Configure an AI by Zapier step to extract structured data by mapping earlier-step content, defining output fields, previewing the result, and sending those fields to later Zap steps.
Where an AI by Zapier Step Fits
AI by Zapier is a built-in tool that adds AI action steps to Zap workflows. Zapier describes a typical AI-powered Zap as a trigger, an AI action that works with trigger data, and a follow-up action that sends the transformed data to another app.
For structured data extraction, Zapier gives examples such as pulling invoice details, collecting action items from meeting notes, and capturing dates or deadlines from messages. Its AI by Zapier guide also describes extracting a project name, status summary, and owner from a chat message before writing those details to a spreadsheet.
Before adding the AI step, consider whether the task needs interpretation. Zapier says conventional automation handles work governed by fixed rules more reliably and less expensively than AI. As a practical recommendation, use predefined rules when they can produce the required fields consistently; reserve AI by Zapier for text that requires contextual interpretation.

Add the AI by Zapier Action
Zapier’s documented setup begins by logging in and creating a new Zap or modifying an existing one. Create the trigger first: select the trigger app and event, connect the account, configure the necessary fields, test the trigger, and click Continue. Then add an action step and select AI by Zapier from the built-in tools.
You can start with a prewritten prompt by clicking Use a template, or write a prompt in the text field. To insert data from an earlier step, type a forward slash (/) in that field and select the relevant information from the modal.
For an extraction workflow, a useful editorial suggestion is to map the source text that contains the details you need—for example, the message, transcript, or invoice text exposed by the earlier step. Ask for the specific values required downstream instead of requesting an unrestricted response.

Define Structured Output Fields
AI by Zapier automatically structures results into output fields that can be mapped into later Zap steps, and it also lets you define custom fields. From the AI step configuration page, Zapier says you can adjust or add output fields, choose whether to return output as line items, and preview the step output.
A practical suggestion is to create one output field for each value the next action actually needs. For example, an invoice workflow might use fields for an invoice identifier, date, and total, while a project-update workflow might use project name, status, and owner. These examples are possible field designs, not guaranteed extraction results.
Describe the desired content of each field in the prompt. If a downstream system expects a particular representation, explicitly request that representation. It is also sensible to state how the AI should respond when the source text lacks a value. These are editorial prompting recommendations; the resulting output should still be reviewed for suitability before it drives consequential actions.
Preview and Refine the Extraction
Zapier documents a preview-regeneration feature: after editing the instructions, you can press a button to see an updated result without running the entire Zap. The configuration page also provides a preview of the step output.
As an editorial recommendation, inspect whether the preview contains the intended fields and whether their values match the requested form. If it does not, revise the prompt or field definitions and regenerate the preview. This review method is a suggested quality check, not a guarantee that future inputs will produce identical results.
Zapier’s setup guide says to click Continue to return to the Zap editor, then add further steps that send the AI-generated content to its destination. Because AI by Zapier output fields are available for mapping, select the appropriate extracted field wherever a later action needs it. A cautious recommendation is to include human review when incorrect extraction could create a meaningful business, financial, or customer-facing consequence.
Conclusion
An AI by Zapier step can support structured data extraction by receiving mapped information from an earlier step, returning defined output fields, and making those fields available to later actions. Zapier’s current documented flow is to create the trigger, add AI by Zapier, map earlier-step data with /, adjust output fields, preview the output, and continue building the Zap.
The practical recommendation is to begin with a narrow extraction task and only the fields required by the destination action. Write explicit field instructions, preview the result, and refine the configuration before connecting it to higher-impact actions. If fixed rules can handle the input consistently, prefer conventional automation; if the text requires interpretation, use the AI step with an appropriate review process.
Sources
- How to build an AI-powered automated workflow | Zapier
- AI by Zapier: Easily add AI steps to your workflows | Zapier
Disclosure
This article was prepared with AI assistance.
Improve this guide
Report a problem or suggest the next question
Submissions enter a private editorial queue. They are reviewed and deduplicated before any public update or new page is created.
