The answer is becoming an interface

Sources checked on 8 October 2026.

Ask a simple geometry question: you have 24 metres of fencing for a rectangle. Which dimensions give you the largest area? A text answer can give you the result, the equation and a short proof. An interactive answer can let you stretch the rectangle, watch the area change and notice where it stops increasing.

Those are different experiences of the same explanation. The second gives the reader something to do with the relationship. Whether that helps depends on what the interaction reveals, and what the reader is supposed to learn from it.

OpenAI's 7 October announcement puts this change directly inside ChatGPT. Intelligent UI can compose answers from text, visuals and interactive controls, using a native component library that appears progressively. [1] The product announcement is a timely example of the format. By itself, it is no evidence that an interactive answer produces better understanding.

What changes when you can move the answer

For the fencing question, call one side x. The other side is 12 − x because the perimeter is fixed at 24. The area is x(12 − x). A 3-by-9 rectangle encloses 27 square metres; a 6-by-6 square encloses 36. You can explain that perfectly well in words.

A useful control makes one dependency visible: increasing the width forces the other side to shrink. The learner can check 3 and 9, then discover that swapping them leaves the area unchanged. The peak at 6 has a reason. The interaction is doing useful work when it makes that reason easier to inspect.

A decorative slider does less. If the graphic changes without showing the constraint, the user can move it around and miss the explanation. If the answer begins at the optimum and celebrates the result, there may be very little left to discover. An interface needs an instructional purpose as much as a visual one.

Make the assumptions visible beside the control

The example assumes all four sides need fencing, the shape is rectangular, the available length is exactly 24 metres and we ignore thickness and gates. Change the problem to a rectangle beside an existing wall, with fencing needed on only three sides, and the calculation changes. A fluent explanation can hide that assumption. A polished calculator can hide it even more effectively.

Against the wall, the best width is still 6 metres, but the enclosure becomes a 6-by-12 rectangle of 72 square metres instead of a square.

Keep the fixed perimeter next to the input. Show metres for lengths and square metres for area. Let the person inspect the equation. Give them a way to reset the example, and make the current values readable without relying on the shape alone. These choices do more for the explanation than adding another animated transition.

The chart uses integer widths from 1 to 11 metres, and you can hover over or tap any width to read its area. That is a teaching choice, not a restriction in the underlying mathematics. The full relationship is continuous between 0 and 12, excluding the degenerate endpoints. A good interface tells the reader when it has simplified the problem.

Choose the format from the question

The model does not have to build a small application for every question. An unnecessary interaction adds a discovery task: the reader now has to work out what can be clicked, what changed and how to get back. Sometimes a table solves the problem without asking the reader to learn a new interface.

There is a cost to variation too. If the reset button changes location or the same visual symbol means something different in the next answer, familiarity becomes less useful. For a learning product, I would keep common controls stable and let the explanation vary inside that structure. The content can adapt without making the learner rediscover navigation.

The Arabic version needs its own inspection

An English interface with translated labels is not the end of the work. In an Arabic explanation, text flows right to left while an equation or a Latin variable may need a different direction. A number must stay attached to the right unit. The control should make it clear which change increases a value, and that behaviour should make sense with both touch and a keyboard.

Those details matter when the interface carries the explanation. If the label is readable but the relationship is hard to follow, translation has preserved the words and lost part of the teaching. Check the actual layout at a narrow width, with realistic Arabic text, rather than judging it from short placeholder labels.

Accessibility belongs in this check from the start. W3C's guidance covers keyboard operation, visible focus, labels and programmatically exposed status messages. [2] For this example, the person should be able to adjust the value without dragging, read the changing result and obtain the same quantities from a text or table alternative. Colour alone should not carry the result.

Test what the person understands afterwards

A high click count would tell us that people clicked. It would not tell us that they understood why the square was the optimum. After the interaction, give them a related question with a different perimeter and ask for a prediction before they touch a control. Ask which assumption would change if one side used an existing wall. Those are proposed checks of transfer; nobody has run them on this example yet.

For a utility, the test might instead be whether someone completes the task correctly, notices a bad input and can recover from it. The evaluation should follow the purpose of the answer. A learning tool and a calculator can share an interface while having different definitions of success.

I would start a learning product with one recurring explanation where changing an input exposes something useful. Build that interaction, make its assumptions inspectable and compare what people can do after using it with what they can do after reading the text. That gives the interface a job. Without one, it is very easy to make an answer more elaborate while leaving the reader in the same place.

Sources

• [1] OpenAI: GPT-6 and Intelligent UI for everyone. 7 October 2026. Announcement and description of the native component library.

• [2] W3C: How to Meet WCAG 2.2. Reference for keyboard access, focus, labels and status messages. The cited principles are not a conformance claim for this illustrative example.