- Most circulating audit percentages trace to other agency blogs, not to published datasets. Treat them as folklore.
- The failures they describe are real. They just have to be measured in your account, not quoted.
- Without a transaction ID, a reloaded confirmation page can fire the tag again and record a second conversion for the same order. [1]
- Analytics-created conversions arrive secondary by default: reported, but not fed to bidding. [2]
- The cleanest first check is platform-reported conversions against orders your business actually recorded.
- Only 11.8% of consent pop-ups met a peer-reviewed study’s minimal legal test, across 680 UK sites in 2020. [3]
- A disapproved product is not ranking badly. It is ineligible, and no bid fixes that. [4]
- Google publishes 30 conversions, 50 for Target ROAS, as an evaluation window, not a switch-on threshold. [5]
Pick any statistic from the last audit deck you were shown and try to trace it. A third of spend wasted. Seventy percent of accounts carrying a tracking error. Fifteen to thirty-five percent recoverable. Follow each one back and the trail almost always ends the same way: a blog citing a blog citing a third that has since been rewritten.
This is worth saying carefully, because the obvious conclusion is the wrong one. The failures those numbers describe are real and common. What is missing is the evidence for how common, and the distinction matters because an audit built on borrowed percentages is a sales document, while an audit built on your own arithmetic is a decision.
That is what a real number looks like: a named method, a stated sample, a peer-reviewed venue, and a claim narrow enough to be wrong. [3] Almost nothing else in the audit literature is built like this.
The six, and what each one leaves behind
1. Conversions counted more than once
This is the one with documentation behind it, which makes it the best place to start. Google states that if there are two conversions for the same conversion action with the same transaction ID, Google Ads will know the second is a duplicate. [1] Google states the reverse case directly: if the customer returns to or reloads the conversion page, some tags can fire again and two conversions can be recorded for the same order. [1]
The second route is structural. Google states that conversions created from Google Analytics are secondary by default and can only be changed to primary in Google Ads, and that a secondary action is for observation only, reported in the All conversions column but not used for bidding unless it sits inside a custom goal. [2] That default protects the bidder. It does nothing for a dashboard where somebody summed both columns.
The measurement: total platform-reported conversions for a period, against orders your finance system actually recorded for the same period. Not a benchmark. A subtraction.
2. Brand demand reported as performance
Brand campaigns return spectacular numbers because they intercept demand that already exists. Blended into one performance figure, they flatter everything around them and hide whether the rest of the account does anything.
Be careful here, because this is where the folklore is thickest. You will read that brand terms account for 60 to 80% of search conversions. That range is not traceable to any published multi-account dataset, and repeating it is exactly the failure this article is about.
The measurement: recompute your blended return with brand excluded, then look at the gap. The gap is yours, it is real, and it needs no industry average to interpret.
3. Campaigns too thin to read
Google recommends measuring performance over periods with at least 30 conversions, and 50 for Target ROAS, in order to evaluate results accurately. [5] Note that this is guidance about evaluation. It is routinely reported as the volume Smart Bidding needs to function. Google does say some strategies rely on a minimum of historical conversion data, but it never publishes 30 and 50 as that minimum, and the difference changes the fix. [5] A thin campaign is not necessarily broken. It is unreadable, and the remedy is consolidation for legibility. Trade coverage reaches the same fix from the opposite direction, arguing that consolidation is what lets the algorithm exit its learning phase. [6]
The measurement: count campaigns whose trailing 30-day conversions fall below Google’s evaluation floor, and what share of spend they hold. That does not mean they are failing. It means you cannot yet tell.
4. Products that cannot serve
Disapproved items are the cheapest finding in any audit because there is nothing to interpret. Google Merchant Center reports product issues and their status directly. [4] A disapproved product is not performing badly. It is absent, and since the feed decides which auctions you enter at all, no budget or bid change will ever surface it.
The measurement: disapproved products as a share of the catalog, weighted by the revenue those items produced when they last served.
5. Consent and the conversions that never arrive
The peer-reviewed evidence, from a 2020 scrape of UK sites, says most implementations failed a minimal reading of European law. It says nothing about what your opt-in rate is. [3] Those are separate claims and only the first one has a study behind it.
The measurement: your own consent rate by market, compared against sessions, and the modeled share of conversions your platforms are filling in. This is the same discipline that consent mode turns into a reporting problem the moment it is half-configured.
6. The value you send back
Every automated bidder optimizes toward the number you give it, which makes that number the highest-leverage line in the account and the least frequently audited. An account sending order totals is asking the system to buy revenue, not profit.
The measurement: compare the value your conversion actions send against contribution margin on the same orders. If they differ, everything downstream was optimized toward the wrong target, which is why what a conversion is worth belongs in an audit rather than in a strategy deck.
