One Claude agent audits your Meta and Google accounts, finds waste, writes briefs, plans tests and reports weekly, and nothing goes live until you approve it.

<aside> 🚀

🚀 FREE STRATEGY CALL Want to grow your product through viral reach? We help teams across iGaming, mobile apps, AI products, and SaaS build creator-led organic distribution generating tens of millions of views at CPMs starting from $0.03.

<aside> 🚀

🚀 Book a free strategy call

</aside>

On the call you'll receive:

Contents

In the post Count Where it lives on this page
A Meta + Google Ads agent in Claude 1 The agent brief, in the section on the one question
One question: What should I test next? 1 The one question section, with the agent brief
It reads both accounts 2 Setup: a route for Meta, a route for Google, and spreadsheet exports for both
It finds the wasted spend 1 Find waste
It writes the creative briefs 1 Write briefs
It plans the next test 1 Plan tests
Five jobs: Audit accounts, Find waste, Write briefs, Plan tests, Weekly report 5 Five sections, in the post's order
The setup 1 Setup
The prompts 6 The agent brief and one prompt for each of the five jobs
The permission settings 2 Permission settings: one table of what the agent may do, one table of settings
Nothing goes live until you approve it 1 Nothing goes live until you approve it
A worked example, from the brief for this page 1 Northlight, from one question to a finished test plan

Jump to: Start here. The one question. Audit accounts. Find waste. Write briefs. Plan tests. Weekly report. Setup. Permission settings. Nothing goes live until you approve it. Northlight. What is not verified. How to use it every week. How this was made.

Start here

  1. Export the last 28 days of campaign data from Meta and from Google. Use the same dates for both.
  2. Export store orders and store revenue for the same 28 days.
  3. Open a new Claude chat. Paste the agent brief from the next section.
  4. Paste the three exports and your gross margin under the last line of the brief.
  5. Type: What should I test next? Read the first table before you read anything else.

<aside>

Built for. E-commerce media buyers and marketers who run paid ads on Meta and Google and need one clear next test.

Use it when. You have at least four weeks of data, a gross margin figure and a free budget for one or two tests.

How to run the prompts. Open Claude, or any AI chat tool. Copy one block. Paste your own details under the last line, which reads "Paste your inputs below this line:". Read the answer before you act on it.

Before you use any of it. Before using these prompts you should omit any PII (personal identifiable information) and check with your compliance team and/or your AI policy. Ad exports need no customer names or emails. Remove any that appear.

What this agent is. An adviser for paid ads. It never replaces paid. It never changes an account. Organic content sits beside paid and lowers the blended cost of each order.

Units. Every money figure in a worked example is in units. One unit is one pound, dollar or euro. Use your own currency.

</aside>

The one question: What should I test next?

The agent is one brief. You paste it once at the top of a Claude chat. After that, one question starts the chain.

The question runs four jobs in order: Audit accounts, Find waste, Write briefs and Plan tests. The fifth job, Weekly report, runs when you ask for it. The agent stops at each gate and waits for you.

The agent brief

Ask once. Approve every step. Those are the two rules of this agent.