The Forty-Five Day Test

A researcher gave 120 students the same assignment. Half used ChatGPT. They finished in 3.2 hours instead of 5.8 — and scored 5.75 against 6.85 on a surprise test six weeks later.

It isn't one study. MIT researchers found AI-assisted writers showed reduced neural connectivity in memory networks — and struggled to recall their own writing moments after producing it. Work from UT Austin, Georgia Tech and Hugging Face found short-term gains alongside weakened long-term retention and critical thinking.

The obvious conclusion is to stop using AI to study. It's the wrong one. The AI group didn't fail — they scored 5.75, in half the time. The problem was that the effort saved came from exactly the part where learning happens, and nothing was put back.

This 23-page guide keeps the speed and restores the difficulty, in one specific place.

Inside:

• Why retrieval difficulty is the mechanism — and why fluency fools you
• The rule: before and after, never during
• Six workflows that help, several better than studying without AI at all
• Six that quietly cost you — including the most popular one
• Three-pass retrieval, interleaving, and the gap log
• A study week, with adaptations for people studying around a job
• How to run your own forty-five day test

Instant PDF download · 23 pages · 2026 edition

Informational only. Check your institution's AI policy — rules vary and are changing.