What separates the two kinds of finding
A finding with evidence
- Names the number as it currently stands in the account
- Names the number it should be, and where that value comes from
- Can be recomputed by someone else from the same data
- States what changes if it is fixed, in money or in decisions
- Survives the question “how do you know”
A finding that is an assertion
- Leads with an industry percentage rather than your figure
- Cites a source that cites a source
- Cannot be reproduced without the auditor present
- Quantifies the upside before measuring the problem
- Would read identically in any other account
The last line in that right-hand column is the fastest test available. Take the audit you were given, imagine handing it to a competitor in a different vertical, and ask how much of it would need changing. If the answer is the logo and three numbers, nobody looked at your account.
| The claim you will hear | What is actually established |
|---|---|
| A third of paid spend is wasted | Nothing published supports a general figure. Yours is computable |
| Most accounts have a tracking error | Plausible, unproven. Your duplicate count is a query |
| Brand drives 60 to 80% of search conversions | Untraceable to any dataset. Your split is a segment |
| Consent implementations are usually non-compliant | Supported for 2020 UK sites, peer-reviewed: 11.8% met the study’s minimal legal test [3] |
| Reloading a confirmation page can double-count | Documented by Google, and preventable with a transaction ID [1] |
The shape of an honest audit
Run the six checks in the order above, because it moves from cheapest to hardest and because the first one changes how you read every number after it. If the conversion count is wrong, every efficiency figure downstream inherited the error, and there is no point arguing about campaign structure on top of a broken denominator.
Then write it up with four columns and no adjectives: what the number is, what it should be, where the correct value comes from, and what changes when it is fixed. Anything that cannot fill all four columns goes in a separate section labeled as judgment, which is an honest thing to include and a dishonest thing to disguise.
The uncomfortable part is that a good audit is usually shorter than a bad one. Six measured findings with arithmetic behind them will lose a pitch against forty checkboxes and a waste percentage every time. That is a sales problem, not an evidence problem, and confusing the two is how the industry ended up quoting each other’s blog posts as data.
Sources
- Google Ads Help · Use a transaction ID to minimize duplicate conversionsThe duplicate-detection rule, in Google’s own words.
- Google Ads Help · About primary and secondary conversion actionsSecondary means observation only, and Analytics-created conversions arrive secondary by default.
- ACM · Dark patterns after the GDPR: scraping consent pop-ups and demonstrating their influenceNouwens, Liccardi, Veale, Karger, and Kagal, CHI 2020. The 11.8% figure and its sample.
- Google Merchant Center Help · Issues in Merchant CenterWhere product-level warnings and disapprovals are reported.
- Google Ads Help · About Smart BiddingThe 30 and 50 conversion guidance, framed as evaluation.
- Search Engine Land · How campaign structure shapes Google Ads performanceTrade coverage arguing for consolidation on learning-phase grounds, cited as the opposite route to the same fix.
Frequently asked questions
Why are audit benchmarks unreliable?
Because most of them have no primary behind them. Figures like the share of accounts with tracking errors, or the percentage of spend typically recoverable, circulate between agency blogs that cite each other rather than a published dataset. They may be directionally right. They are not evidence, and they should never be the basis of a decision about your account.
How does conversion double counting actually happen?
Two common ways. A customer reloads the order confirmation page, and without a transaction ID Google Ads counts a second conversion for the same order. Or the same event arrives twice, once from the tag and once from an Analytics import, and both land in reporting even though only one is biddable. Google’s fix for the first is a transaction ID, which lets it recognize the second conversion as a duplicate.
What is a secondary conversion action?
A conversion action that is reported in the All conversions column but not used for bidding, with one exception: it does count for bidding if it sits inside a custom goal. Conversions created from Google Analytics arrive as secondary by default and can only be promoted to primary in Google Ads.
How do I check whether brand is inflating my performance numbers?
Segment brand and non-brand into separate campaigns or at minimum separate reporting, then recompute your blended return with brand excluded. You are not looking for a benchmark. You are looking at the gap between the two numbers, which is the part of your reported performance that was demand you already had.
How much consent loss should I assume?
Measure it rather than assume it. The strongest published evidence on consent design is a 2020 peer-reviewed study of the five most popular consent platforms across the top 10,000 UK sites, which found only 11.8% met the minimal requirements the researchers drew from European law. That says implementations were usually wrong, not what your own opt-in rate is.
Does a disapproved product just rank badly?
No, it is absent. A disapproved item is not eligible to serve at all, so no bid change or budget increase will surface it. This is why the disapproval count is one of the first numbers worth pulling: it separates a performance problem from an eligibility problem.
What is the fastest check that finds real money?
Compare the sum of platform-reported conversions against the orders your business actually recorded for the same period. If the platforms claim more than you shipped, you have double counting, overlapping attribution, or both, and every optimization decision made on that number inherited the error.
Is 30 conversions a month a requirement for Smart Bidding?
It is published as an evaluation guideline, not an activation threshold. Google recommends measuring performance over periods with at least 30 conversions, and 50 for Target ROAS, in order to evaluate results accurately. Campaigns below that are not broken. They are unreadable, a different finding with a different fix.
What should an audit deliverable contain?
For every finding, the number as it stands in the account, the number it should be, the source of the correct value, and what changes if it is fixed. A finding without those four things is an opinion. Opinions are allowed, but they should be labeled.
Mikołaj Salecki
Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.
