How to Use Gemini in Zapier with AI by Zapier

Introduction

Gemini is most people’s first stop for a quick draft or a fast answer, but a chat window is only the start of what it can do. The real payoff comes when Gemini reasons inside a workflow you already run: classifying every support ticket, summarizing every call recording, or drafting replies to new emails without anyone opening a tab. Zapier added exactly that in November 2026. Gemini models are now available as steps inside any Zap through AI by Zapier, and Zapier is also live as a connector inside Gemini Enterprise. This tutorial shows how to add that step, pick the right model, and keep the task bill under control.

Quick answer

Add an AI by Zapier step to a Zap, pick a Gemini model, write a prompt that maps fields from steps, and define output fields before publishing. Flash Lite costs 1 task per run, Flash models 3, and Pro 5.

Evidence

Zapier’s own announcement puts the scale of the integration in plain numbers. Gemini steps run inside Zaps that can reach more than 9,000 apps in Zapier’s library, and the cost of a Gemini step is billed in Zapier tasks rather than on a separate Google bill. The model lineup, published on the Zapier blog’s Gemini automation guide, is short enough to memorize: Gemini 3.5 Flash Lite costs 1 task per run, Gemini 3.5, 3.6, 3.7, and 3.8 Flash cost 3 tasks each, Gemini 3.1 Pro costs 5 tasks, and Gemini 4 Argon requires bringing your own API key. AI by Zapier is a paid-plan feature: Zapier’s pricing page lists Professional at $19.99 a month billed annually for 750 tasks, while the Free plan’s 100 tasks exclude AI steps.

Zapier states the design rule this tutorial follows: “AI only shows up when it’s needed, which helps you save tokens (and avoid unnecessary hallucinations).” Deterministic steps should stay deterministic: if a Zap only logs form responses into a spreadsheet, no model needs to be involved. Gemini earns its tasks when the work needs judgment: classifying hundreds of support tickets a day, scanning competitor pages for keyword opportunities, or summarizing a pile of customer reviews. Zapier also publishes AutomationBench, its own benchmark for how AI models handle real business workflows, which is the fastest way to compare Gemini against other models for a specific task type before you commit a Zap to one.

Prerequisites

• A Zapier account on a paid plan (Professional, Team, or Enterprise). AI by Zapier steps cannot be published on the Free plan.
• An existing Zap to add the Gemini step to, or a trigger app such as Google Forms, Gmail, or Slack.
• A rough idea of the monthly volume. A Zap that runs 400 times a month with a 1-task Gemini step plus one Slack action costs 800 tasks a month, which already exceeds the Professional plan’s 750 tasks, so volume decides the model tier.
• Optional: a Google AI Studio API key, only if you want the dedicated Google AI Studio (Gemini) app or a bring-your-own-key model such as Gemini 4 Argon.

Step-by-step guide

Step 1. Pick a Zap where judgment is the bottleneck

Open a Zap that already works end to end and find the step where a human currently reads something and decides what happens next. The worked example in this tutorial is the one Zapier itself uses in its guide: a Google Form collects feedback all week, Gemini reads each new submission and decides whether it needs attention from the product team, and only flagged submissions go to Slack. The pattern transfers directly to support ticket triage, lead scoring, and inbox sorting. If the Zap has no such decision point, do not add a Gemini step. A model that copies a field into a sheet on every run is a task bill with no benefit.

Step 2. Add the AI by Zapier step after the trigger

In the Zap editor, click the plus icon after the trigger step and search for “AI by Zapier”. Select it as an action step. In the step setup, open the model dropdown and choose a Gemini model. For a first run, pick Gemini 3.5 Flash Lite at 1 task per run: it is usually good enough for classification and short summaries, and you can change the model later without rebuilding the Zap.

Step 3. Write a prompt that names its inputs and its output

A good AI by Zapier prompt has three parts: the job, the input fields, and the exact output shape. Map the trigger’s fields into the prompt rather than pasting static text. For the feedback triage example, the prompt looks like this:

You are triaging product feedback. Read the submission below and decide
whether it needs attention from the product team this week.

