This module teaches how to build grounded prompts in AI Builder and reuse them across Microsoft Power Platform so that natural language responses are based on your own organizational data and business logic. A grounded prompt is one whose output is informed not only by the maker's natural language instructions but also by inputs the prompt receives and by Dataverse knowledge that the prompt is allowed to read. The course frames the work around a sales scenario built on three sample Dataverse tables (Proposal, Issuer, and RelatedParty), where the goal is to generate a short summary of a named proposal. Makers create prompts in the Prompt Builder, reached through the AI hub in Power Apps or Power Automate by choosing Prompts and Build your own prompt. In the prompt settings you add content of two kinds: inputs, such as a ProposalName text input that callers will supply at run time, and data grounding, where you connect a Dataverse table and select specific attributes. The instruction text is written in plain natural language, and you insert input tokens and Dataverse attributes (for example Proposal.description, Proposal RelatedParty Name, and Issuer Name) directly into the prompt so the model fills the response with live record data. Adding inputs and grounding lets you fine-tune and focus the response, and filters narrow the data to what is relevant. After defining an Output of type Text, you test the prompt with sample data and then save and publish it. The module then shows three consumption paths. In a Power Automate cloud flow you add the AI Builder action Create text with GPT using a prompt, pick the published prompt, pass the input, and read the Output Text downstream. In a Power Apps canvas app you add the prompt as a data source and call its Predict function from a control, binding the returned Text to a label. In Copilot Studio you finalize the prompt, create an AI plug-in, and call it as an action inside a Topic of a custom copilot, passing a user response variable and sending the result back to the user.
A grounded prompt combines natural language instructions, run-time inputs, and Dataverse data so the model generates responses based on your own records and business logic.
Makers build prompts in AI Builder's Prompt Builder, reached through the AI hub in Power Apps or Power Automate by selecting Prompts and then Build your own prompt.
In prompt settings you add inputs (such as a ProposalName text input) and ground the prompt by connecting a Dataverse table and inserting specific attributes like description and related Name fields into the instruction text.
You define an Output of type Text, test the prompt against sample data, and then save and publish it before it can be consumed.
In a Power Automate cloud flow the prompt is invoked with the Create text with GPT using a prompt action, where you select the prompt, supply the input, and use the Output Text in later steps such as Compose.
The same published prompt is reused in a Power Apps canvas app by calling its Predict function on a control and in a Copilot Studio custom copilot by creating an AI plug-in and calling it as an action within a Topic.