Why AI can raise your marks but not your skills (and how to learn with it properly)

AI can make your work look better today while teaching you very little for tomorrow. That's the central finding of the OECD's Digital Education Outlook 2026: general-purpose AI tools can raise how well people perform a task without building lasting skill. Tools and habits that make you think, through hints, questions and explaining, do lead to real learning. The difference is in how you use them.

What did the OECD find?

The OECD's 2026 report, Exploring Effective Uses of Generative AI in Education, draws a line between two things we usually treat as the same:

• Performance: how good your output is right now.
• Learning: what you can still do later, on your own.

Its message is that general chatbots can push performance up while learning stays flat, because the tool does the hard part. Tools designed for learning, which ask guiding questions, give hints and prompt strategies, did better at building skill that lasts. The report urges governments to use AI selectively and for a clear teaching purpose.

Why does AI raise marks without building skills?

Learning needs effort in the right place. When you struggle a little, recall something, or explain it back, your brain stores it. When a tool skips that step, you get the result without the memory.

I see it constantly. Two learners build the same small app with AI. One accepts every suggestion. The other asks "why this?" each time. A week later only the second can change the app without help.

Five ways to learn with AI, not around it

1. Try first, then ask

Spend five minutes on the problem yourself before you ask. Even a wrong attempt makes the answer stick.

2. Ask for hints, not answers

Say: "Don't give me the answer. Give me one hint." Ask for another only if you're still stuck.

3. Explain it back

After the AI helps, explain the idea in your own words and ask it to check you. If you can't explain it, you haven't learned it yet.

4. Make it quiz you

End each session with: "Ask me three questions about what we just did." It takes two minutes and roughly doubles what you keep.

5. Do it once without the tool

The real test is doing a small version alone the next day. If you can, you learned it. If you can't, go back to step 2.

Where does AI genuinely help beginners?

None of this means "avoid AI". Used well, it removes the parts of learning that only cause frustration:

• Unsticking you. A hint at the right moment keeps you going instead of giving up.
• Explaining jargon. Asking "explain this like I'm new to it" saves hours of searching.
• Checking your work. "What's wrong with my reasoning here?" is one of the best questions you can ask.
• Making practice. It can generate endless small exercises at exactly your level.

All four keep you doing the thinking. That's the line to stay on.

How do you set up a chat so it teaches?

Most chat tools let you give an instruction at the start of a conversation. Paste something like this before you begin:

"I'm learning this topic. Don't give me full answers. When I ask something, give me one hint, ask me a question that makes me think, and only explain fully if I ask twice. At the end, quiz me with three questions."

This one paragraph changes the tool from an answer machine into a patient tutor. Save it somewhere you can copy it each time. If the AI slips back into giving full answers, remind it of the instruction. You're in charge of how it helps you.

Common mistakes I see

• Asking for the whole thing at once. Big requests give big answers you don't understand. Ask for one step at a time.
• Skipping the "why". Every time the AI changes something, ask why. It takes ten seconds.
• Never working without it. If you never try alone, you never find out what you know.
• Trusting confident answers. AI can be wrong and sound certain. Check anything that matters.

A simple weekly routine

• Monday: learn one new idea with AI as a tutor (hints, explain-back, quiz).
• Wednesday: use that idea in a small piece of real work.
• Friday: do a tiny version alone, with no AI, and note what you forgot.

Three short sessions a week, done this way, build more skill than a weekend of copying answers.

What if you're learning to build software?

Building with AI is a great way to learn, as long as you stay the one making decisions. In our Academy we keep the AI doing the typing and the learner doing the thinking: you describe, predict what will happen, check the result and explain the fix. That's the "designed for learning" pattern the OECD describes, applied to building real things.

A fifteen-minute exercise

Take something you asked an AI to do this week. Close the chat. Write down, from memory, the three main steps it took. Then open the chat and compare. The gaps are exactly what to learn next.

Takeaway

Better output isn't the same as more skill. Use AI to coach your thinking, not replace it, and measure progress by what you can do without it.

Source: OECD Digital Education Outlook 2026