Almost every week we are handed an account to review with the same premise: "the campaigns aren’t performing, maybe it’s the bids". Sometimes that is true. Far more often the problem sits three levels lower, in a conversion action counting the same lead twice or a cookie banner that never passes consent to the tags. Changing bids on bad data fixes nothing; it just moves the problem.
That is why a serious audit is not a list of checks in no particular order, but a sequence in which each step assumes the previous one has been verified. This is the sequence we use, across nine areas. For each: what to look at, what bad looks like, and why it matters.
| # | Area | The typical red flag |
|---|---|---|
| 1 | Tracking and consent | Google Ads conversions far from back-office figures |
| 2 | Conversion goals and values | Micro-conversions set as primary |
| 3 | Account structure | Many campaigns, few conversions each |
| 4 | Bidding and budgets | Targets unconnected to real margins |
| 5 | Search terms and negatives | Search terms report unopened for months |
| 6 | Ads and assets | One ad per ad group, assets missing |
| 7 | Landing pages | All traffic sent to the home page |
| 8 | Performance Max and Demand Gen | Budget absorbed by brand or by unchecked channels |
| 9 | Change history and recommendations | Recommendations auto-applied without review |
1. Tracking and consent
What to check. That every conversion action records a real event, exactly once, and that consent signals reach the tags. Start with conversion action diagnostics in Google Ads, then run a live test: a form submission or test purchase, accepting and then refusing cookies, with tag manager debug tools open.
What bad looks like. The same lead counted by both the Google Ads tag and a GA4 import. A thank-you page that reloads and fires a second conversion. Modelled conversions at zero in an account that claims to run Consent Mode in advanced mode. A ratio between recorded conversions and real orders that worsens month on month for no apparent reason.
Why it matters. Because every piece of automation in the account learns from here. It is the point we made in Consent Mode v2 and the DMA: misconfigured consent does not produce visible errors; it produces a system that optimises confidently in the wrong direction. If this area does not hold up, the audit stops here and restarts from here.
2. Conversion goals and values
What to check. Which actions are set as primary (the ones bidding optimises towards) and which as secondary (observation only). Which goals each campaign uses: the account defaults or campaign-specific goals. Whether conversion values exist, and whether they reflect real economic value or are a number entered to fill the field.
What bad looks like. Phone number clicks, page views or scroll depth set as primary conversions alongside quote requests. Every action valued at 1. An ecommerce account optimising for order count rather than revenue. A lead generation business treating a qualified lead and a spam enquiry identically.
Why it matters. Smart Bidding does not know what you care about; it only knows what you told it to count. If the cheapest micro-conversion is primary, the system will learn to buy that. Attribution weighs in here too: data-driven attribution redistributes credit across clicks, and comparisons with GA4 need the caveats set out in data-driven attribution across Google Ads and GA4. For lead generation, the real step up is feeding the sales outcome back into the account, as explained in the offline conversions guide.
3. Account structure
What to check. How many campaigns exist, why they are separate, and how many conversions each one generates. How ad groups are organised, whether brand and non-brand are split, and whether location and language settings match the market you actually serve.
What bad looks like. One campaign per important keyword, inherited from years of management in layers. Campaigns overlapping on the same searches. The location option set to presence or interest when presence was needed, with clicks arriving from countries you do not ship to. Brand and generic terms mixed in one campaign, so the blended CPA looks excellent and hides the true non-brand figure.
Why it matters. Fragmentation was good practice when bids were managed by hand. Today it splits the signal: ten campaigns with three conversions a month each learn worse than two with fifteen. Our rule: a separate campaign needs a budget, goal or reporting reason to exist, not merely a different theme.
4. Bidding and budgets
What to check. Each campaign’s bid strategy, its targets (CPA or ROAS) and where they came from. Impression share lost to budget and to rank. How often targets have been changed. Whether the budget-limited campaigns are the best or the worst performers.
What bad looks like. A target CPA picked by feel, with no calculation against margin. Targets changed every week, so learning never settles. The best campaign limited by budget while a mediocre one spends everything it has. Manual bidding on campaigns with the volume for automation, or the reverse: Smart Bidding on campaigns with two conversions a month.
Why it matters. The target is the most direct instruction you give the system, and since 17 August 2026 it carries more weight in delivery, as we covered in Smart Bidding Exploration and promotion mode. How to derive a sensible budget from your goals rather than your wallet is in the guide to how much budget you actually need.
