On this page
Where the waste actually sits
This page measures how zero-conversion spend is distributed inside an account. For its share of the whole budget, use the separate study of how much Google Ads budget is wasted.
The size and the shape of the waste are separate questions, and our Wasted Spend & Search Terms studies take them in turn.
Order every zero-conversion search term in an account from most expensive to least, then walk down the list. In the median account you reach 48.8% of the wasted spend after the first 10% of terms. Take only the top 5% and you already hold 34.6%. Widen to 20% and you hold 66.2%.
The comparison worth holding is the dashed line. If waste were spread evenly across terms, the top tenth would hold a tenth. It holds close to half.
Two checks before that number travels anywhere. Leaving any single account out of the portfolio moves the median only between 48.5% and 49.2%, so no one account is carrying it. And accounts with long term lists show the same concentration as accounts with short ones (48.8% against 49.8%), so this is not an artifact of some accounts simply having a longer tail.
That tenth is still 13,806 queries
Here is where the usual reading breaks. “Concentrated” normally implies “small”, and therefore “fixable in an afternoon”. Not here.
The median account in this portfolio carries 138,066 search terms that spent money and returned no conversion. Ten percent of that is 13,806 terms. That is the head of the distribution, the part holding half the waste, and it is still five figures of individual queries.
So both of these are true at once, and most audits only carry the first one home:
- Half your wasted spend is reachable through a small fraction of your terms.
- That fraction is far too large to work through by hand, one negative keyword at a time.
The concentration curve tells you which rows to inspect first. With thousands of terms in its expensive head, group recurring patterns before reviewing individual exclusions.
Update, October 2026: a weaker concentration, still a large head
Newer data still shows a concentrated head, just a lighter one. In the latest usable audit exports, from late 2025 to early 2026, the most expensive 10% of zero-conversion search-term rows held a median 35.7% of the zero-conversion spend. That is well below the original 48.8%, though still far above the 10% an even spread would give. So the head is still where to start, even if it carries less of the problem than the original figure suggests.
The head is still large. Its median size in those exports was 13,038 exported rows, five figures again, so the lesson of the previous section stands: concentrated in money, enormous in count. We also tested a list you could actually keep by hand: the 100 most expensive zero-conversion rows held a median of just 2.2% of that spend. In these exports, a hand-maintained list reaches very little of the measured waste.
A newer snapshot of the managed portfolio, with the same top-10% rule, lands in between: 43.6% across 2025 and a provisional 42.0% for the first half of 2026. These are different groups of accounts and windows, so read the three new figures as separate measurements, not as a trend or as proof that any account improved. Concentration survives every check. The original 48.8% is a measurement of one portfolio, not a benchmark for yours.
What to do with a head this size
The finding does not tell you what to exclude. It tells you what shape of work actually matches the problem.
- Sort by cost, not by count. Open the search terms report, filter to zero conversions, sort by spend descending. The first screen is worth more than the next hundred. Anyone auditing alphabetically or by impressions is working the wrong end of the curve.
- Work in patterns, not in terms. With five figures of rows in the head, the unit of work has to be a pattern that catches many terms at once, not the individual query. That is a different job from maintaining a list.
- Measure your own curve before you judge it. Take your zero-conversion terms, sort by cost, and check what share of the wasted spend your top 10% holds. If it is well under 45%, your waste is flatter than this portfolio’s and the head-first approach will pay you back less.
- Do not read zero conversions as zero value. A term that converted nothing in a six-month window may still have assisted, and the export cannot see that. Treat the head as a list of candidates to examine, not a list to delete.
What your own account cannot give you is the comparison: whether your curve is normal, whether your head is unusually heavy, and how much of it repeats across campaigns. Running that against a portfolio is what a Profit Forensics examination does.
When this does not apply
Your account is small. With a few hundred zero-conversion terms rather than a hundred thousand, the head is genuinely small and a maintained list can keep up. The concentration still holds; the practical conclusion flips.
Your conversion window is long. Considered purchases that convert weeks later will show as zero-conversion terms inside a reporting window and inflate what looks like waste. The concentration shape survives this, the label “wasted” does not.
