Feature request: Generative UI for interactive, stateful workflows in ChatGPT

ChatGPT can already create highly personalized plans based on a user’s goals, history, constraints, equipment, and schedule. However, executing those plans is still limited by the text-only interface.

Fitness is a good example. ChatGPT can generate a workout plan tailored to me, but recording each set still requires manually typing the weight, repetitions, completion status, pain level, and other feedback into the chat.

I would like ChatGPT to support Generative UI: interactive components generated dynamically from the conversation and the current task.

For example, a workout plan could include an interactive workout card with:

  • Exercise name and instructions

  • Target sets and repetitions

  • Editable weight and repetition fields

  • “Complete set,” “Add set,” and “Skip” actions

  • Rest timer

  • RPE, pain, and physical-condition inputs

  • Progress compared with previous sessions

Most importantly, the information entered through these components should become structured conversation state that ChatGPT can use later. It should be available for follow-up reasoning, memory, progress tracking, and automatic adjustment of future plans.

The interface should support both directions: users could modify it by clicking controls or simply telling ChatGPT what to change in natural language.

This should not be limited to fixed workout forms. The UI should be generated from the user’s intent and context, making the same capability useful for nutrition tracking, learning plans, project management, travel planning, routines, and many other workflows.

Generative UI could bridge the gap between conversational intelligence and the usability of traditional apps, allowing ChatGPT to become a genuinely personalized, AI-native application platform.

Thanks for laying this out so clearly, @zitongzhang098. The workout example makes the gap easy to understand: ChatGPT can create a personalized plan, but following and updating it still requires too much manual text entry.

We’re sending this to the team for logging as a request for dynamically generated, interactive UI components that can capture structured data, respond to both clicks and natural-language changes, and carry that state into future reasoning, memory, tracking, and plan adjustments across workflows like fitness, nutrition, learning, travel, and project management.

-Mark G.