The Four-Prompt Weekly PMax Review

Performance Max rewards the operator who checks weekly and punishes the one who fiddles daily. The campaign type runs on machine learning that needs stable inputs — every panicked change feeds it noise — but it also drifts if ignored, spending into query territory and audiences you never intended. The working cadence is a disciplined weekly review, and because PMax reporting is scattered across insight panels, a connected AI that reads it all in one pass turns that review into four prompts and about twenty minutes.

Why PMax reviews should be weekly, not daily

Daily numbers in PMax are noise wearing a trend costume. Conversion lag makes yesterday look worse than it was; auction-time bidding makes single days swing; the learning system is mid-adjustment more often than not. React daily and you are managing static.

Weekly is where signal accumulates enough to act on, while drift is still cheap to correct. The failure mode on the other side — the monthly glance — lets a bad search category or a decayed asset group compound for weeks. Weekly, timeboxed, prompt-driven: that is the shape that survives contact with a real calendar. What you can and cannot control in this campaign type is its own subject — the Performance Max control guide maps it — but the weekly loop below assumes the honest premise: you steer PMax by inputs and inspection, not by daily levers.

The twenty minutes, budgeted:

Minutes	Check	You are deciding
5	Spend and conversion delta	Is any movement real and material?
5	Search categories	Did the budget wander off-intent?
5	Asset coverage	Which expensive gaps get this month's refresh?
3	Brand leakage	Are the numbers still honest?
2	Notes and thresholds	Does anything clear the intervention bar?

Prompt 1: spend and conversion delta versus prior period

For each Performance Max campaign in account 123-456-7890, compare
the last 7 full days with the 7 before: spend, conversions,
conversion value, cost per conversion. Note conversion lag — flag
which movements might still fill in. Anything moving materially,
give the most likely cause from the data you can read.

Five minutes. You are looking for two things: material movement (against thresholds you set once — a share of weekly budget, not a universal number), and the lag caveat applied honestly so you do not diagnose a dip that is actually just conversions still arriving. Write the thresholds into the saved prompt and the review inherits consistency for free. Most weeks, this prompt returns "within normal range" and you move on — that is the review working, not failing.

Prompt 2: search categories and the queries you never intended

Show search category insights for the same period, ranked by spend.
Which categories are new this week, which converted, and which spent
without converting? Flag any category that does not match what we
sell.

This is where PMax drift shows first. The campaign has wide matching latitude, and the categories report — backed since 2025 by a full search terms report, the same granularity Search campaigns get — is your window into where the budget wandered. A category that is plainly wrong is actionable: tighten the feed and asset language that attracted it, and add campaign-level negative keywords, which Google has now rolled out to every PMax campaign — no rep required. A category that is merely unexpected but converting is a discovery, not a leak — note it, and consider whether it deserves dedicated Search coverage.

What good looks like: most weeks the new categories are adjacent and small, the converting ones are familiar, and this check is thirty seconds of nodding. The week it is not — a plainly foreign category with real spend — you caught drift seven days into it instead of seven weeks, which is the entire return on the cadence.

Prompt 3: asset group coverage and stale creative

For each asset group: which asset slots are unfilled or rated Low,
which assets have not changed in 90+ days, and where is video
missing so Google generates it for me? Rank asset groups by how much
spend flows through them, so I fix the expensive gaps first.

An asset group is the creative surface PMax assembles from, and it decays quietly — the launch-week assets still running in month six, the missing video slot Google fills with an auto-generated one you have never watched. The spend-ranked ordering matters: a threadbare asset group carrying most of the budget is this week's job; the same gaps in a trickle-spend group are next quarter's. One habit worth adding: actually watch the auto-generated video once — it represents your brand in placements you will never browse, and thirty seconds of viewing is the entire cost of knowing what it says. Refresh drafting itself is a separate session — the connected prompt library covers copy workflows.

Prompt 4: brand traffic leakage check

Estimate how much of this PMax campaign's performance is brand:
compare its conversion volume and efficiency against my brand Search
campaigns, check whether brand terms appear in the top search
categories, and confirm brand exclusion lists are in place and
attached.

PMax looks brilliant when it quietly absorbs brand traffic that would have converted anyway. The tell is efficiency too good against every other campaign, brand-flavoured categories in the insights, or a brand Search campaign losing volume in step with PMax gains. The fix is structural — brand exclusions attached and maintained — and worth confirming weekly precisely because it makes every other number in the review honest. Uncaught, brand leakage inflates PMax and misprices every budget decision you make downstream.

What stays out of the weekly loop

The review works because it is small, so guard its edges. Feed health, audience signals, goal changes, asset group restructures and campaign-level experiments belong in a monthly session with more runway — they are decisions with learning consequences, not observations. The weekly loop observes, notes and occasionally intervenes; the monthly session is where structure changes. Mixing the two produces the daily-fiddling failure mode wearing a weekly costume.

When to actually intervene — and what to leave alone

The review produces notes; intervention needs a bar. A workable set of thresholds, agreed once:

Signal	Intervene when	Otherwise
Cost per conversion drift	Sustained above target for 2-3 weeks	Note and watch
Wrong search categories	Plainly outside what you sell, with spend	Note if small
Asset gaps	High-spend group with Low-rated or missing assets	Batch for monthly refresh
Brand leakage	Exclusions missing or brand terms in categories	Confirmed clean — move on

Leave alone: single bad days, asset groups mid-learning, and anything you changed less than two weeks ago — changes need runway before they can be judged. The discipline of the twenty-minute review is mostly the discipline of not acting: four prompts, honest thresholds, and the patience to let a learning system learn. Intervene rarely, on evidence, and PMax repays the restraint. The prompts do the reading; the thresholds do the restraining; the twenty minutes is what remains.