AKAntonios Kioksoglou
Contact
Examples

ChatGPT Intelligent UI examples

On October 7, 2026, OpenAI said GPT-6 can answer in ChatGPT with buttons, charts and small tools instead of a paragraph. These are three mock-ups of my own, so you can feel the difference between a text answer and an interface answer.

What this page is, and is not. The mock-ups are mine. I built them after reading OpenAI’s post, GPT-6 and Intelligent UI for everyone, and I have not used Intelligent UI myself. They are not screenshots of ChatGPT, ChatGPT did not make them, and what it shows may look and behave differently. No model runs here, and the “streaming” is a timer. The arithmetic in each one has tests. The pizza recipe is generic, and I have not cooked from it at scale.

1 · An everyday question

A pizza night when you do not know how many are coming

You askPlan a pizza night for friends. I do not know yet how many are coming.

Answer as text

Pizza night for friends

Plan on about 1.5 pizzas a person, from 250 g dough balls (one 12-inch pizza each). Each pizza takes 150 g flour, 95 g water, 4 g fine salt, 0.3 g instant yeast, 60 g tomato passata and 90 g mozzarella, plus a few basil leaves.

Because the headcount is open, here are four sizes of party:

GuestsPizzasFlourMozzarella
46900 g540 g
691.35 kg810 g
8121.8 kg1.08 kg
10152.25 kg1.35 kg

If your count lands between two rows, take the larger one. A home oven with two trays bakes two pizzas in about 8 minutes, so 9 pizzas take five rounds, around 40 minutes.

Make the dough the evening before and chill it. Take it out of the fridge two hours before the first round, heat the oven as hot as it goes an hour before, and start baking when the first guests are hungry.

Answer as an interface

mock-up by me

Pizza night for 6 people

9pizzas
5rounds at the oven
40 minof baking

Shopping list

Flour
1.35 kg
Water
855 ml
Fine salt
36 g
Instant yeast
2.7 g
Tomato passata
540 g
Mozzarella
810 g

Plan for the day

Assumes 250 g dough balls, 8 minutes a round and a first round at 7:00 pm.

What the interface adds. The headcount is a control, so the list, the oven time and the plan follow it. The list copies in one tap, and the plan keeps track of what you have ticked off.

What it can show that the text did not say. Sixteen hungry people and one tray is more than four hours at the oven. Try it.

What to check. The amounts. Here they come from six numbers per pizza, all listed in the text answer.

2 · A tool, just by asking

A break-even ROAS calculator, built from one sentence

You askMy margin is 40%. What ROAS do I need on Meta to break even? Build me something I can play with.

Answer as text

Break-even ROAS is 1 divided by your margin. With a 40% margin that is 1 ÷ 0.40 = 2.5.

For every $1 you spend on ads you need $2.50 of sales just to cover the ads, because only 40 cents of each sales dollar is left after product cost, shipping and fees. At a ROAS of 3.0 you keep 20 cents of every ad dollar. At 2.0 you lose 20 cents.

Margin here means what is left of a sale before ad spend. This ignores returns, repeat purchases and tax, and the ROAS Meta reports may not match your store’s revenue.

Answer as an interface

mock-up by me

Break-even ROAS calculator

2.50

break-even ROAS at 40% margin

ProfitAt a ROAS of 3.0 you keep $20 for every $100 you spend on ads.

Break-even ROAS by margin

MarginBreak-even ROAS
20%5.00
30%3.33
40%2.50
50%2.00
60%1.67

Simplified on purpose: it ignores returns, repeat purchases and tax, and the ROAS Meta reports may not match your store’s revenue. Here is how I reconcile the numbers.

What the interface adds. Break-even stops being one number and becomes a relationship. Drag the margin down and watch the line you have to clear climb.

What to check. The formula and the definition of margin. Both are in the text answer, which is how you would catch a calculator that used different ones.

3 · Learning something hard

The Monty Hall problem, explained by letting you play

You askExplain the Monty Hall problem.

Answer as text

You are on a game show with three doors. Behind one is a car, behind the other two are goats. You pick a door. The host, who knows where the car is, opens one of the other doors and shows you a goat. Then you are offered a deal: stay, or switch to the last closed door.

You should switch. Switching wins two times out of three. Your first pick is right one time in three, and the host opening a goat door does not change that. The other two times, the car is behind a door you did not pick. The host must open the goat door, so the door that stays closed is the car.

Many people feel it is a coin flip, because two doors are left. It is not, because the host’s choice is not random: it always avoids the car.

Answer as an interface

mock-up by me

Pick a door. One hides a car, the other two hide goats.

Scores: staying against switching

Stay
0 of 0 won · none yet
Switch
0 of 0 won · none yet

The dashed lines mark one in three and two in three.

Count every starting case, with no luck involved
Car behindYou pickStay winsSwitch wins
Door 1Door 1YesNo
Door 1Door 2NoYes
Door 1Door 3NoYes
Door 2Door 1NoYes
Door 2Door 2YesNo
Door 2Door 3NoYes
Door 3Door 1NoYes
Door 3Door 2NoYes
Door 3Door 3YesNo
Cases won, out of 936

What the interface adds. You can lose, which is the point. After a few rounds the feeling that it is a coin flip gets hard to defend, and after a thousand games it is gone.

What to check. That the game follows the real rules: the host knows where the car is and never opens it, or your door. The table counts the same thing with no luck at all.

What the three have in common

  • The question sets the shape. A plan became a control panel, a formula became a calculator and an idea became a game.
  • A number is a control. Change one input and everything that depends on it follows, which a paragraph cannot do.
  • The interface can show what the text left out. The oven time, the line you are above or below, the thousand games.

The article explains what OpenAI announced, how it says Intelligent UI works and what I think it could change: ChatGPT Intelligent UI explained. These mock-ups are plain HTML, CSS and JavaScript with no model behind them.