How to audit a Google Ads account, properly

A Google Ads account audit is supposed to answer one question: where is this account leaking money, and what is the cheapest thing to fix first? Yet most audits bury that answer under a hundred green ticks. So here is the version that leads with impact: the checks that actually recover spend, in the order a senior buyer runs them.

Start with the number you can't trust

Before you judge a single campaign, judge the measurement. An account optimising toward a misconfigured or value-less conversion will make every downstream number lie, and no amount of bid tuning fixes a broken signal. So the first check is not performance; it is whether performance is even measurable.

An audit that trusts a broken number just optimises the account deeper into the wrong thing.

This is the same discipline behind leading with the verdict: say what you can stand behind, flag the numbers you cannot, and never manufacture confidence.

The 12-point audit, ranked by recoverable spend

1. Conversion tracking integrity. Are conversions firing once, with the right value, on the right action? Duplicate or value-less conversions corrupt every optimisation downstream. Verify against Google's conversion tracking setup.
2. Wasted search terms. Pull the search terms report and find spend on queries that will never convert. This is usually the single largest pool of instantly recoverable budget.
3. Placement leakage. On Display, Video and Performance Max, spend drifts to low-quality placements at a fraction of core performance: the Google-side twin of the placement drift that hides inside a blended average.
4. Bid strategy fit. Is the strategy matched to the data volume? Target CPA and ROAS need enough conversions to learn; below that threshold they thrash.
5. Budget pacing. Which campaigns are capped while winners starve? Reallocation is free performance.
6. Search vs Performance Max overlap. Performance Max can cannibalise your branded and high-intent search. Check where they compete for the same query.
7. Negative keyword hygiene. Missing negatives let broad match bleed into irrelevant queries, which compounds with check 2.
8. Audience and geo signal. Are you feeding the algorithm useful audience signals, or letting it guess?
9. Ad strength and asset coverage. Thin responsive assets cap reach and quality, but chase this after the money checks, not before.
10. Landing page and tracking parity. Do the pages match the promise, and does analytics agree with Ads on conversions?
11. Attribution maturity. Recent numbers are still settling. A campaign judged inside its window reads pessimistic: the Google cousin of Meta's 7-day window problem.
12. Cross-platform comparability. If you also run Meta, remember the two cannot be compared directly before you shift budget between them.

Rank, don't just list

The difference between a useful audit and a report nobody acts on is ordering. Every finding should carry an estimate of recoverable spend so the account owner knows what to do on Monday morning, not a flat list where a $12,000-a-month leak sits next to a cosmetic asset warning.

• Attach a number to each finding. “Wasted search terms: about $9,000 a month” beats “improve keyword targeting.”
• Sequence the fixes. Some changes reset learning; order them so you do not undo one fix with the next.
• Make every figure verifiable. Each number should point back to where you can confirm it in the account. Trust is the whole point.

What a ranked audit looks like when it comes out of the account itself rather than a template: one run on one live account, every item priced and sorted into classes before anyone opened a spreadsheet.

Why this doesn't stay done

The catch with a one-off audit is that accounts drift. Search terms shift weekly, placements leak again, pacing changes with seasonality. A PDF from last quarter describes an account that no longer exists. The real fix is not a better audit template; it is a standing analyst that keeps re-reading the account and flags what changed.

That is the job Adgent is built for. It is a team of AI agents working toward your marketing goals, and it reads every connected Google and Meta account each day. Several of its ten specialists map straight onto this list: Measurement Integrity on tracking, Placement & Network Optimizer on placements, CPC-Spike Autopsy on bidding, Budget-Pacing Cop on budget exhaustion, and Account Architect on structure. What they find comes back ranked by an estimated monthly amount, with the evidence and next move attached.

Red flags: what a broken account looks like

Some findings you have to calculate. Others you can spot in thirty seconds: the tells that an account has been left on autopilot. If you see any of these, you have found recoverable spend before you have done a single sum.

