How to Run Google Ads From ChatGPT

What actually connects ChatGPT to Google Ads

ChatGPT can't reach into Google Ads on its own, there is no built-in integration, and pasting your API credentials into a chat would be reckless. The bridge is the Model Context Protocol (MCP), an open standard that lets an assistant call external tools through a server it trusts. A Google Ads MCP server exposes concrete actions, create a campaign, add keywords, adjust a budget, pull a search-terms report, as callable tools with typed inputs.

When you add that server to ChatGPT as a connector, the model can invoke those tools against your real account. Authentication happens between the server and Google, using OAuth, ChatGPT never sees your Google password, and the server never sees anything of your chat beyond the specific tool call. That separation is the whole point: the model proposes, the server (and you) dispose. VigilDog's Ads MCP is built on exactly this model, and the same server drives Google, Meta, and LinkedIn.

Setting up the connection

Setup is a one-time job and takes a few minutes. You point ChatGPT at the MCP server, then authorize the Google Ads account you want it to manage. From then on, the account is available in any conversation.

• Open ChatGPT's connector settings and add a custom connector using the Ads MCP server URL (developer mode / connectors, depending on your plan).
• Approve the connection. ChatGPT will show you the tools the server exposes so you can see its scope before anything runs.
• Trigger the Google Ads OAuth flow. You sign in to Google directly and grant access to the specific accounts you choose, nothing broader.
• Confirm the account is linked by asking for something read-only first, like "list my Google Ads campaigns and their status."

Dry-run first, then approve

The difference between a helpful assistant and a dangerous one is what happens the moment you say "raise the budget on the brand campaign." A good Ads MCP never applies that silently. It returns a preview, the exact change, the campaign it touches, the before-and-after value, and waits. Only when you confirm does the write actually reach Google's API.

This preview-and-confirm gate is what makes running ads from a chat window safe rather than nerve-wracking. You get the speed of natural language with the deliberateness of a human clicking "save." If the model misreads you, you catch it in the preview, not in tomorrow's spend.

What you can actually do from chat

Reporting is where most people start, because it is read-only and instantly useful. But the write actions are where the time savings compound, routine account hygiene that normally means a dozen clicks becomes one sentence.

• Pull performance: "Show me last 30 days by campaign, sorted by cost, with conversions and CPA."
• Mine search terms and add negatives: "List search terms with spend over $50 and no conversions, then propose negatives."
• Build campaigns: describe the structure in plain English and review the generated campaign, ad groups, and RSAs before anything publishes.
• Adjust bids and budgets, pause underperformers, and apply labels, each as a previewed change you approve.
• Audit an account for wasted spend, missing conversion tracking, or thin ad strength before a client call.

Guardrails worth keeping

Two honest cautions. First, an assistant is only as good as your review of its previews, treat every confirmation like signing off on a change a junior would make. Read the numbers, not just the summary. The gate protects you from accidents, not from approving a bad idea quickly.

Second, keep the connection scoped. Authorize only the accounts you intend to manage from chat, and use accounts with the permissions the work needs, no more. The workflow is identical in Claude if you prefer that assistant; the run Google Ads from Claude walkthrough covers the same preview-gated flow, and pricing for account volume lives on the plans page.