The Google Ads audit checklist for high-spend accounts.

The reason most Google Ads audits are useless is not that the checks are wrong. It is that no check has a number attached, so every observation survives contact with judgment and the output is a list of things that could be better — which describes every account ever built. An audit is only worth reading if its items can come back clean, and an item can only come back clean if it has a threshold.

So this is the same checklist we run on accounts spending $50,000 a month and up, with the numbers put back in. Each check states what makes it a finding rather than an observation. Underneath most of them sits a significance floor, because on a high-spend account there are thousands of entities available to generate a false positive, and an audit that reports variance as waste is worse than no audit — it spends real management attention on noise and burns credibility you will need for the real findings.

The floor underneath every waste check

Before the checklist, the single piece of arithmetic that decides whether your findings are real. The classic waste check is "this search term, product or placement has spent money and produced no conversions." The question nobody asks is how much spend without a conversion is actually surprising.

If a term converts at your account average, the expected number of conversions after spending one target CPA is exactly one. Conversions arrive as a count process, so the probability of seeing none is not small:

Spend, in multiples     Expected      Chance of zero conversions
of target CPA          conversions   on a perfectly healthy term

1x                     1.0           37%
2x                     2.0           14%
3x                     3.0            5%
4x                     4.0            2%

A term that has spent exactly one target CPA with nothing to show for it is not a finding. It is a coin flip you would lose more than a third of the time on a term with perfectly average performance. Flag those and you will spend your audit killing healthy terms and calling it optimization.

Two target CPAs is the usual working threshold — you are wrong roughly one time in seven, which is tolerable when the action is a negative rather than a deletion. Three is the right bar when the action is irreversible or politically expensive. The point is that the threshold is a choice about how many good terms you are willing to kill, and picking it deliberately is what separates an audit from a purge.

1. Measurement and conversion integrity

• →Reconcile platform-reported conversions against back-end truth for the same window. Finding if the gap exceeds roughly 5%. This is the highest-value ten minutes in the entire audit and it cannot be done inside the ad platform, which is precisely why it is almost never done. Everything below is graded in a currency this check validates.
• →Confirm every conversion action is deliberately primary or secondary. Finding if any secondary-quality action (newsletter signup, page view, add-to-cart on a purchase account) is set as primary. One mis-flagged action redirects the whole account, and the account will look like it is performing while doing it.
• →Check for double-counting between page and server events, or across GA4 and Ads without dedup. Finding if reported conversions exceed back-end count in the same direction every month — a consistent overcount is structural, an inconsistent one is attribution lag.
• →Verify conversion values are real rather than a static placeholder. Finding if value variance across conversions is near zero on a business with variable order values. Value-based bidding on a flat value is conversion-count bidding wearing a costume.
• →For lead gen: confirm offline conversion import is live and the imported event is a qualified stage rather than a form fill. Finding if the account is optimizing to raw form fills, because at scale the machine will find you an unlimited supply of cheap ones.

2. Brand and non-brand separation

• →Confirm brand is isolated into its own campaigns and reported separately. Finding if any campaign carrying non-brand budget also serves brand queries — the blend makes every other number in the account uninterpretable, which is the argument in brand versus non-brand.
• →Check whether PMax or Shopping is absorbing brand queries with no brand exclusion applied. Finding if brand terms appear in the search-term view of a campaign whose reported performance is being used to justify budget. This is the single largest source of overstated performance at scale.
• →Quantify the brand share of reported conversions. Finding if nobody in the business knows this number. It is not a threshold so much as a test of whether the account is being read honestly at all.

3. Structure and budget concentration

• →Check conversion volume per campaign per month. Finding below roughly 30 conversions, because under that the bidding model is learning from a sample too small to generalize from and the campaign is a rounding error with a name. The fix is consolidation, not a target change.
• →Identify overlap where multiple campaigns are eligible for the same queries. Finding if the same query appears in the search-term view of two campaigns with materially different targets — you are bidding against yourself and paying the spread.
• →Check that spend concentration matches intent. Pareto-shaped spend is normal and not a finding. Finding if upper-funnel prospecting is funded ahead of proven capture, or if a single unsegmented group carries the majority of spend.
• →Confirm the structure reflects business economics — margin band, market, objective — rather than legacy folders. Finding if two products with materially different margins sit in the same bid group under one target, which is the mixed-margin error in the Shopping guide applied to Search.

