How to Prompt ChatGPT Intelligent UI for Forms and Charts

Intelligent UI lets ChatGPT answer with a working form, a chart or a row of buttons instead of a wall of text. This guide shows how to prompt ChatGPT Intelligent UI so you end up with an interactive mockup you can click through with a stakeholder in the same meeting. You'll leave with four copy-ready prompts and a way to turn the feature down when it gets overeager.

One framing note for FDEs: this is a ChatGPT product feature, not an integration path. OpenAI has not announced API availability, so use it to prototype and pin down requirements, then build the real thing in your own stack. The OpenAI help center article is the primary source for everything below.

Prompt ChatGPT Intelligent UI in five steps

1. Confirm you have it

OpenAI began rolling it out on October 7, 2026 in the Chat tab for Plus, Pro, Business and Enterprise, and on October 8 for Free and Go, according to launch coverage. Paid plans run it on GPT-6 Sol, Free and Go on Luna. Use the web app, since that is where the settings are documented.

Two things don't support it: the Pro reasoning option (GPT-6 Astra) and Voice. Pick another thinking level. Work and Codex are unchanged. On Enterprise, access depends on admin settings, so if it's missing on a work account, ask your admin instead of rewording the prompt.

There is no magic phrase. ChatGPT decides when an interface helps, so name the thing you want.

2. Ask for an interactive form

State the fields, the rules and what should update live. Requirements meetings go faster when the customer can type into something.

Build an interactive intake form for scoping an AI pilot with a customer.
Fields: use case (dropdown: summarization, classification, extraction, other), monthly request volume (number), data sensitivity (dropdown: public, internal, regulated), human review required (yes/no), success metric (text).
Show a live summary card that updates as I edit. Flag any field left empty. Add a reset button. Keep the layout compact.

The summary card is the useful part: it shows the customer how their answers turn into a scope. Compact layout is worth asking for, because the default can stack inputs awkwardly.

3. Ask for a chart you can steer

Give it inputs to change and a result to watch, not just data to plot.

Make a monthly cost calculator for an LLM-backed support assistant.
Inputs with sliders: tickets per month (1,000 to 100,000), average input tokens (200 to 4,000), average output tokens (50 to 1,000), price per million input tokens, price per million output tokens.
Show the formula, a bar chart of monthly cost at 25%, 50%, 100% and 200% of ticket volume, and a warning if cost passes a budget I can type in.
Use placeholder prices that I can edit. Do not claim they are current.

Use made-up or sanitized numbers. Prototype data doesn't need to come from the customer's systems.

4. Add buttons for a review flow

Buttons are good for mocking a human-in-the-loop step before anyone writes code.

Create a review-queue mockup. Show 6 invented support tickets, one at a time, each with a drafted AI reply.
Under each reply put three buttons: Approve, Edit, Escalate. Keep a running tally at the top and show a summary of counts when the last ticket is done.

ChatGPT chooses how the buttons behave, so click through every one before the meeting.

5. Revise, then test it

Follow-ups work like normal chat. Change one thing at a time.

Add a buffer percentage input, default 15%, that adds to the total. Show the arithmetic.
Then list every assumption as plain text so I can paste it into the project doc.

Then try zero, negative and boundary values. If you only get text back, reply with a request to show it as an interactive tool.

Dial it down with Layout and Visuals: Simple

When every answer sprouts a chart, go to Personalization > Layout and Visuals > Simple on the web. OpenAI says this reduces visual and interactive elements, but some may still appear, so it is not a text-only switch. For a single chat, ask for plain text in the prompt.

Gotchas

• It picks the format. A simple question can still get a plain answer, and no documented setting forces an interface.
• It looks more authoritative than it is. OpenAI says the model's design judgment still needs work. Check the formula and every input that feeds a result, not just the headline number.
• Treat it as chat-bound. The coverage I found doesn't describe saving or exporting a tool, so copy your assumptions into your own docs.
• Mind the data. Don't paste customer records into a prototype unless their policy and your plan allow it.
• Availability varies. Rollout was gradual and a better prompt can't unlock a feature your account doesn't have.

Intelligent UI won't ship to production, but it can get a customer to say what they actually need before you have written a line of the real system.