You have narrowed the gaming PC search to three machines. Then you decide to keep the old monitor, or spend less, or insist on something quieter. The comparison you just read needs revising. So does the next one. A few controls beside the answer start to sound appealing.
That is the sort of mess where a more flexible answer could help. OpenAI introduced GPT-6 with Intelligent UI on October 7, letting ChatGPT combine text, visuals and interactive controls inside a conversation. Think charts, buttons and small tools assembled around a question. The response can also stay plain text.
Let me change one thing
Imagine asking for a comparison of three PCs, then adjusting the budget, resolution or importance of noise. This is a proposed use case, not a test we have run. Its appeal is easy to understand: the questions you ask after seeing a comparison are often more revealing than the question you started with.
Perhaps the expensive option only makes sense if you intend to replace the monitor. Perhaps the cheapest tower stops being cheap once you add the storage you actually need. A comparison worth using would expose those dependencies and let you change your priorities without ordering a fresh essay every time.
The assumptions would need to stay visible as the comparison changes. A control labelled “performance” could mean frame rate in one game, an average across several, or an estimate with little evidence behind it. Moving the slider should make it easier to understand the recommendation, including what it leaves out.
The graph still owes you an explanation
The prettiest chart in the room can still be confidently wrong.
A polished interface gives numbers an air of authority that plain prose may not have. Before trusting it, the reader still needs to know how those numbers got there.
For a hardware comparison, start with the prices: where were they found, and when? Then check whether benchmark figures are measured results, estimates or illustrations. If changing the resolution moves a performance bar, there should be relevant evidence for that movement.
Ask for sources, a clear distinction between measurements and assumptions, and visible gaps where information is missing. Leaving a field empty is a perfectly useful response when the alternative is an unsupported figure.
A useful control earns its place by changing something that matters.
It also needs to be usable: readable labels, discoverable options and a way to operate it without a mouse. The conclusion and its reasoning should remain accessible to someone who cannot use the controls.
The rollout began globally on October 7 for Plus, Pro, Business and Enterprise, with Free and Go starting October 8. Enterprise access depends on admin settings. Crucially, the Pro reasoning option uses Astra and does not support Intelligent UI. A Pro subscription and that reasoning setting are different things.
For readers trying it, a modest first request seems more revealing than an elaborate demo: take a real comparison you already understand and ask for one meaningful adjustment. Check whether the result remains clear when you change it. Can you see why the recommendation moved? Can you get back to the original assumptions?
The test is whether adjusting the answer helps you understand your choice. In the PC example, that means being able to explain why one machine still fits after the budget changes, or why none of them does. The button has done its job when the decision becomes clearer.
Got a different take? I’d love to hear it. hello@maisfps.com
