Technologies.