Better answers with AI.
But are we learning?

Students score higher with an AI tutor.
Before a university calls it better learning, what should it check?

90 seconds · One fictional case

A fictional case. The study and AI-style reply below were written for this demonstration.

First, read the claim.

The claim

“Using an AI tutor improves what students can do on their own a week later.”

An illustrative AI-style reply

Supported.

“The AI group’s average score was 20% higher. That shows the tutor improved their lasting learning.”

Confident wording. A conclusion worth checking.

Now, examine the evidence.

What would you need to check before calling this better learning?

Choose one, then explore the reasoning.

Repeat the same study and assessment with 2,400 students. A week later, test both groups without AI on new, comparable questions about the same material. Ask whether the students who used AI feel more confident about the topic.

What remains when the AI is gone?

What was measuredWith AI
today

What needs testingWithout AI
a week later

The study measured performance while assistance was available. The reply turned that into a claim about what students could do independently later. Those are different outcomes.

A delayed test without AI would address the gap. If the group that studied with AI also performed better on new, comparable questions a week later, that would support a benefit for retained, independent performance on these tasks.

AI may help students learn. This result alone does not tell us whether it did.

Behind the question

From an answer to your own judgement.

My Nature correspondence asks universities to measure both performance with AI and what students can do independently. This fictional case lets you examine why that distinction matters.

01 · Nature · 3 November 2026

An educational illustration, not an assessment of your ability. No scores, saved answers or live AI calls.