You run mostly Performance Max or Shopping. This measures the search terms report. Where you get no term-level visibility, there is no curve to sort.
Researcher’s take
Two things are true here and most audits carry home only the first. Half the money sits in the first tenth of the list. That tenth is 13,806 rows in the median account. Read the first fact alone and you plan an afternoon with a spreadsheet. Read both and you plan a different unit of work: patterns that catch many terms at once. In the accounts I audit the ones with a genuinely small head are the small accounts, though that is my observation from auditing rather than a measurement from this slice.
Key takeaways
- In the original data, the most expensive 10% of zero-conversion search terms hold a median 48.8% of the wasted spend (middle half of accounts: 45.8% to 58.9%).
- The top 5% hold 34.6%; the top 20% hold 66.2%. Even spread would put the top 10% at 10%.
- The result survives leave-one-account-out (48.5% to 49.2%) and is not an artifact of list length.
- That top tenth is a median of 13,806 terms, out of 138,066 zero-conversion terms in the median account. Concentrated in money, not in count.
- Sort by cost and work in patterns. A hand-maintained negative list cannot cover a head of that size.
Frequently asked questions
Where is wasted spend concentrated in a Google Ads account?
In the expensive head of the zero-conversion search terms. Across the managed portfolio, the top 10% of those terms by cost held a median 48.8% of the wasted spend, against the 10% you would see if waste were spread evenly. The top 5% already held 34.6%.
How many negative keywords would it take to cover the worst offenders?
More than a list can hold. The top decile that carries half the waste is a median of 13,806 search terms per account, because the median account has 138,066 terms that spent and never converted. This is why the work has to be done in patterns rather than one term at a time.
Method and sources
Doctor Ads Profit Forensics is the research desk of a Google Ads practice with 19 years and a large managed spend sample behind it.
Figures come from a forensic analysis by Igor Ivitskiy of a managed portfolio of Google Ads accounts. The population is every account with at least 20 paid search terms carrying zero conversions, deduplicated to one reporting window per account: the accounts we analysed and a large sample of zero-conversion terms, with windows mostly spanning September 2024 to January 2025. Counts are of exported search terms, and Google truncates the search terms report, so the tail is longer than any export shows and the head is measured against what was exported. For each account the zero-conversion terms are ordered by cost and the cumulative share of that account’s zero-conversion spend is measured at fixed cut points; the figures reported are medians across accounts, with the middle half shown as a range. The 10% cut point was fixed before the run, alongside 5% and 20% precisely so that the choice of threshold could not be tuned afterwards, and a kill condition was registered in advance: a median at or below 20% would have closed the topic. Robustness: leave-one-account-out moves the median between 48.5% and 49.2%, and splitting accounts by list length gives 48.8% against 49.8%, which rules out the concentration being an artifact of longer tails. This is a managed-agency portfolio with client self-selection, so it describes what these accounts contained rather than what any market must contain, and nothing here is a causal claim. Definitions follow Google’s documentation for the search terms report, negative keywords and keyword improvement. Related reading from the same portfolio: half the competitors in your auction insights report have no impression share, another case of a report that hides its own structure.
The October 2026 update keeps the original unit and rules. It uses paid search-term rows with zero recorded conversions, in accounts with at least 20 such rows. Rows are ranked by cost within each account, and the top 10% is rounded to a whole number of rows (halves to even), as in the original study. The fixed-shortlist check takes the first 100 rows of each account’s ranking. The audit-export check covers 848,056 such rows from two windows, 1 November 2025 to 4 February 2026 and 1 October 2025 to 28 January 2026, with one window per account. The original study’s rule treated an empty conversion field as zero; this update leaves those rows out rather than label unknown rows as waste. The other check uses a 10 July 2026 snapshot of the managed portfolio, with one account refreshed in September 2026: January to June 2026 covers 191,941 zero-conversion rows and January to December 2025 covers 462,925 rows, with no empty conversion fields among paid rows in either window. Medians give each account equal weight. Counts are exported rows, not distinct queries, so a query that appears in several ad groups is counted once for each. The first-half 2026 figure is provisional, because recent conversions may still arrive.