The 20-Minute Weekly Google Ads AI Routine

The weekly routine is five prompts, run every Monday, in twenty minutes: numbers, search terms, budgets, ad health, one improvement. The agent pulls every number and argues every case in seconds; you stay in control — approving, and skipping. This is the maintenance layer that sits on top of everything Google already automates, covering the judgement calls no Smart Bidding model makes.

The routine assumes you have run a first audit and cleared the backlog; a weekly pass maintains a clean account, it does not clean a messy one. One conversation, all five prompts, in order.

The routine at a glance

Slot	Prompt	Minutes	Ends with
1	Week-over-week summary	4	Understanding
2	Search-term sweep	5	Approved negatives
3	Budget pacing check	4	At most one budget proposal
4	Ad and asset health	3	Fix list
5	One improvement	4	One approved or parked change

Prompt 1 — the week in numbers

In account <account>: last 7 days vs the 7 before — spend, conversions,
cost per conversion, by campaign. Three sentences on what moved and the
most likely cause, checking change history before you speculate.

Read it, don't act on it. The summary exists to load the week's context into your head — and the agent's conversation — before anything is decided. The "check change history" clause is what separates a cause from a guess: last Monday's approved budget change should be the first suspect for this Monday's movement.

Approve: nothing here — it is a read. Skip: any urge to react to a single week's cost-per-conversion wobble on low volume.

Prompt 2 — the search-term sweep

Pull search terms from the last 7 days with cost and no conversions.
Classify them, recommend negatives with match types, show me the list —
then add the ones I confirm to the campaigns where they appeared.

The compressed version of the full sweep: classify, judge, one add_negative_keywords approval per campaign.

Approve: clearly irrelevant terms, on sight — negatives can only stop spend. Skip: "plausibly fine" terms and the competitor-blocking debate; the first needs time, the second is a quarterly decision, and neither belongs in a Monday slot.

Prompt 3 — budget pacing

For each enabled campaign: daily budget, average daily spend last 14
days, lost impression share to budget, cost per conversion. Flag
campaigns that are capped AND converting well, or uncapped AND wasting.
If one budget change is clearly justified, propose it with the evidence.

The two flag conditions are the only budget situations that deserve weekly attention: a winner being throttled, or a loser left running.

Approve: at most one budget change per week, and only when the evidence column convinces you independently. Skip: simultaneous changes to several budgets — batch spend changes destroy your ability to read next week's Prompt 1, and the one-change cap is the whole reason this routine stays trustworthy. The reasoning is the same as in the approval workflow: spend changes are gated because they are the only expensive mistakes.

Prompt 4 — ad and asset health

Any ads disapproved or limited by policy this week? Any ad group left
with zero or one enabled RSA? Any campaign missing sitelinks or
callouts? Short list, worst first.

A three-minute smoke test for silent breakage. Disapprovals are the urgent class — a disapproved ad is blocked distribution wearing an "enabled" label.

Approve: asset fixes on the spot — new sitelinks or callouts drafted, reviewed, created. Skip: rewriting healthy ads because you are bored of them; creative changes belong in a deliberate session with a hypothesis, not a health check.

Prompt 5 — one improvement

Based on everything this session, propose the single highest-value
change to account <account> this week. State the evidence, the exact
change, the risk, and how we'll know in 14 days whether it worked.
Wait for my decision.

The open slot is what makes the routine an optimisation practice rather than pure maintenance. Some weeks it proposes a device bid adjustment with a table behind it; some weeks a keyword graduation from the search terms; some weeks it says "nothing clears the bar this week" — which is a correct answer and worth praising.

Approve: the proposal if — and only if — the "how we'll know" criterion is concrete. Skip: proposals whose evidence is a recommendation rather than a measurement. The agent argues from your account's data, and the standard to hold it to is the one from the audit method: findings, not vibes.

Close the session

Log this session: everything changed, everything proposed and declined,
and the one thing to watch by next Monday.

Thirty seconds, and next week's Prompt 1 has a written baseline to check against. Over months, these logs become the account's management history in plain language — every change explained at the moment it was made, which is a record most manually-run accounts never have.

Start this Monday

Set a recurring twenty-minute slot, paste the five prompts into a note, and run the routine once — this week's session will be slower than twenty minutes and that is fine; the second one will not be. If nothing is connected yet, Start a free pilot and you can run the full routine twice inside the free week — enough to know whether Monday-by-prompt beats Monday-by-dashboard for you.