Worked Examples: How AI Can Help You Teach Problem-Solving (with a Geometry Walkthrough)

Ask a class to "study hard" for a math test and most learners will reread their notes until the words blur. But reading is not the same as learning to solve. One of the most reliable findings in learning science offers a simpler tool: the worked example — a problem shown solved, step by step, with the thinking made visible.

This post covers what the research actually says, what a worked example is and how it works, a quick recipe for building one for your own topic, and a simple geometry walkthrough. At the end we look at how AI can help you produce these in minutes — without handing over your judgment.

What the research actually says

The worked-example approach is not a trend. It grows out of Cognitive Load Theory, developed by educational psychologist John Sweller in the 1980s. The foundational study — Sweller and Cooper (1985) — taught algebra two ways: one group solved a long set of equations the usual way, while the other studied worked examples paired with just a few problems to try. The group that studied worked examples spent less time, made fewer errors, and performed better on later problems.

The reason is working memory — the small mental "workspace" we think in. Facing a blank problem, a beginner has to hold the goal, search for a method, and do the calculation all at once. Much of that effort goes into searching, not learning. A worked example removes the search, so a learner's limited attention can go to how the steps connect — the part that carries over to the next problem. Researchers call this dependable result the worked-example effect.

There is one important limit, confirmed by later research (the expertise-reversal effect): worked examples help beginners most. As learners grow more skilled, fully worked steps start to get in the way, and they learn more by doing. So the support is meant to fade as competence grows — we will come back to that.

What a worked example is, and how it works

A worked example is simply a problem shown fully solved, step by step, that learners study before they try one alone. Instead of "Here are 20 problems, go," you first show one problem solved correctly — every step visible, and, crucially, the reason for each step spelled out.

It helps in three ways:

• It lowers the load. The hardest part for a beginner — figuring out where to start — is already done, so attention goes to the method.
• It models the thinking. A good example shows not just what to do but why: the quiet self-talk an expert uses without noticing.
• It builds a pattern. After studying a few, learners begin to recognize the type of problem and the path it calls for, instead of meeting each one as brand new.

How to build one for your own topic

You can turn almost any lesson into a worked example in a few minutes. The subject does not matter — the recipe is the same:

1. Pick one typical problem. Not the hardest, not a trick question — a clear, representative example of the skill you want.
2. Solve it correctly yourself, slowly. Write out every step you would normally do in your head.
3. Label each step. Number them or give each a short heading so the structure is visible at a glance.
4. Add the reason to each step. This is the part most teachers skip, and the part that matters most. Beside each step, say why you did it.
5. Cut the clutter. Keep the diagram, the numbers, and the words close together. Anything a learner has to hunt for is wasted effort.
6. Add one "why" question. A single self-explanation prompt turns passive reading into active processing.
7. Plan the fade. Make a near-identical problem with the last step or two removed — the faded example — for learners to finish.

Keep these seven in mind and the example almost writes itself. Here is the recipe applied to a geometry lesson.

A worked example in geometry

This follows the recipe above: one typical problem, labelled steps, and a reason on every line.

Problem. A right triangle has legs of 6 cm and 8 cm. Find the length of the hypotenuse, then find the area of the triangle.

Step 1 — Choose the tool. The triangle has a right angle and we know both legs, so the Pythagorean theorem applies: a² + b² = c², where c is the hypotenuse.

Step 2 — Substitute the known values. 6² + 8² = c².

Step 3 — Compute the squares. 36 + 64 = c², so 100 = c².

Step 4 — Solve for c. Take the square root of both sides: c = √100 = 10. The hypotenuse is 10 cm.

Step 5 — Find the area. For a right triangle the two legs are the base and height, so Area = ½ × base × height = ½ × 6 × 8 = 24. The area is 24 cm².

The two answers come from two different tools applied to the same figure — a small but important habit to model out loud. And the "why" question to leave learners with: why can we use the legs directly as the base and height here?

Turn it into practice (the faded example)

Once learners have studied the example, hand them a near-identical problem with the final steps removed. They do the part they are ready for, while the structure carries the rest.

Your turn. A right triangle has legs of 5 cm and 12 cm.

1. Write the Pythagorean theorem.
2. Substitute the legs: 5² + 12² = c².
3. Compute: ____ + ____ = c².
4. Solve for c: c = √____ = ____ cm.
5. Find the area: ½ × 5 × 12 = ____ cm².

(Answers: c = 13 cm, area = 30 cm².)

As learners grow more confident, remove more of the scaffold — give only the problem and the diagram, then eventually just the problem. That gradual fade is the whole point: the example is the training wheels, not the bicycle.

Where AI fits

Worked examples have one cost: good ones take time to write, and you need many. This is where AI earns its place — as a drafting partner, never the final authority. Once you have one solid example, AI can:

• Generate variations at the same difficulty, so practice never runs dry.
• Build faded versions — ask it to strip the last step, then the last two.
• Add the reasons to each step, modelling the self-talk learners need to see.
• Write self-explanation prompts that make learners process the example instead of skimming it.

The rule that keeps this safe: check every line. A model can produce a clean-looking solution with a wrong number or a skipped assumption. Treat its output as a first draft you edit — the way you would check a colleague's worksheet before it reaches learners.

The takeaway

Worked examples are a small change with decades of evidence behind them: show the solution first, make the reasoning visible, then fade the support as skill grows. The recipe is quick, it works in any subject, and AI makes it cheap to produce — as long as you stay the editor who checks every step.

Start with one topic this week. Solve one problem completely, write the reason beside each step, and build a faded twin next to it. Your learners get a clearer model of how to think — and you get a reusable set of materials instead of another evening lost to a blank page.