How to audit a Google Ads account that is already spending

When we take over a Google Ads account, the first thing we look at is not performance. It is whether the performance is real.

On one inherited account last spring, Google Ads reported 214 conversions for the quarter. The sales manager’s spreadsheet, the one she actually ran the team from, had 51. Hold on to those two numbers, because they come back later and they turn out not to mean what everybody in the room assumed they meant.

That gap, in some size, is the most common finding in an inherited account. Not a bad keyword. Not a lazy advert. A reported conversion figure that is wrong, usually in the advertiser’s favour, with a year of decisions stacked on top of it. Which is why the order below is a dependency chain rather than a preference. Each step is only meaningful if the one above it checked out.

Why tracking has to come first

Every automated bidding strategy in Google Ads optimises toward the conversions you report. Target CPA, Target ROAS, Maximize conversions, the bidding inside Performance Max and Demand Gen: all of it. The system has no concept of a real customer. It knows what you told it a conversion is, and it will go and buy more of exactly that, competently and at scale.

Which makes a tracking error a spending error, compounding daily. If the tag fires on a page refresh, the algorithm learns to buy traffic from people who refresh pages. If a newsletter signup sits alongside a purchase as a primary action, the algorithm chases the cheap one, because that is how it hits its target. If half your European traffic arrives with no consent signal, those conversions never make it back and the bidding quietly bids down the segments that were working.

What to check, specifically

Which conversion actions are primary. Goals, Conversions, Summary. Only primary actions are counted in the Conversions column and used for bidding by default. We regularly open accounts with five or six primary actions including phone number clicks, PDF downloads and time on page. Anything that is not the money event belongs on secondary, observe only.
Custom goals, which are the trapdoor. A campaign-specific custom goal can include a conversion action that is labelled secondary and bid on it anyway. So the primary and secondary labels alone do not tell you what the campaign is optimising for. Open each campaign’s conversion goal settings and read what is actually inside.
Duplicate counting. The same event defined once in Google Ads and once imported from Analytics, both counted. This roughly doubles reported conversions and is very common in accounts that have changed hands.
Count settings. One conversion per click for lead generation, every conversion for ecommerce. A lead form set to every counts one person submitting three times as three leads.
Conversion values. With no values, value based bidding is impossible. With one flat arbitrary value on every lead, Target ROAS is Target CPA wearing a hat. That is fine as a starting position, as long as nobody is presenting it as revenue optimisation.
Consent signals. For European traffic, confirm Consent Mode v2 is implemented and that ad_user_data and ad_personalization are genuinely reaching the tag, not merely that a cookie banner exists. Without them you lose remarketing audiences, conversion modelling and enhanced conversions for that traffic.
Enhanced conversions. Check whether it is on and matching. Note that Google has said web and leads will be combined into one setting, which its documentation still lists as coming, and that from 15 June 2026 offline conversion uploads moved to the Data Manager API. If the previous agency was importing closed deals through the old Google Ads API path, that pipeline may have stopped working this summer and nobody noticed.
Attribution and lookback windows. Write them down. They explain a great deal of apparent performance change that has nothing to do with the campaigns.

Then do the part nobody enjoys. Take one month of reported Google Ads conversions and reconcile them against the CRM or the inbox, counting only what paid search can claim. If the two numbers are within a reasonable margin, carry on. If they are not, stop the audit here and fix the measurement, because everything below this line is built on sand. The mechanics are in our piece on conversion tracking that counts real enquiries.

Step two: find where the money actually went

Once you trust the numbers, go after the waste. In most accounts this is the fastest source of real savings.

Pull the search terms report for 90 days, sorted by cost, under Campaigns, Insights and reports, Search terms. Read it rather than skim it. You are looking for wrong intent, things the business does not sell, and people hunting for free or DIY versions of the service. There is usually a block of spend sitting on terms the owner would reject on sight. Our guide to reading search terms reports covers how to bucket them.

Where the waste comes from is predictable.

