How to build an AI-powered automated workflow

If you set out to improve your AI fluency this year, and you've been using AI tools often and comfortably, fantastic. You're well on your way toward your goal. If you're ready to progress beyond submitting prompts manually, I recommend thinking about your work through an automation-first lens and building a Zap—what we call an automated workflow—that has an AI step.

In this post, I'll explain what AI-powered workflows are. I'll also show you two ways to build them—whether you use ChatGPT, Claude, Gemini, or any of the hundreds of other AI apps we integrate with. Let's get to it.

Table of contents

• What is an AI-powered Zap?

• What you can do with AI steps

• How to build an AI-powered Zap

What is an AI-powered Zap?

First, the basics: Zaps are automated workflows that have at least two steps. They need a trigger, an event that begins the workflow. And they need an action, an event that automatically happens after a trigger (or after other actions).

An AI-powered Zap has at least three steps. It still needs a trigger, which is usually followed by an AI action that somehow manipulates the data from the trigger. Then, a follow-up action sends that manipulated data to another app.

Here's one example:

In this Zap, whenever you receive an email that includes a certain keyword, Gemini will summarize the key points of that email and add the summary to a running list in a Google Doc of your choice. You get a neat, scannable document of whatever you want—like newsletter excerpts or team updates. 

Want to build this Zap yourself? Skip ahead to see how.

What you can do with AI steps

AI tools are scary good at so many different kinds of tasks. I write about AI all day long, and even I struggle with decision paralysis when thinking about the best way to pair AI with automation.

A helpful framework I always come back to? AI excels at these five things: analyzing, classifying, extracting, generating, and summarizing data. Here are some examples of what that could look like:

• Analysis: Evaluating sentiment in customer feedback, identifying trends in data, assessing the urgency of support tickets

• Classification: Sorting emails into categories, tagging content by topic, organizing leads by priority level

• Extraction: Pulling key details from invoices, grabbing action items from meeting notes, capturing dates and deadlines from messages

• Generation: Drafting email responses, creating images from text descriptions, writing product descriptions from specs

• Summarization: Condensing long emails into key points, turning meeting transcripts into action items, distilling articles into highlights

If you're still not sure which automated workflows would benefit from the addition of AI, ask yourself these questions: What tasks do I do repeatedly? What work drains my energy? Where could delegating to AI free me up to do work that actually needs my creative, human judgment?

Your answers are going to point you to solid candidates for AI automation.

How to build an AI-powered Zap

Let's practice building these kinds of Zaps using the example I showed you in the previous section: a workflow that connects Gmail, Gemini, and Google Docs. (But you can feel free to replace Gemini with a different app, like ChatGPT or Claude.)

To get started, click into the Zap template below. If you want to practice using other apps, you can go straight to the Zap editor and begin from scratch.

First, set up your trigger. If you're using the Zap template, this will already be set up for you. If not, pick Gmail as your app and New Email Matching Search as your trigger event, then connect your account. When you're done, click Continue.

You'll be taken to the Configure tab. This Zap will only trigger if an email contains a certain keyword—and this is where you'll specify that keyword. For multiple keywords or to limit your searches, use search operators. For example:

• urgent OR ASAP (this finds emails containing either keyword)

• in: newsletters (this limits the search to your "newsletters" mailbox)

• subject: "new meeting request" (this searches for this exact phrase in subject lines)

• from: junebug@woof.com (this finds emails from this specific sender)

I want to filter for emails with "update" in the subject line, so I configured that field accordingly.

Click Continue, then click Test trigger. Zapier will look for emails that match your search keywords. Select a test record and click Continue with selected record.

Now it's time to set up your action app. If you're using the Zap template, this will already be done for you. If not, search for and select Google AI Studio (Gemini) in the modal that appears. (Or whatever other app you're using.) Choose Send Prompt as your action event and connect your account.

If it's your first time connecting your AI tool to Zapier, click Connect a New Account. You'll be prompted to enter your API key—the unique string of characters tied to your account.

Finding your API key is similar across AI tools, but for Gemini, it's specifically located at aistudio.google.com. Once you're there, go to the bottom-left corner, above your email address, and click Get API key. Pick a key if you have one set up already. Otherwise, click Create API key.

A modal with two dropdown menus will appear. Give your key a name, assign it to a project, and click Create key.

Copy the generated key, then return to Zapier and paste it into the field. Then click Yes, Continue to Google AI Studio (Gemini).

Back in the Zap editor, you should see your connected account. Click Continue.

In the Configure tab, you can modify a few settings, including your Model and Prompt. I'll go with the newest model and write a quick prompt for summarizing an email. After the prompt, I need to map the field to my previous Gmail step, so Gemini actually has content to work with.

To map data, click the plus sign (+) or type a forward slash (/) in the field, then select the data you want to map. (I'll pick the body of the email, since that's what I want Gemini to summarize.)

You should now see the mapped data in the field.

When you're done configuring your step, click Continue. You'll see the data that'll be sent to your Gemini action step when it runs. Click Test step—this will show you the AI's output. If you want to tweak it, go back to the Configure tab and play with your prompt a little. If you're happy with the result, it's time to add your last action step.

If you used a Zap template to build this Zap, this will already be set up for you. If not, click the plus sign (+) under the Gemini action step to add a new step. Pick Google Docs as your app and Append Text to Document as your action event, then connect your account. When you're done, click Continue.

From the Configure tab, you can pick the document you want your AI-generated summary to write to. And you can specify what, beyond that summary, you want to add. I want to keep a running list of team updates, displaying the subject line and date of the email before the summary—so I'm going to fill out that field and map data accordingly. This field supports HTML tags, so you can adjust your text to make sure it looks exactly the way you want it to.

When you're done, click Continue. And in the next tab, click Test step. Check the document you created to make sure you see what you expected. Then, don't forget to click Publish to let your Zap go live.

Here's what my document looks like:

Get inspiration with Zapier Copilot

What if you don't know what you want to build yet? Zapier Copilot is perfect for this situation. It’s our built in AI-powered assistant, capable of not just building and configuring workflows, but also brainstorming ones for you too.

To get started, click Send this prompt to Copilot.

Alternatively, you can go to your Zapier dashboard and provide Copilot with a more specific request, such as:

• "Suggest AI-powered Zaps I can build using Slack, Gmail, and ChatGPT."

• "What RevOps workflows can I build with Claude?"

• "Show me AI workflow ideas based on the apps I have connected."

After you submit your request, Copilot will share some ideas with you.

Pick one and ask Copilot to build it (you can use natural language!). It may ask you some follow-up questions to learn more about what you want to happen in your workflow. Or it'll just confirm your request. When prompted, click Continue with Zap editor to let Copilot start building.

Copilot will take care of setting up the steps for you. You can even chat with it to make adjustments or troubleshoot issues. Just remember to test and publish your Zap when you're done.

Harness the power of AI and automation

By automating with AI steps, you can squeeze more out of your favorite AI models, leveling up your fluency in the process. As you build your skills, you'll naturally progress to creating more sophisticated workflows. And with Copilot guiding you along the way, you can tackle complex challenges without even needing a developer. So, what will you automate first? 

This article was originally published in June 2023 by Elena Alston. It was most recently updated in January 2026.