Where Google Ads Budgets Actually Go: A Search Terms Analysis of 4 Accounts

Yesterday we looked at how eight advertising accounts were built. Today's question matters more to anyone paying the invoices: what exactly is the money buying. We exported twelve months of search terms from the four accounts where conversion tracking demonstrably works, and sorted every query into buckets. Here is what came out, with the numbers, and without the generalisations that fall apart under scrutiny.

How we counted

Of the eight accounts we audited yesterday, only four qualified for this analysis. The criterion is simple: conversions have to be recorded. Where tracking is silent, every query automatically looks worthless and the conclusions lose their meaning.

The search terms report covers search campaigns. Queries that triggered Performance Max are largely invisible there, so the figures below describe not the whole budget but the part of it where we know for certain what was paid for.

Every query went into one of three buckets:

• Converting. The query produced at least one conversion during the year.
• Long tail. Fewer than ten clicks and no conversions. Judging these one by one is pointless: ten clicks is too few to claim anything.
• Assessable non-converting. Ten or more clicks and zero conversions. This is where conclusions become possible.

We then classified the third bucket using dictionaries built separately for each account: product words, other companies' brands, own brand, personal names. The classification is approximate, and where it goes wrong we say so plainly.

Three buckets: how the spend divides

The first thing that stands out is the size of the tail. Between 22.6% and 56% of spend went on queries that each gathered fewer than ten clicks. In one of the online stores that means 43,000 distinct phrasings. No specialist will read those one at a time, which is why the tail is managed through structure rather than negatives: match types, and how loosely the system is allowed to interpret each keyword.

The second is the spread in the third bucket: from 1.6% in one account to a third of spend in two others. That is not an industry difference. It is a difference in how the account is configured.

Why one account is clean and the others are not

The obvious hypothesis is that more negative keywords mean less spend on junk. The data supports it only partly.

The cleanest account does have by far the densest negative list, at 873 per thousand clicks. But it has a second distinguishing feature: broad match covers only 12.9% of its keywords, with the rest on phrase and exact.

Healthcare B is the telling counterexample. It carries plenty of negatives, 159 per thousand clicks, more than twenty times the density of either online store. Yet a third of its spend still goes on non-converting queries. The reason is that 86.7% of its keywords sit on broad match: the system finds new phrasings every day faster than anyone can exclude them.

What sits inside the non-converting queries

The interesting part begins when the third bucket is broken into its components. The shape of the leak differs in every account, which is exactly why generic advice about negative keywords is so often imprecise.

Queries with no product word

In the first online store, two thirds of non-converting spend went on queries containing no word connected to the product or to buying. People were searching for content on a subject adjacent to the range, not for the product itself. To put it with a hypothetical example from another category, it is like a coffee machine retailer paying for "cappuccino recipe".

These queries gather hundreds and thousands of clicks because they are high-volume and cheap. Each click costs little, but together they form the largest single leak in the account.

What to do: this is the one category that genuinely warrants mass exclusion as negatives. At the same time, check whether the keywords pulling these queries in are set too broadly.

Other companies' brands

In the second online store, 59% of non-converting spend went on queries naming other companies: product manufacturers and other retailers. In Healthcare B the classifier directly recognised 17% of that spend, but a manual review turned up several more laboratories missing from the dictionary. Counting those, the competitor share there is about 21%.

This category mixes three different query types, and each needs different handling. The first is navigational: a company name with "website", "official site", "address" or "phone". The searcher wants that specific company, and a click on your ad rarely ends in a purchase. The second is a product plus a manufacturer's brand: if you stock that brand, the query is entirely relevant. The third is a company name with "jobs" or "reviews". In the second store those queries took 7% of all non-converting spend.

What to do: navigational queries about other companies go on the negative list. Keep "product + manufacturer brand" queries if you stock the brand, and send them to that brand's page. "Jobs" always goes on the negative list. "Reviews" of a brand you sell can stay, provided the landing page actually shows reviews; reviews of another retailer are a negative. And if you deliberately want to appear on competitor brands, do it in a separate campaign with its own budget and lower conversion expectations.

Relevant queries that did not convert

This is the most important category because it is the easiest one to get wrong. Across the four accounts, between 32% and 100% of non-converting spend went on entirely appropriate queries: product names, specific models, services with a price or a city attached. In the second store we also counted model-number queries with no generic words here: the classifier initially put them under "other", but they express direct product interest.

Healthcare A is the clearest case. Only six queries in that account gathered ten or more clicks without converting, and all six ask about the price of a specific service. None of them is junk.

What to do: exclude nothing. A relevant query that does not convert signals a problem with the landing page, the price or the competitive offer, not with the query. Check which page the ad points to, whether it shows a price, and how your offer looks next to the neighbouring results.

Your own brand not converting

In Healthcare B, 10.6% of non-converting spend went on queries containing the company's own name. Someone searches for you by name, clicks your ad and does nothing measurable. That does not happen under normal conditions: branded traffic is usually the best-converting traffic in the account.

The cause here turned out not to be the queries at all. Part of the landing page addresses passed through a redirect that dropped the click parameters. People reached the site and could phone, but the link between the call and the ad was lost. That likely also explains part of why only 11% of this account's spend sits on converting queries.

What to do: if your own brand is not converting, stop looking at queries and check the path from click to conversion. Take the final URL from the ad, append a test click parameter, walk the whole route to an enquiry, and confirm the parameter survives to the final page.

Names of specific people

A category specific to healthcare: 7.1% of non-converting spend in Healthcare B went on queries containing the full names of individual doctors. Someone is looking for a specific doctor, and broad match sees a resemblance to your services.

What to do: names of doctors who do not work for you belong on the negative list. Names of your own specialists, on the other hand, deserve their own ad group: they are among the warmest queries there are.

How to find this in your own account in an hour

1. Open the search terms report for a full year. A shorter period gives too few clicks to conclude anything.
2. Filter to 10+ clicks and zero conversions. Sort by cost. That is your third bucket.
3. Work through the first hundred and give each query one of five labels: no product word, other brand, relevant, own brand, personal name.
4. Calculate the shares. Even the first hundred will show which category dominates in your account.
5. Look at the tail separately. If queries with fewer than ten clicks take more than a third of spend, the problem lies in match types, not in particular words.

What this data does not prove

Four accounts in four industries are an illustration, not statistics. We do not present these shares as a market norm.

The search terms report does not show Performance Max, so leaks inside automated campaigns are absent from this analysis. The classification is approximate: the dictionaries were built by hand for each account, and some category boundaries are a judgement call. And "zero conversions" depends on the attribution window: some queries may have assisted a purchase that was credited to another click.

One thing the data does show with confidence: the same report holds genuine junk right next to perfectly good queries that simply have not converted yet. Telling one from the other can only be done by hand.

This continues yesterday's structural review: what we found in 8 ad accounts. We showed how the review process itself works in the case study, and collected the basic rules for negative keywords in a dedicated guide. If you want us to sort the queries in your own account into these buckets, talk to our team.