How college professors are actually using Claude

While the world frets about students using AI to cheat on term papers, professors are quietly doing the exact same thing to survive their workweeks.

Teachers already save an average of 5.9 hours per week using AI tools, according to Gallup. Now, students are the ones getting nervous about faculty using AI in class.

Anthropic teamed up with Northeastern University to see what higher education professionals actually do with Claude. Researchers analyzed roughly 74,000 anonymized higher ed conversations from Claude Free and Pro accounts across May and June, combined with qualitative surveys from 22 faculty early adopters.

Heavy collaboration on lectures, hands-off on grades

The findings show educators use Claude primarily for curriculum development, followed by academic research and assessing student performance.

Some professors use Claude to build mock legal simulations, write vocational training modules, or draft recommendation letters. Northeastern faculty also reported spending an average of 29% of their AI time on personal learning.

Anthropic charted this behavior along an augmentation versus automation spectrum, revealing a massive split in how professors treat different duties.

When it comes to actual teaching and creating practice problems, educators stick close to the wheel, with 77.4% of sessions classified as augmentation. Writing grant proposals was 70.0% collaborative. Professors treat Claude as an intellectual sparring partner, iterating through multiple turns rather than accepting the first draft.

Administrative work flips the script. Managing institutional finances and fundraising saw a 65.0% automation rate, with professors happily offloading repetitive paperwork entirely.

Grading, however, sits in an uncomfortable gray zone. While grading accounts for just 7% of total educator conversations, 48.9% of those interactions leaned toward full automation, delegating rubric scoring and student feedback to Claude.

That dynamic has sparked immediate ethical pushback. As one Northeastern professor noted, students do not pay tuition for an LLM's time, they pay for the professor's time, and using AI to score papers remains unreliable and ethically fraught.

Interactive courseware and shifting curriculums

Educators are also moving beyond conversational chat. Using Claude Artifacts, faculty are whipping up interactive web simulations and customized classroom experiments that would have been prohibitively expensive or technically impossible before.

This shift is fundamentally altering what gets taught. In coding and analytics courses, professors report spending far less time debugging missed semicolons and commas, freeing up classroom hours for broader architectural and business concepts.

The new challenge is cognitive offloading. As AI output explodes across campus, faculty are racing to teach students how to fact-check synthetic material rather than blindly trusting the model.

Why it matters

The education debate has spent two years fixated on student plagiarism while ignoring faculty automation. This report reveals an uncomfortable reality: universities risk becoming closed loops where AI writes the assignments, AI designs the curriculum, and AI grades the submissions.

We think Anthropic deserves credit for openly flagging the 48.9% automation figure in grading. It exposes a genuine blind spot. If institutions allow automated grading to quietly normalize, they gut the core value proposition of higher education.

For educators, the immediate value here is clear. Treating Claude as an interactive courseware builder or a syllabus sounding board saves real time and enriches the classroom. But attempting to offload evaluation is a trap that risks eroding student trust.

For edtech and rival AI labs, the race is no longer just about catching cheaters. The real battlefield is building reliable, collaborative pedagogical tools that keep human judgment firmly in the loop.