Feedback: {{Google Forms response}}
Submitted by: {{respondent email}}

Reply with exactly two fields:
- needs_attention: yes or no
- reason: one sentence explaining the decision

Keep the prompt short and specific: vague prompts produce vague output that cannot be mapped reliably into later steps. If the step needs reference material, attach it as a knowledge source instead of pasting it into the prompt. AI by Zapier supports sources such as Google Docs, Notion pages, and uploaded documents.

Step 4. Define output fields before you test

This is the step most people skip, and it is the one that decides whether the Zap is reliable. In the AI by Zapier step, add an output field for each value later steps need. For the triage example, add needs_attention as a yes/no field and reason as a text field, and mark both required. Without output fields, the step returns one combined block of text, and the filter in the next step has to parse sentences instead of reading a clean value. Test the step with a real submission from your trigger app and read the actual output before moving on.

Step 5. Add a filter and the follow-up action

After the Gemini step, add a Filter step that only continues when needs_attention is yes. Then add the Slack action that posts the flagged submission and Gemini’s one-sentence reason to the product channel. The filter keeps the Zap quiet: unflagged submissions end the run after the Gemini step, so the team only sees what matters. Test the full Zap with one submission that should be flagged and one that should not, and confirm that only the first one reaches Slack.

Step 6. Check the model tier and the task math, then publish

Before publishing, open the model dropdown one more time and confirm the tier. New AI steps can default to the most expensive tier, so a step left on Premium costs 5 tasks per run when the same job on Flash Lite would cost 1. Then do the monthly math. Take the tasks per run, which is the Gemini step plus every action step after it, and multiply by the expected runs per month:

At 400 runs a month, even the cheapest setup needs more than the Professional plan’s 750 tasks, so either raise the plan, cut the volume with a filter before the Gemini step, or accept that this Zap belongs on a higher tier. Publish only after the math fits your plan. If it does not, move the filter in front of the Gemini step so the model only runs on submissions that pass a cheap keyword or form-field check.

Step 7. Use the Gemini Enterprise connector when the team works from chat

If your organization has a Gemini Enterprise license, the integration runs in the other direction too. An admin opens the Gemini Enterprise admin console, goes to Settings and help, then Manage team, then Connected apps, and toggles Zapier on. If Zapier is not listed yet, the admin clicks Add MCP Server and pastes in the Zapier MCP URL instead. Once it is on, anyone on the team can ask Gemini to pull the last five HubSpot opportunities and send a summary to Slack, or check Gmail for anything missed this morning, all from the chat window. Zapier notes that its connector adds thousands of apps on top of Gemini Enterprise’s 100+ native connectors, and the same governed permissions apply. This route is only available inside Gemini Enterprise, the paid admin-managed edition formerly known as Agentspace. Personal Gemini accounts can still use Gemini inside Zaps, but they cannot use the Zapier MCP connector.

Troubleshooting

The AI by Zapier step is missing or greyed out

The step is only available on paid Zapier plans. On the Free plan you can add the step and preview results with the Standard tier, but you cannot add tools or publish a Zap that contains it. Upgrade to Professional, Team, or Enterprise, or use the dedicated Google AI Studio (Gemini) app with your own API key, which is a separate integration and has its own billing. If you are on a paid plan and the step still does not appear, check that you are editing a Zap and not a Zapier Table or Interface, where AI steps are not available.

Task usage is far higher than expected

Check the model dropdown first. A step that defaulted to Premium costs 5 tasks per run, and every tool call the step makes costs the same rate again, so a Premium step that calls two tools costs 15 tasks per run. Switch to Gemini 3.5 Flash Lite for classification and short summaries, and turn off agentic tool calling when the step only reads its prompt and replies. Also check whether the Zap runs on events that do not need a model at all. Adding a filter before the Gemini step, so the model only runs on submissions matching a keyword or a form value, is usually the biggest saving.