• Conversion action set to “Every” instead of “One.” A lead form counting every submission inflates your conversion count and teaches Smart Bidding to chase repeat form-fillers. For lead gen, this single toggle can materially overstate conversions.
• No audience exclusions on retargeting or Performance Max. You are paying full CPA to re-serve ads to people who already converted. On accounts with a healthy return rate, that is spend re-buying customers you already have.
• Performance Max with zero audience signals. An empty signal means the algorithm cold-starts on your money. Performance Max will find the cheapest conversions it can, usually your existing branded traffic, and take credit for demand you would have captured for free.
• A search terms report nobody has opened in 90 days. Broad match without weekly negative-keyword pruning bleeds into junk queries continuously. If the last negative was added six months ago, assume a meaningful share of search spend is landing on queries that will never convert.
• Target CPA or ROAS running on under 30 conversions in 30 days. Google's own guidance is to judge a Target CPA campaign over at least 30 conversions in the last 30 days. Below that, per campaign, Smart Bidding does not have enough signal to learn: it thrashes, and CPA swings wildly week to week.
• One catch-all ad group with 50+ keywords. Structure this loose means no query gets a relevant ad or landing page. Quality Score suffers, CPCs climb, and you cannot diagnose anything because everything is averaged together.
• Auto-applied recommendations turned on. Google's auto-apply will quietly expand match types and raise budgets on its own schedule. An account where Google is making the changes is not an account you are auditing; it is one you are discovering.
• Display and Search fused in one campaign. “Search with Display Select” pools two different funnels under one budget and one CPA target. The Display half almost always drags, hidden inside the blended number: the same trap as placement drift inside an average.
• Ads-reported conversions that do not match analytics. When Google Ads claims 400 conversions and GA4 sees 250, someone is optimising on a number that is not real. Past roughly 15 to 20% (our working threshold for investigate versus ignore), we treat it as a measurement problem to fix before the bids; inside that band, call it attribution noise.

Manual audit vs continuous audit

The format of the audit decides what it can actually catch. A person running a checklist once a quarter and software reading the account through the API every day are looking for the same leaks, but they see very different amounts of them.

Neither wins outright. A manual audit brings judgment a script cannot; a continuous one catches the leaks that reopen the week after a manual audit ships. The strongest setup is a standing analyst for the high-drift checks and a person for the strategy calls, which is exactly what an AI media buyer is meant to be.

How often should you audit?

Audit cadence should match how fast each check goes stale. Some things barely move; others rot within a week. A single quarterly review treats them all the same, which is why so many audits feel out of date the moment they land.

• Full audit: quarterly. The complete 12-point pass: structure, bid strategy, attribution, account hygiene. These are slow-moving and worth a deep, unhurried look four times a year.
• Light review: monthly. A focused pass on budget pacing, ad strength and the campaigns that shifted most. Thirty minutes to catch anything the quarter would let drift too far.
• Continuous: for the high-drift checks. Wasted search terms, placement leakage and budget pacing move week to week. These want watching, not scheduling: the moment they slip, spend leaks, and a quarterly cadence means you find out 89 days late.

The blunt version: audit the slow things on a calendar and watch the fast things continuously. Anything you only check quarterly, you are effectively not checking.

Common Google Ads audit mistakes

A bad audit can be worse than an incomplete one: it draws confident conclusions from bad inputs. These are the mistakes that turn a review into a list of wrong decisions.

• Auditing before you have checked conversion tracking. Every performance judgment rests on the conversion signal. If you rank campaigns before verifying it fires once, with the right value, you are ranking noise. Measurement is step one for a reason.
• Treating small-sample placements as decisions. A placement with 40 impressions and one conversion is not a winner, and one with 200 clicks and no sale is not dead: it is just noise. Cutting or scaling on samples this thin creates the very volatility you were trying to fix.
• Ignoring the attribution window. Recent conversions are still landing. Judge a campaign from inside its window and it always reads pessimistic: the Google cousin of Meta's day-one data problem. Let the window close before you call a verdict.
• Comparing across platforms as if they measure the same thing. Google and Meta count conversions differently, so a raw ROAS comparison is apples to oranges. Understand how each one counts before you move budget between them.
• Producing a to-do list instead of a ranking. A flat 40-point list where a $12,000-a-month leak sits beside a cosmetic asset warning gets skimmed, not acted on. Rank by recoverable spend, or the audit changes nothing.

Frequently asked