Broad match with no maintained negative list. Broad match plus Smart Bidding can be good, but it needs a mature conversion signal and someone tending the negatives. On a thin account it is an expensive way to discover your category’s adjacencies.
AI Max arriving without an invitation. From September 2026 Google began automatically upgrading campaigns that used automatically created assets or the campaign level broad match setting, switching on keywordless search term matching. Dynamic Search Ads were excluded from that wave; their sunset and auto-upgrade begin in February 2027. Check whether this happened to the account, and compare the search terms mix either side of the upgrade date. If the account widened in autumn with no edit in the change history, this is why.
Search partners and display expansion left on. Segment by network and compare cost and conversions per network. These are frequently on by default from the setup wizard.
Location targeting on presence or interest, which is the default and includes people merely interested in your area. A dental clinic in Varna does not need clicks from people reading about Varna from another continent. Campaign settings, Locations, Location options, set to presence.
Performance Max absorbing brand traffic. If Performance Max runs alongside search, check whether it is serving on your own brand terms and taking credit for conversions the brand campaign would have won far cheaper. Campaign level negative keywords and brand exclusions are both self-serve now, so this is controllable rather than something to complain about. Also open the channel performance report, at Campaigns, Insights and reports, channel performance, which breaks spend out across Search, Display, YouTube, Discover, Maps, Gmail and Search partners for any date range after 6 June 2025. Before that report existed this was guesswork. It is not any more, so look.
Automatically applied recommendations. Check Admin, Account settings, Auto-apply, or the Auto-apply control at the top of the Recommendations page. The list of what can be auto-applied includes adding broad match keywords, adding keywords, expanding reach with search partners and adjusting CPA and ROAS targets, which is to say most of the decisions you were hired to make. We turn almost all of them off. One useful note: from 26 January 2026 the add responsive search ads recommendation stopped being auto-applicable, so that particular surprise is retired.

Quantify the waste as a share of total spend and write the figure down. It is the clearest single thing you will show the client.

Step three: the structural faults that actually matter

Fix structure only where it stops you controlling budget or reading results. Nobody has ever been thanked for an elegant restructure.

Brand and non-brand sharing a campaign. The most damaging fault in the list, because brand search converts cheaply and subsidises the real cost of acquisition. Mixed together, the account looks profitable while the acquisition half is not. Separate them before judging anything else.
Fragmentation. Twenty campaigns each producing two conversions a month. No bidding strategy has enough signal to work with and none of them ever will. Consolidate.
The reverse problem. One campaign holding three unrelated services on a shared budget, where the highest volume service eats everything and the highest margin one never serves.
Ad group bloat. Fifty keywords sharing one set of adverts that cannot be relevant to all of them.
Geography and language mixed. Countries with very different click prices and conversion rates on one budget are always misallocated, in the direction of the cheapest clicks.

Also check the account level things people forget: shared negative lists actually applied, conversion goals set at the right level, ad schedules, and whether anything is paused that should not be.

Step four: bidding and where the budget sits

Now compare where the money goes against where it works. Sort campaigns by spend, then by cost per acquisition. The usual pattern is that the biggest spender is not the best performer and nobody has moved money, because nobody had a number they trusted enough to move it on.

Impression share lost to budget. A campaign hitting its target cost per acquisition while losing significant impression share to budget is money sitting on the table. Raise the budget before touching anything else.
Impression share lost to rank on the best converting terms. That is a bid, quality or relevance problem, and it usually means the advert or the page is weaker than the competition rather than the bid being too low.
The 17 August 2026 change, which reframes a lot of old reporting. Budget limited campaigns using Target CPA or Target ROAS now deliver toward the target as set instead of appearing to beat it. Google’s worked example is a campaign running at a 5 dollar cost per acquisition against a 10 dollar target, which now lands nearer 10. So if the previous manager’s reports show a Target CPA campaign comfortably outperforming its target, that reading expired in August. Check whether reported cost per acquisition stepped up in late August for no campaign reason, and then check what the target says, because in a budget limited campaign the target is now an instruction rather than a ceiling.
Targets set once and never revisited. A Target CPA set two years ago at a figure the account has beaten every month since was throttling volume before August and is now simply spending to it. Move targets in steps of 10 to 15 per cent.
Strategies on campaigns with no signal. Target ROAS on a campaign producing four conversions a month is a random number generator. Google’s own documentation says Target CPA can run with no conversion history, so the familiar rule about needing thirty conversions to switch is a misreading; what thirty conversions is actually for is evaluation, over a window of at least thirty days. The practical point stands either way. If you cannot gather thirty conversions in a month, you cannot judge the strategy, so stop pretending the numbers mean something and consolidate until you can.

