Google Ads with Claude: a one-week working routine

Setup guides usually end with “you are connected, congratulations”. The real question comes next: what does managing an ad account with Claude look like day to day? This article proposes a concrete one-week routine. Tecrube's connector URL is added to Claude once; after that the same chat structure is used every day.

Claude's Projects feature is useful here: write your account's context (industry, target cost per conversion, brand rules, campaigns that must not be touched) into the project instructions. Every new chat starts with that knowledge and you stop repeating yourself.

Prompt: “Yesterday's summary: spend, conversions, cost per conversion; flag campaigns deviating more than 25 percent from the 7-day average.” Claude reads the live data, lists the deviations and writes likely causes: budget-capped, disapproved ad, competitor pressure.

If nothing deviates, the chat ends there. If something does, you dig in within the same conversation: “campaign X's ten highest-spending search terms yesterday”. The morning routine is read-heavy; writes are rarely needed.

One day a week belongs to search terms. “Group terms from the last 7 days that spent above the account's average cost per conversion without converting, and suggest which of my shared negative lists each goes to.” The preview arrives, you read the rows, you approve.

Claude's strength is reasoning: it explains why each term counts as waste. You remove the rows you disagree with; thanks to the preview → approve → apply sequence, a wrong negative needs your approval before it can enter the account.

“List ad groups with low click-through rate; for each, show the current RSA headlines and suggest three new ones.” Claude is comfortable with copy, and the suggested headlines follow the brand rules you put in the project instructions. Adding a new RSA is a write, so it comes with a preview.

On the quality score side, asking for “keywords whose landing page experience is below average” shows that the work is on the page, not the ad. If your WordPress site is connected to Tecrube, the page edit can be made from the same chat with a preview and approval.

In the last week of the month: “Against my monthly budget target, which campaigns under- or over-spent; preview suggested daily budgets for the remaining days.” Old and new values appear per campaign. Because single-step increases above fifty percent are rejected, large corrections happen in two steps.

Every approved change is written to the activity log; next month the answer to “what did we change at month-end?” is ready.

Say the morning report shows a campaign's spend doubled overnight. The first reflex is to pause it; ask why first instead: “show campaign X's spend yesterday by hour and by search term; what happened in the change history yesterday?” Most of the time the answer is a budget or match type change, and it shows what needs reverting.

If the cause is unclear, lowering the daily budget as a temporary measure does less damage than pausing the campaign outright, because learning data is preserved. In Tecrube that is a preview too: old budget, new budget, approval. You look again in the evening and make the permanent decision once the cause is clear.

In moments like these the activity log is the most valuable source. When “who changed what yesterday?” is answered in minutes, troubleshooting drops from an hour to a minute.

Save the four prompts as templates in the project instructions, so each day you only type “morning report”. Because the same Tecrube URL also works in Claude Code, the morning report can be triggered by a command as well; writes still wait for approval. The 75 free credits are enough to try this routine for a week; after that Pro is $49 a month.