The Gemini step returns one messy block of text

You skipped the output fields. Open the step, define one field per value with a name, a type, and a description, and mark required any field a later step depends on. Retest with a real trigger event and confirm the output arrives as separate fields before you republish. If the output is still unreliable, shorten the prompt and remove instructions that ask for two things at once: a step asked to classify and draft a reply in one pass will do one of them badly.

The model I want is not in the dropdown

Not every Gemini model is built into AI by Zapier. Gemini 4 Argon, for example, is only available as a bring-your-own-key model, which means you connect your own Google account and pay Google directly for its tokens. The other route is the dedicated Google AI Studio (Gemini) app, which connects with an API key from Google AI Studio and supports actions AI by Zapier does not, such as generating images, speech, and video. If a model is listed in Zapier’s blog but not in your editor, check Zapier’s model list again, since new models are added as they become available from Google.

Zapier does not appear in my Gemini connectors

Zapier MCP is only live inside Gemini Enterprise, the paid admin-managed edition. It does not appear in personal Gemini accounts, and there is no setting on the Zapier side that enables it. If your organization has a Gemini Enterprise license and Zapier still does not show in Connected apps, ask an admin to add it manually with Add MCP Server and the Zapier MCP URL, then authenticate through Zapier when prompted.

FAQ

Do I need a Google AI Studio account to use Gemini in Zapier?

No. When you use Gemini through AI by Zapier, the step runs on Zapier’s own connection, so no Google AI Studio account or API key is needed. You only need your own key if you use the dedicated Google AI Studio (Gemini) app, or if you want a bring-your-own-key model such as Gemini 4 Argon.

How many Zapier tasks does a Gemini step cost?

It depends on the model. Gemini 3.5 Flash Lite costs 1 task per run, Gemini 3.5, 3.6, 3.7, and 3.8 Flash cost 3 tasks each, and Gemini 3.1 Pro costs 5 tasks. Gemini 4 Argon is only available with your own API key. Each tool call the step makes costs the same rate again.

Can I use Gemini in Zapier on the Free plan?

No. AI by Zapier is only available on paid plans: Professional, Team, and Enterprise. The Free plan includes 100 tasks a month and does not include AI steps. The Professional plan starts at $19.99 a month billed annually and includes 750 tasks.

What is the difference between AI by Zapier and the Google AI Studio (Gemini) app?

AI by Zapier is a built-in step that runs Gemini on Zapier’s connection and bills in Zapier tasks, with prompt testing, knowledge sources, and agentic tool calling included. The Google AI Studio (Gemini) app connects with your own Google API key and bills through Google, and it supports actions AI by Zapier does not, such as generating images, speech, and video.

Can I run Zapier automations from inside the Gemini chat window?

Yes, but only with Gemini Enterprise, Google’s admin-managed edition formerly known as Agentspace. An admin enables the Zapier connector there, and the team can then trigger automations across Zapier’s 9,000+ apps from the Gemini chat window. Personal Gemini accounts cannot use the Zapier MCP connector yet.

Which Gemini model should I pick for a high-volume Zap?

Start with Gemini 3.5 Flash Lite at 1 task per run for classification, triage, and short summaries. Move up to a 3-task Flash model only when the output quality is visibly worse on Lite, and reserve Gemini 3.1 Pro at 5 tasks for work that needs deeper reasoning. Check Zapier’s AutomationBench if you want benchmark scores for your exact task type.

Conclusion

Gemini inside Zapier is at its best when it does one judgment call inside a workflow that already works, and the bill stays small when the model matches the job. Start with Gemini 3.5 Flash Lite on a single Zap, define output fields before you test, filter before the model when volume is high, and check the tier before you publish. If you are new to AI steps in general, read how to add an AI step to your Zapier workflow first, and if you are coming from the retired Zapier Agents product, our guide to migrating Zapier Agents to AI by Zapier explains what changed. For a broader comparison of automation platforms, the complete n8n guide shows where a self-hosted workflow engine fits instead.

Frequently asked questions