Map controlled inputs into an AI Builder prompt and inspect its output before taking action.
Workflow & automation · Documented · Documentation checked 2026-10-04
Cloud flows, desktop flows, AI Builder prompts, and Power Platform environments.
AI Builder access, licensing or capacity, supported region, and connector policy apply. The action is named Run a prompt; older guides use Create text with GPT using a prompt.
The worked example is illustrative and has not been tested in a signed-in product account.
What it does and how it connects
Map controlled inputs into an AI Builder prompt and inspect its output before taking action.
AI Builder’s action is currently named Run a prompt, replacing older Create text with GPT using a prompt wording. It accepts mapped inputs and returns model output inside a flow. Capacity, region, and policy apply; downstream validation and review are separate actions.
Walkthrough
Use the product named in the access panel. The steps below combine documented controls with the editorial example.
- Create a sample instant cloud flow with a manual trigger.
- Add the documented Run a prompt action and create or select a supported prompt.
- Map controlled input fields and state the required output.
- Test and inspect the result before adding any consequential downstream action.
Worked example
Use a fictional workshop question with an approved agenda excerpt. Ask for a draft answer plus missing details. The model should not invent enrollment or payment status because those fields are absent from the input.
From the approved agenda text and question, draft an answer. Include the supporting agenda passage and missing details. If the agenda cannot answer, state needs_review. Do not infer registration, payment, identity, or consent. Return a draft only; no action is authorized by this response.
Inspect and verify the result
Compare output with the agenda and inspect the actual action result fields before mapping them onward. Test an absent answer and contradictory context. Validate shape and values outside the prompt.
Limits and troubleshooting
AI Builder capacity and regional availability can block execution. Older action names may appear in tutorials. A successful model call does not establish factual correctness or consent to a later action.
Data, permissions, and actions
Prompt inputs are processed through the documented AI Builder service. Flow run records can retain inputs and outputs. Use permitted minimal context, protect connections, and review data policy before adding private work material.
Related reading
Approval flows and human decisions
Route a proposed decision to an authorized reviewer and branch on the actual approval outcome.
From an AI prompt to a human-approved action
Place a reviewed draft between AI interpretation and a consequential cloud-flow action.
A desktop draft task with a safe failure path
Create a new draft in a sample desktop application under explicit machine boundaries.
Permissions and meaningful approvals
Decide what a system may read, what it may change, and when someone needs to authorize a concrete action.
Sources and verification
Documentation checked 2026-10-04. Examples are illustrative unless described otherwise.
Run a custom prompt in a flow · Microsoft