A practical listener note

AI Tutor Checklist: The 4C Test

Do not judge an AI tutor by how impressive its answer looks. Judge it by what the learner can still do after the tool closes.

Use the checklist Download the one-page PDF Hear the evidence · 17:17

For parents, teachers, students and anyone evaluating learning technology.

The 4C test

Run all four checks. A fast answer may improve today’s task without building tomorrow’s independent skill.

  1. Coach Ask the tool for one hint or question—not the completed solution. The learner should make the next attempt. Check: does the AI create useful effort?
  2. Close Close or hide the AI. Continue the same task without copying, prompting or reopening the answer. Check: can the learner keep going alone?
  3. Change Try a genuinely different problem, not a near-copy. Ask the learner to explain why the method still applies. Check: did the idea transfer?
  4. Check later Return the next day. Retry or explain the reasoning before looking at notes or asking the AI again. Check: what remained after a delay?

Why the checklist focuses on independence

Two trials point in different directions because they tested different tools in different settings.

Structured tutor, positive result

A 2025 crossover trial included 194 eligible Harvard physics students. Its purpose-built tutor used expert scaffolding, targeted feedback and self-pacing. Students learned more in less time than in active-learning class sessions.

Read the Scientific Reports study.

Answer machine, later harm

In a separate field experiment with nearly 1,000 high-school mathematics students in Turkey, an answer-giving GPT-4 tool raised practice performance. After the AI was removed, those students scored about 17% lower than the no-AI group. Teacher-designed guardrails largely mitigated the harm, but did not produce a positive unaided-test effect.

Read the PNAS study.

Do not treat these as a head-to-head product comparison. They involved different ages, subjects, designs and tutoring systems. Together they make one evaluation question more useful: what can the learner do when the tool is no longer helping?

Hear the complete evidence review

We finished this 17:17 Evidence Verdict before making the checklist. It weighs positive, negative and synthesis evidence, names the limits of each study and excludes a retracted meta-analysis.

Start the full episodeFree · no sign-up · 17:17

Episode details, transcript and sources

Educational research commentary, not advice for a specific learner or a product endorsement. Published 2026-07-30.