5. Search terms and negative keywords
What to check. The search terms report for the last 90 days, sorted by spend with no conversions. Existing negative lists, the level they are applied at and when they were last updated. Whether any negatives are blocking good searches.
What bad looks like. Informational queries, job searches ("vacancies", "salary"), unwanted competitors or products you do not sell absorbing a visible share of spend. No shared lists at all. Or, at the opposite extreme, hundreds of broad match negatives added in a hurry that also cut traffic that converts.
Why it matters. This is where waste is recovered fastest, without touching the budget. With broad match and AI Max, the keyword alone no longer decides where you appear; control shifts to search terms and negatives. Architecture and cadence are covered in negative keywords in 2026.
6. Ads and assets
What to check. The responsive search ads live in each ad group, the real variety across headlines, and the account- and campaign-level assets (sitelinks, callouts, structured snippets, images, logo). The Quality Score components: ad relevance, expected CTR and landing page experience.
What bad looks like. Fifteen headlines saying the same thing fifteen ways. Ads that never mention the ad group’s theme. Sitelinks all pointing to the home page. Assets disapproved for months without anyone noticing. Ad groups with "below average" relevance on keywords carrying most of the spend.
Why it matters. The ad drives CTR and, in part, cost per click. But Quality Score should be read as a diagnostic, not chased as a target: what is worth fixing and what is not is in the piece on Quality Score in 2026.
7. Landing pages
What to check. Where traffic actually lands: the landing pages report, split by campaign and device. Mobile speed, consistency between ad message and page, and how easy the form or checkout path is. Conversion rate per page, not just per campaign.
What bad looks like. All traffic sent to the home page. A twelve-field form on mobile. A page promising a discount the ad never mentions, or the reverse. A landing page converting far below the others while receiving the largest share of clicks.
Why it matters. Conversion rate multiplies everything else: moving from 3% to 4% cuts cost per conversion by a quarter without an extra penny of budget. It is the subject of the landing page is half the campaign.
8. Performance Max and Demand Gen
What to check. For Performance Max: the channel performance report, search term insights, brand exclusions and negatives, final URL settings, feed quality where Merchant Center is involved, asset groups and audience signals. For Demand Gen: which channels are active, which audiences, and against which conversion goal.
What bad looks like. A Performance Max campaign living largely on brand searches, inflating ROAS with demand you would have captured anyway. Final URL expansion switched on and sending traffic to blog or support pages. A single asset group for a mixed catalogue. A Demand Gen campaign judged on last-click conversions, as if it were Search.
Why it matters. These are the campaigns where the aggregate result most easily looks good while the substance does not. Channel reporting has made visible much of what used to be hidden; how to read it is in Performance Max in 2026, and the proper role of upper-funnel work is in the guide to Demand Gen campaigns.
9. Change history and recommendations
What to check. Change history over recent months: who changed what, when, and whether shifts in performance line up with any of those moments. Then the Recommendations page, and in particular which recommendation types are set to auto-apply.
What bad looks like. Broad match keywords added by an auto-applied recommendation. Budgets raised without anyone deciding to. A drop in conversions that begins the day someone edited a conversion action. An account managed with 100% optimisation score as the goal.
Why it matters. Change history is the one place where the cause of a problem is written down with a date and an author: it is the most underrated diagnostic tool in the interface. As for recommendations, some are useful, but optimisation score is an estimate of how closely the account is set up according to Google’s suggestions, not a measure of how well it performs. Assess them one by one; never apply them in bulk.
What we do with the output. An audit that produces eighty equally weighted recommendations helps nobody. The useful result is a short list ranked by impact: usually two or three fixes on tracking and goals, one or two on structure and bidding, and the rest as maintenance. You always start from the top of the sequence.
How often to audit
A full audit makes sense when an account changes hands, when results move without an obvious cause, or at least once a year. In between, lighter recurring checks are enough: search terms and negatives on a regular cadence, change history after every significant intervention, and a tracking check after every change to the site or the cookie banner. It is these, not the annual audit, that stop a problem running for months.
If you want to see this sequence applied to your own account, it is exactly the path our free audit follows: it starts with tracking and ends with a short list of fixes, ranked by impact. You can see how we work, and the answers to the questions we get most often in the site’s FAQ.