Google Ads with ChatGPT: eight prompts and safe edits

On its own ChatGPT cannot see your ad account; an MCP connector gives it that ability. Tecrube's hosted server opens your Google Ads account to the chat through a single URL added in ChatGPT's connector settings. Reads answer instantly; writes return a preview and wait for your approval.

Setup is three steps: add a connector in ChatGPT settings and paste the URL from your Tecrube member panel (/app); on the first tool call choose your account on Google's OAuth screen; start chatting. Your password is never shared anywhere.

“Tabulate campaign performance for the last 30 days with spend, clicks, conversions and cost per conversion, compared with the previous 30 days.” The data comes from the live account and ChatGPT adds commentary: which campaign got more expensive, which got cheaper.

“List search terms from the last 14 days that spent over $20 without converting and group them by pattern.” In most accounts this prompt reveals savings on day one. Follow up with “add these as negatives” and a preview arrives.

“List keywords with a quality score below 5, including the ad relevance and landing page experience components.” A low score usually comes from a disconnect between ad copy and keyword; ChatGPT writes the likely cause for each row.

“Pull 30 new keyword ideas from Keyword Planner for this ad group, with volume and bid range.” The suggestions come from Google's planner, not from the model's memory; you then group them by intent and choose.

“Raise budgets by 20 percent on campaigns with below-average cost per conversion and lower them by 20 percent on the rest; preview it.” Old and new budgets appear per campaign; single-step increases above fifty percent are rejected anyway. You approve, it is applied.

“Who changed what in the account over the last 7 days?” When working with a team or agency, this prompt lets you tie a performance shift to an operation. Tecrube's own approval log is also available in the panel.

“Prepare a preview of a paused search campaign for this product with a $40 daily budget, targeting Turkey, three ad groups and two RSAs each.” The structure arrives in one preview; you review and approve, the campaign is created paused, and enabling it is a separate approval.

“Audit my account: conversion tracking, disapproved ads, campaigns with budget but no impressions, missing extensions.” The audit output is a to-do list, and every item can be resolved in the same chat.

First, state a date range and a threshold: not “expensive terms” but “non-converting terms that spent over $20 in the last 14 days”. A vague prompt returns a vague table. Second, ask for the output format: a table, the sort criterion, the columns. ChatGPT applies it and keeps the same format in the next prompt.

Third, use the word “preview” explicitly in write requests. Tecrube returns every write as a preview anyway, but phrasing the prompt that way strengthens your own habit of reading it. Save the eight prompts in ChatGPT's custom instructions or a project and you can call each with a single word every week.

Writes follow preview → approve → apply, and the budget cap and activity log apply to every prompt. For Google Ads management, the difference between ChatGPT and Claude is not in the connector but in the assistant's style; the same Tecrube URL works in both. Use whichever you think better in; part of the team can use ChatGPT and part Claude.

For a trial, sign-up includes 75 free credits that never expire. After that Pro is $49 a month for one account, Agency $99 a month for several, and Enterprise starts at $299.