How to Connect Google Ads to ChatGPT

What "connecting Google Ads to ChatGPT" really means

ChatGPT can't talk to the Google Ads API on its own, it needs a bridge that exposes Ads operations as tools the model is allowed to call. That bridge is an MCP (Model Context Protocol) server. You add it to ChatGPT as a connector; from then on, when you ask about your account, ChatGPT calls the connector's tools, the connector calls Google Ads with your authorization, and the results come back into the chat.

So you're not pasting an API key into a prompt or exporting spreadsheets. You're adding a governed connector that speaks Ads on your behalf. The diagram below shows the path a request takes from your message to your account and back.

What you'll need first

The setup is short, but a couple of prerequisites save you a stalled afternoon:

• A ChatGPT plan that supports custom connectors / MCP (this lives in the connector or developer settings, and availability depends on your plan tier).
• A VigilDog account with the Ads MCP enabled, which handles the Google Ads authorization for you.
• Access to the Google Ads account(s) you want to manage, you'll authorize them once during connection.
• A clear idea of read-only vs. change access: it's fine (and smart) to start read-only while you learn what it can do.

Adding the connector in ChatGPT

The exact labels shift as ChatGPT's UI evolves, but the flow is consistent. You point ChatGPT at the VigilDog Ads MCP endpoint, authorize, and connect your Google Ads accounts.

• In ChatGPT, open Settings → Connectors (or Developer/MCP settings, depending on your plan) and choose to add a connector.
• Enter the VigilDog Ads MCP server URL from your VigilDog dashboard.
• Complete the OAuth prompt, this links the connector to your VigilDog account and, through it, to Google Ads.
• Select which Google Ads accounts the connector may access.
• Start a new chat and confirm the connection with a harmless read, e.g. "List my Google Ads campaigns and their status."

Dry-run and approval gates: why this isn't scary

The reasonable fear with letting an AI touch a live ad account is that it confidently does the wrong thing at scale. The VigilDog Ads MCP is built around that fear. Write operations run as a dry run first: ChatGPT shows you exactly what it intends to change, the campaign, the field, the old value, the new value, and nothing is sent to Google until you explicitly approve.

That means the model can propose "pause these three underperforming ad groups" or "raise this budget to $80/day," but the actual mutation waits behind an approval step you control. You get the speed of natural-language control with a hard stop before anything real happens, the same guardrails whether you're in ChatGPT or Claude.

What you can actually ask it

Once connected, the useful range is wider than most people expect. Reads are immediate and safe; changes go through the approval gate.

• "Show me search terms from the last 14 days that spent over $50 with zero conversions."
• "Which campaigns are limited by budget right now?"
• "Draft three new responsive search ad headlines for the brand campaign" (drafts, not published).
• "Add these five terms as negative keywords at the campaign level" (proposed, then approved).
• "Compare this month's CPA to last month's by campaign."

Where to go from here

Google Ads is the starting point, not the ceiling, the same VigilDog Ads MCP runs Meta and LinkedIn accounts through the identical dry-run-and-approve model, so one connection pattern covers your whole paid stack. If you also work in Claude, the mechanics are nearly identical; our run Google Ads from Claude guide walks through that side.

Start read-only, get a feel for how it answers, then enable changes once you trust the approval flow. When you're ready to connect live accounts, pricing and plans lay out what's included at each tier.