Step five: adverts and landing pages

By now you know what is true, what is wasted and where the money should sit. Only now is creative worth judging, because a good advert on a broken account still loses.

On adverts: check that each ad group has responsive search ads with genuinely distinct headlines rather than ten paraphrases of one sentence. Check pinning, which is the most common self-inflicted wound we find, because heavy pinning removes the only mechanism the format has. Treat ad strength as a rough signal and not a target. Then check whether sitelinks, callouts and structured snippets exist at all. Missing assets are among the cheapest fixes available.

On pages: click every advert yourself, on a phone, on mobile data. Look for a gap between the promise in the advert and the first line on the page, and for traffic landing on a homepage that should be going to a service page. Time the load. Count the form fields. Check the telephone number is tappable. A slow, generic page behind a nine field form caps the account regardless of how well the campaigns are run, and no amount of bid tuning repairs it. If cost per lead has been drifting up without an obvious cause, the page is often where it went, which we covered separately in cost per lead is climbing and nothing obvious changed.

The audit checklist, in order

List every conversion action and confirm which are primary. Demote everything that is not the money event.
Open each campaign’s conversion goal settings and check for custom goals containing secondary actions.
Look for the same event counted twice, once natively and once imported.
Check count settings: one per click for leads, every for ecommerce.
Check conversion values and whether value based bidding is even possible.
Confirm Consent Mode v2 is implemented and the advertising consent signals actually fire.
Check enhanced conversions status, and whether any CRM import still runs since the move to the Data Manager API.
Reconcile one month of reported conversions against the CRM or inbox. Stop here if they do not match.
Read 90 days of search terms sorted by cost. Quantify the waste as a share of spend.
Check whether the account was auto-upgraded to AI Max, and compare the search terms mix either side of the date.
Segment by network. Turn off search partners and display expansion if they underperform.
Set location targeting to presence only.
Open the Performance Max channel performance report and check brand absorption, then add campaign level negatives and brand exclusions.
Turn off the automatically applied recommendations nobody chose.
Separate brand from non-brand.
Consolidate campaigns too fragmented to gather signal, and split the ones where one service or country is eating a shared budget.
Review impression share lost to budget and to rank, campaign by campaign.
Review every bid target against the last 90 days of actual performance, and against the August 2026 change.
Move low volume campaigns off target based strategies.
Check responsive search ads for distinct headlines, minimal pinning and complete assets.
Click every advert on a phone and grade the page for match, speed, form length and clarity.
Write one page: what is wasted, what is misallocated, what is broken.

Then make the call the client is actually paying for

Most audits stop at the list and hand it over. What the client needed was a decision about whether this account is worth repairing, and the honest version of that decision sometimes costs you the easy retainer.

Repair it if the tracking is correct or correctable, the account holds a real history of genuine conversions, and the problems are waste, misallocation and neglect. History has value. Two years of clean conversion data is an asset, and rebuilding resets the learning, which you will pay for in a worse first month.

Rebuild it if the structure mixes brand, geography and services so thoroughly that separating them means touching everything anyway, or if the account was built for a product range, business model or market the client no longer has. Either way, build the replacement in parallel rather than by deletion: run the new campaigns small until they produce, then move the money across, and keep the old account because its history is still the only benchmark you have.

And rebuild it, without much hesitation, if the conversion data was always wrong. Which brings us back to the 214 and the 51.

Everyone in that meeting treated the gap as a reporting problem, something to be corrected and apologised for. Those 163 conversions that never happened were the cheap part of the damage. The expensive part was eighteen months of Smart Bidding doing its job perfectly: studying the people who triggered a tag that fired on page refresh, learning what they had in common, and going out every morning to buy more of them. The account was not mismeasured. It was trained. Every pound of that budget bought it a little more skill at finding people who were never going to buy anything, and that skill is what you would be keeping if you fixed the tag and carried on.