Researchers using millions of student interactions on McGraw Hill’s ALEKS platform found behavioral shifts after ChatGPT’s launch that appear tied to outsourced problem-solving. The analysis differentiates “word problems” that can be copied into chatbots from “graphing problems” that are harder to outsource without additional reconstruction work. After ChatGPT’s arrival, students spent less time on word problems while time on graphing problems stayed relatively stable, with the gap widening each quarter. In supervised comparisons, time reduction on word problems largely returned to historical norms, but under unsupervised testing the faster work raised concerns about learning. The findings add to ongoing institutional debates over AI-assisted instruction, assessment integrity, and how to measure mastery when students can complete assignments faster.