Connect Google Ads to ChatGPT and ask better campaign questions

Start with the project and accounts

In Adport, select the project for the brand or client. Connect Google, save the specific accounts to include, and enable agent access. Use a recognizable project name so the authorization screen is easy to check.

Choose the business question before connecting the assistant. Examples include understanding a change in acquisition cost, finding a campaign that needs a closer review, or preparing a weekly client summary. Start read-only while you verify the data. A model can summarize an account without needing permission to change its campaigns.

Add the custom MCP plugin in ChatGPT

In ChatGPT's Plugins area, use the add control to create a custom MCP connection. Name it Adport and supply:

https://app.adport.dev/mcp

Select OAuth, complete the Adport sign-in, and review the project being authorized. Finish creating the plugin, install it in the appropriate personal or workspace context, and select it in your conversation. OpenAI publishes the current custom MCP setup and authentication options.

Workspace administrators can limit which plugins are available. If the controls differ from this guide, use the current documentation and your workspace settings rather than entering advertising credentials into a chat. You do not need to give ChatGPT your Google password or an Adport platform secret.

Verify the connection with a small read

Confirm a known account before asking for a large analysis. If the account is missing, check the Adport project and enabled account selection. Reauthorizing an agent for one client should be a deliberate choice, not a side effect of changing a dashboard tab.

For the next question, name the account, reporting period, and metrics. Ask ChatGPT to cite the account and campaign rows it used. You should be able to trace a recommendation back to a specific result, rather than an unsupported general best practice.

Use prompts that lead to a decision

Ask follow-up questions about conversion definitions and timing. A purchase value, a lead value, and a sign-up count are different inputs. Google explains how assigned conversion values influence reporting and optimization.

A useful client summary ends with decisions: what changed, what you checked, what you recommend, and what still needs evidence. Avoid asking for a single performance score. A score can conceal missing data, a change in conversion quality, or a result that matters differently to the business.

Keep write requests separate from analysis

When you want a supported change, ask for a preview of one exact operation. Check the account, campaign, old and new values, currency, and any policy adjustments. Read the preview before you authorize the applying call.

Adport's write gate is separate from ChatGPT's own permission UI. The matching pending operation and arguments are still required. An expired preview or a revised budget needs a new preview. A viewer or Free cloud plan remains read-only even if the assistant can discover the write tool.

After applying an approved change, ask for a fresh read and check the audit record. This closes the loop between your decision and the platform's actual state.

Keep the workflow useful over time

Reuse a short briefing with the account IDs, primary conversion, currency, normal reporting period, and current hypothesis. Update it when the business changes its offer or measurement. Clear context usually improves the review more than adding another paragraph of generic optimization instructions.

For team permissions, use the read-only client access guide. If you prefer a terminal workflow, the Codex guide uses the same Adport tools and project boundaries.