ChatGPT’s answers have buttons now. When are they actually useful?

A paragraph is a slightly ridiculous way to explore a decision that keeps changing. Move one appointment, add a guest, change a budget, and suddenly you are asking for the whole answer again.

That is where ChatGPT's new interactive replies make sense. OpenAI announced GPT-6 with Intelligent UI on October 7: answers can combine writing with charts, forms, buttons and other elements you can use inside the conversation. It can still give a plain text reply. Sometimes that is exactly enough.

Our interest is less in the decoration than in the chance to change one assumption without starting over.

Ask for the part you want to move

Imagine planning a Saturday with two errands, lunch and a train to catch. A useful request might be: “Show the schedule and let me change the time I leave home. Keep the train departure fixed.” That is a prompt to try, not a result we have tested.

The distinction matters. “Make this interactive” gives the assistant a styling assignment. Naming the moving part gives it a job.

OpenAI's launch examples include visual explanations, comparisons and small calculators. The same release lets ChatGPT begin replying while it continues thinking or using tools. Both changes aim to make the wait feel less like staring at a locked door.

A nice chart can still be wrong

An interface has a quiet power: it looks finished. A slider with a tidy number beside it can feel more authoritative than a sentence, even when both came from the same uncertain input.

For a low-stakes calculation, check a case you can do in your head. If four people are splitting an $80 bill evenly, the base share should be $20 each before any extras. For a schedule, ask where the travel times came from. For a comparison, ask which details were verified and when.

Those checks are ordinary arithmetic and source checking. They do not become unnecessary because the answer has rounded corners.

Where the feature works—and how to turn it down

OpenAI’s feature guide places Intelligent UI in Chat, rather than Work or Voice. It supports reasoning settings from Instant through Extra High; the Pro reasoning option using Astra is outside that support. Account access and rollouts vary, so a missing control does not automatically mean something is broken.

You can ask for a different format in the conversation. The web guide also describes a Simple setting under Personalization, Layout and Visuals. It reduces interactive elements rather than promising to remove every one. Labels can differ as the interface changes.

Some components can retain their state after a refresh within the same thread, but that state does not follow you into another conversation. The model help page also says existing plan and tool allowances apply; this is not a separate unlimited playground.

Our take: the best version of this feature should make the controls almost forgettable. You change the thing you care about, see the consequence, and get on with it. If you are spending longer operating the answer than understanding it, ask for a paragraph.

For a more conventional starting point, try giving one familiar AI tool a small, repeatable job.

Checked October 23, 2026. Examples are suggested uses, not independent product tests.