Anthropic Education Report How Educators Use Claude

Anthropic found substantial delegation to AI within grading-related conversations in its 2025 educator report. It classified 48.9% of those conversations as automation-heavy, meaning the user appeared to ask Claude to perform the task directly rather than develop the answer collaboratively. Grading-related conversations were about 7% of those studied.

The analysis covered about 74,000 conversations selected for educator-specific tasks from Claude.ai Free and Pro accounts with higher-education email addresses. It therefore describes a filtered set of interactions, not the share of professors who automate grading. Nor could the researchers determine how much the generated responses influenced final grades. The finding establishes that delegation was being attempted, leaving its actual role in assessment unresolved.