4. Bidding and target hygiene

• →Check when each target was last derived from unit economics rather than adjusted by feel. Finding if any target is older than the last material change in margins, pricing or shipping costs. A target set eighteen months ago is an artifact, not a decision.
• →Read the change history for target whipsaw. Finding if targets moved more than once inside a single learning cycle, or by more than roughly 15% in a step. Each reset costs a learning period and the account never reaches a steady state to be judged from.
• →Examine performance by spend tier rather than the account average. Finding if the marginal tranche converts materially worse than target while the average sits inside it — see the arithmetic in how to scale without breaking it, where an average stays green through three value-destroying steps.
• →Confirm changes move in disciplined steps: roughly 20% on budget, 10-15% on target, one per cycle, never both in the same week. Finding if the history shows leaps, because you then cannot attribute any subsequent movement to anything.

5. Performance Max control

• →Brand exclusions applied at campaign level. Finding if absent on any PMax campaign whose reported ROAS is being used for a budget decision.
• →Asset groups mapped to genuine product or margin segments with real creative variety. Finding if one catch-all group carries most of the spend, which reduces PMax to a black box with no levers.
• →Channel and search-term insight reviewed for where money actually lands. Finding if the Display share is high and nobody has checked whether it is incremental — full treatment in controlling PMax.
• →New-customer goals or value rules in place where acquisition rather than repeat purchase is the objective. Finding if absent on an acquisition account, because PMax will otherwise happily rebuy your existing customers and report it as growth.

6. Waste patterns that scale with budget

Apply the floor from the top of this page to every item here. Each check is a spend-with-no-return check, and each is worthless without the multiple attached.

• →Search Partners and Display spend hiding inside Search campaigns. Finding if either carries spend past your chosen CPA multiple with no conversions, or if nobody has ever segmented the campaign by network to look.
• →Broad match without a maintained negative list and a conversion feedback loop. Finding if the search-term view shows meaningful spend on queries no reasonable buyer would call relevant, above the floor.
• →Geographic and device spend misaligned with where margin comes from. Finding if a geo or device segment is past the CPA multiple with no conversions, allowing for genuinely long consideration cycles.
• →Auto-applied recommendations changing the account unreviewed. Finding if any are enabled and the change history shows applications nobody in the business can account for. This one has no significance floor — it is a governance failure regardless of outcome.

Ranking what you found, in dollars

A finished audit is a list of findings, and a list is not a plan. The ordering that matters is expected recoverable dollars, which is not the same as the dollars at risk — and here is where nearly every audit report, including the ones agencies sell, gets the arithmetic backward in both directions.

Excluding waste does not save money. It redirects it. If you negative $12,000 a month of genuinely wasted spend, your bill is unchanged — that budget now flows to inventory that earns your target return. At a 4.0 target, the finding is worth roughly $48,000 of incremental revenue, not $12,000 of savings. Reporting it as a saving understates it by a factor of your target ROAS, which is why waste findings are routinely undersold to finance.

The overstatement runs the other way. A finding is only worth its recoverable share: the portion above the significance floor, multiplied by the probability the fix actually works. "Twenty percent of spend went to zero-converting terms" is not a $20,000 finding if half that spend sits below 2x CPA, because half of it was never a finding at all. Size the recoverable part honestly and you will usually discover that the small measurement finding at the top of the audit outranks the dramatic waste number, because a mis-flagged primary conversion misprices every decision in the account rather than one slice of the budget.

Where these thresholds stop applying

• →The numbers here are calibrated for $50k a month and up. Below roughly $10k a month, the 30-conversion campaign floor is unreachable and should be read as "consolidate until it is true" rather than as a failure, and the CPA-multiple floors will flag almost nothing because no single entity accumulates enough spend to clear them.
• →The Poisson arithmetic assumes the term converts at the account average, which is the null hypothesis rather than a fact. A term genuinely worse than average will clear the threshold sooner, and one genuinely better will occasionally get caught. The threshold manages the error rate; it does not eliminate it.
• →Long consideration cycles break every zero-conversion check. If your typical path to purchase runs weeks, a 30-day window will manufacture waste findings out of ordinary lag. Widen the window to at least two purchase cycles before trusting any of them.
• →A clean audit does not mean good performance. It means the account is being decided well and read honestly, which is a precondition for good performance rather than evidence of it. The definition of what good looks like — scored, weighted, gated — is a separate document: the account standard.
• →Self-auditing fails in a predictable direction. You will find the problems you already suspected and miss the ones your own past decisions created, because the motivated reasoning is invisible from inside it.

If several checks come back as findings, the problem is rarely one campaign — it is the foundation, which is exactly the pattern in why most Google Ads audits miss the real issue. To size what a fixed account should return, run the breakeven ROAS and ad budget calculators. And if you would rather have this run against your actual account by someone with no stake in the previous decisions, our free audit does exactly that — read-only access, structural findings, no countdown timer.