iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
AI can turn a plain-language description into a proposed automation, but the result is a draft—not a workflow you should run without checking. Describe the trigger, the actions that should follow, any conditions, and where the result should go. Then review the suggested apps, connections, field mappings, and behavior, and test with safe, representative inputs.
What an AI-generated automation draft includes
Most workflow automations have a trigger—the event that starts the workflow—and one or more actions that happen afterward. An AI drafting feature uses your description to suggest that structure. It can reduce the need to start by choosing every trigger and action yourself, but it cannot guarantee that the suggested steps match your intent or are configured correctly.
A useful prompt describes the workflow in sequence, such as: “When X happens, do Y, then do Z.” Include the starting event, the intended actions, relevant conditions or exceptions, and the destination for the result. For example, say whether an action should happen only when a record meets a particular condition, rather than leaving that decision implicit.
Recommended Free Tools
How Power Automate and Zapier approach drafting
| What to compare | Microsoft Power Automate | Zapier |
|---|---|---|
| Starting point | Enter a detailed description in the “What will your flow do?” field. Copilot proposes a trigger and one or more actions. Microsoft’s instructions | Describe the workflow in natural language. Copilot creates an outline with a trigger and actions. Zapier’s instructions |
| Refining the draft | Add detail to regenerate the suggestion, then continue into the designer to complete and configure the flow. Microsoft’s instructions | Copilot can help configure and refine steps, add or replace steps, suggest field values, and explain fields. Auto-build is the default; ask-as-you-build requests confirmation before acting on suggestions. Zapier’s instructions |
| Setup and testing | Verify connections, finish setup, save, and test the flow. Microsoft recommends testing after saving. Microsoft’s instructions | Copilot can help test and refine the workflow. An unfinished Zap remains a draft if you leave the editor. Zapier’s Copilot instructions and Zapier’s Zap instructions |
| Access and status | For the documented flow-from-description path, Microsoft lists a work or school account and an environment based in Europe or the United States. It says a Microsoft account cannot be used to create cloud flows from a description. Microsoft says Copilot is optimized for English, with limited support for other languages. Microsoft’s instructions | Zapier describes Copilot as open beta and lists Free, Professional, Team, and Enterprise plans. Check your account for current availability because beta access may change. Zapier’s instructions |
Power Automate: describe a flow, then finish it in the designer
In Power Automate, enter the intended workflow in the “What will your flow do?” field. Copilot proposes a trigger and actions; you can give it more detail and regenerate the proposal before moving into the flow designer. There, check the proposed structure, verify connections, configure the required fields, save, and test. The documented language and account limits above apply to the flow-from-description path, not necessarily to every AI feature in Power Automate.
#1 Best Overall
Zapier: outline a Zap and refine its steps
Zapier Copilot starts with a natural-language description and an outline of a trigger and actions. You can use it to configure and refine the steps. Auto-build is the default mode; if you want Copilot to ask before taking action on a suggestion, select ask-as-you-build. Because the feature is documented as open beta, confirm whether it is available in your account rather than assuming access based on plan name alone.
How to write a prompt that produces a more useful draft
- Name the trigger. State the event that should start the workflow, such as a new submission or an updated record.
- Give the steps in order. Specify what should happen after the trigger, and include each action the workflow needs to take.
- State conditions and exceptions. Explain when an action should or should not run. If the workflow needs a decision based on a field value, identify that value and the desired outcome.
- Name the destination. Say where the information or result should go, and identify important details that should be carried over.
- Review the proposed structure before configuring it. Check whether the selected apps, trigger, actions, conditions, and sequence represent the workflow you described.
These details make the request more specific, but they do not ensure a correct draft. Treat the generated structure as a starting point, especially when a workflow has complex conditions or exceptions.
What to inspect before saving or running the workflow
- App and step choices: Confirm that the proposed trigger and actions are the ones you intended, in the right order.
- Connections and permissions: Check that each account connection is valid and has access to the information or operation the step requires. Microsoft specifically instructs users to verify connections.
- Field mappings: Inspect what information each step sends to the next one. Suggested values may not match the fields or data you intended.
- Conditions and exceptions: Confirm that the workflow handles the cases you described and does not take an action when it should be skipped.
- Test behavior: Save the flow where required, then run a test with representative, safe inputs. Check the actual result at the destination before relying on the workflow. Microsoft recommends testing saved flows, and Zapier describes testing and refinement as part of Copilot’s workflow support.
- Confirmation controls: In Zapier, choose ask-as-you-build if you want confirmation before Copilot acts on its suggestions; auto-build is the default.
Generating an outline does not make a Zap ready to run. Zapier says an unfinished Zap is saved as a draft when you leave the editor; review and finish its configuration before relying on it.
Drafting a workflow is different from adding AI to one
Asking an assistant to propose a complete workflow is different from adding an AI-powered step inside a workflow you are building. Microsoft documents generative actions as a preview that can suggest input and output types, connectors, and actions from a description; users can preview with sample inputs before adding an action to a cloud flow. The feature is prerelease and subject to change. Read Microsoft’s generative actions overview.
AI Builder prompts are another separate option: they run as actions within a flow rather than drafting the overall flow. Microsoft says availability can be limited by region, and use may be subject to limits or capacity throttling. The action formerly called “Create text with GPT using a prompt” was renamed “Run a prompt” in May 2025. See Microsoft’s AI Builder prompt guidance. Microsoft describes AI Builder as integrated with Power Apps and Power Automate for adding AI to automated processes. Read the AI Builder overview.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
