Google Ads in ChatGPT

Google Ads is connected once in Adrails, under Ad accounts, when Google has finished verifying Adrails; a workspace already connected keeps its accounts. Adrails reads campaigns with their type, status, budget and bidding, ad groups with their keywords, ads with their copy and final URLs, the account's picture library and audiences, and Keyword Planner volumes and ideas, and ChatGPT reaches the same data through the MCP server, with your sign-in and inside the one workspace you chose.

In ChatGPT, the Google Ads reads come from 7 tools: reporting_overview, performance_period, campaigns_structure, google_ads_library, google_keyword_planner, targeting_options and findings_review. The table above says what each one returns and gives a question to type as it is. Every result comes back twice, as exact data for the model and as a readable table for you, so a figure ChatGPT states can be traced to the call that produced it.

ChatGPT adds remote MCP servers through Developer mode, which depends on your ChatGPT plan and your workspace's policy. The server is added once, then enabled in the conversations where you want it, and ChatGPT's app settings let you switch individual tools off and refresh the list from the server.

Every client uses the same server URL, https://adrails.ai/api/mcp, and the same sign-in: ChatGPT registers itself, you sign in to Adrails, then choose the workspace and the permissions on the consent screen. Google Ads needs nothing on the ChatGPT side, since it is already connected in that workspace under Ad accounts.

• 1. In ChatGPT Settings, turn on Developer mode
• 2. Open Plugins and add a custom MCP server named Adrails, with https://adrails.ai/api/mcp as its URL and OAuth as its authentication
• 3. Sign in to Adrails, choose the workspace and the permissions, and choose Allow access
• 4. Review the tools ChatGPT discovered, then add Adrails to a new conversation

Ask in your own words: ChatGPT picks the tool, Adrails returns the figures, and ChatGPT's model does the reasoning, so a tool call uses no agent credit. Each result carries the exact data in structuredContent.result and the same figures as readable tables.

A question can be narrowed to some of the workspace's ad accounts with accountIds, never widened to accounts outside it. A result is at most 256,000 bytes, so a long period at the ad level is asked for in shorter ranges or one campaign at a time.

• What did the Search campaigns spend last month, and at what CPA?
• What are the monthly searches and top-of-page bids for these keywords in Germany?
• Prepare a Search campaign for our new product page with three ad groups, paused.

No tool creates, edits, pauses, switches on or rebudgets a campaign, ad set or ad by itself. In ChatGPT, google_ads_build creates a Search, Display, Demand Gen or Performance Max campaign with its ad groups, keywords and ads or its asset groups, or new ad groups and ads under what exists, all paused after Google validates the request; google_ads_edit changes ad texts, final URLs, keywords, names, campaign dates, target CPA or ROAS, extensions, places, languages, search themes and audience signals, validated by Google when prepared and again when applied, all or nothing; campaigns_change_budget raises or lowers a campaign budget by at most 30% in one change, and refuses a budget shared between campaigns.

Then campaigns_set_status pauses or resumes one campaign, ad group or ad and measures the result after 24 hours, 3 days and 7 days; campaigns_duplicate_to_draft duplicates a campaign paused, with its targeting, bidding, budget, extensions, keywords and ads. Each of them returns a proposal: a preview of the exact change and a link to the Review advertising change page in Adrails, where you, signed in as an owner or admin of the workspace, apply it or decline it. Adrails reads Google Ads again first and applies the change once, or blocks it if the entity moved in the meantime.

A launch is built paused, switching it on is a second proposal that lists every budget that starts spending, an applied change can be undone while the platform still matches what Adrails wrote, and a proposal expires 30 minutes after it is prepared. ChatGPT asks you to confirm a write action before it runs; that confirmation is ChatGPT's own setting, and the advertising change is still applied from its page in Adrails, never from the confirmation.

Read workspace data is required and covers every ad account of the workspace; Prepare advertising changes is a separate permission that only an owner or admin can grant, and a permission never goes beyond your role. A member who connects ChatGPT reads Google Ads but sees no write tool.

Structure, findings and overviews come from the account's last synchronization, and performance_period reads an exact date range from Google Ads. accounts_list shows each account's last snapshot, and an owner or admin who granted Manage integrations can queue a refresh with meta_sync. Google Ads opens once Google finishes verifying Adrails. Connect Meta for now.

Calls on one connection run one at a time, at most 60 a minute per access token, and a connection works in the one workspace you chose at consent: to work in another, connect ChatGPT again and choose it. Revoke ChatGPT under Your connections on the MCP page in Adrails; removing the server in ChatGPT does not revoke it there.