Most accounts get looked at when something goes wrong. That is three weeks too late: a broken conversion tag, a budget quietly capped, a new query type eating a fifth of spend — none of them announce themselves. A weekly review is the cheapest insurance in paid search, and with an assistant that can read the account it takes about ten minutes.
This is not an audit. The 30-minute audit is what you run on an account you inherited or have not opened in months. The weekly review is what you run on an account you already understand: six numbers, week over week, and a decision about each one.
What a weekly review actually checks
Six things move fast enough to be worth a weekly look. Everything else is a monthly or quarterly job.
The first row is the one that pays for the habit. A conversion tag that stopped firing looks exactly like a bad week, and every automated bidding strategy in the account will spend a fortnight optimising towards nothing before anyone notices.
The one prompt
If your assistant is connected to the account, the whole review is one message. This is the version we run.
Run my weekly Google Ads review. Compare the last 7 days with the 7 days before, for the whole account and for each active campaign, and report: 1. Conversions and conversion value, and flag any campaign that recorded conversions last week and zero this week. 2. Spend, and whether the account is on pace for its monthly budget. 3. Search terms that spent more than $20 with zero conversions in the last 30 days. 4. Cost per acquisition this week against the 4-week average. 5. Any campaign showing as limited by budget, with its current daily budget and its conversion rate. 6. Any disapproved ads, assets, or extensions. For each finding, tell me the number, the change, and whether it needs action this week or can wait. Rank your recommendations by how much money each one is worth. Do not change anything.
Two details in there earn their space. "Do not change anything" keeps the review a review — approvals happen in a second message, deliberately. And "whether it needs action this week or can wait" is what turns a wall of numbers into a short list; without it you get a report, and reports do not get read twice.
Set the twenty in line three to your own threshold. It should be roughly what you would pay for a customer — generous at an $8 cost per acquisition, meaningless at $300.
Put it on a schedule
The review only works if it actually happens, which is an argument for taking yourself out of the loop.
In ChatGPT, save the prompt as a scheduled Task. Open the conversation, ask for it to run every Monday morning, and the answer is waiting when you start the week. The task runs against live account data, not a cached copy of last week's answer.
In Claude, there is no scheduler. The prompt still works exactly the same way; the cadence has to be a habit, or a calendar reminder that tells you to paste it.
Either way the assistant needs access before it can answer anything. Connecting it takes a few clicks — ChatGPT or Claude.
Reading the answer
The report is not the work. Deciding is.
Act this week. Zero conversions where there were conversions. Disapprovals. A profitable campaign capped by budget. Search-term waste above the threshold. These are either broken or free money, and neither improves by waiting.
Watch one more week. A CPA up 20 percent on a normal volume of conversions. A single campaign softening while the account holds. Weekly numbers are noisy, and reacting to noise is how accounts get churned into the ground.
Do nothing. A change of a few percent on any metric. Seasonality you already know about. A week with a public holiday in it.
The mistake is not missing a problem. It is treating every wobble as one and changing four things at once, after which nothing in the account can be attributed to anything.
Approving the fixes
When something needs doing, it is a second message and it names one job.
Take the two items you ranked highest. Show me exactly what you would change — the campaign, the setting, the current value and the new value — and what you expect it to do. Wait for my confirmation before applying anything.
One job at a time, with the before and after visible, is the difference between an assistant and an autopilot. Adding negative keywords deserves particular care in this step: it is the only change that removes traffic silently, so read the proposed list rather than skimming it.
What does not belong in a weekly review
Some jobs need a longer window and get worse when you look at them every seven days.
- Bid strategy changes. A smart bidding strategy needs two to four weeks to learn. Changing target CPA weekly means it never finishes learning.
- Ad copy tests. Statistical significance on a responsive search ad takes longer than a week on most budgets.
- Structural rebuilds. Splitting campaigns or rewriting ad groups is a monthly decision made with a month of data.
- Landing page work. Not a Google Ads job at all, and the weekly numbers cannot tell you much about it.
Keep those on a monthly cadence. The weekly review's job is to catch things that are broken and things that are free, and then stop.
A ten-minute Monday
Paste the prompt, or read the task that already ran. Skim the six rows. Act on the ones that need acting on, ignore the rest, close the tab.
That is the whole routine. It catches a dead conversion tag in week one rather than in the monthly report, and over a year that single habit is usually worth more than any optimisation in it. If you want more prompts in the same shape, the prompt library has fifteen you can paste today.