The obvious fix wasmore negative keywords.It stops half a percent.
A consumer brand selling in markets across Europe, with five years of search terms nobody had read end to end. The plan was the usual one: find the junk searches and exclude them. The client is not named here, and every amount is rounded and converted to euros. It is also a case where the scoring showed the plan was aimed at the wrong problem.
- €800k
- spent over five years on searches that never converted and that nothing excludes, 18% of all search spend
- €66k
- of it went to the searches every proposed negative would block, each search counted once
- 0.5%
- of this year’s dead spend is what the 227 confident negatives stop
- 963
- product phrases converting at under half the expected rate, which no negative fixes
Too many searches to read, and a name the brand does not own alone.
Five years nobody had read
307,662 search terms across 18 accounts, in more languages than anyone on the team reads. A person reviewing a search terms report reads the top of it. Whatever waste there was sat in the long tail, where no single line looks worth excluding.
A name shared with other companies
The brand shares its name with businesses in unrelated industries. Searches for them triggered the brand’s ads, and a keyword rule cannot tell one company from another without knowing what each of them sells.
Negatives nobody could audit
Every account already carried negatives at campaign, ad group and shared list level. Nobody could say which were missing, which sat in the wrong place and which were blocking searches that sold.
Score every suspect, then let the clicks decide.
Every search term, and every negative already live
Five years of search terms came out of the Google Ads API through a read-only gateway: every term with at least one click. The same pull took every negative already in place, at campaign, ad group, shared list and manager level, because recommending something that is already live is noise. The API returns 10,000 rows a request, so the pull was split by month and by campaign until no request hit the ceiling.
A decision model instead of a reviewer
Jev, a decision model, read 12,742 candidates, from single search terms to short phrases repeated across hundreds of them. For each one it returned the probability that it belongs in negatives and a category. It was told what the brand sells and which other companies share its name. At 0.70 or above a negative goes on the list, between 0.45 and 0.70 a person decides, and 1,689 borderline scores were run a second time in a different batch and averaged.
Proof before exclusion
A search with no conversions proves nothing until it has had enough clicks to convert. With zero conversions, a search is worse than comparable traffic once its clicks pass 3 divided by the conversion rate of the same account and the same brand or non-brand segment: between 81 and 1,582 clicks here, depending on the account. Most single searches never get there, so clicks were pooled over every phrase of one to three words the searches share, and 8,152 of those phrases were tested with the correction that testing thousands at once requires.
Checking what is already live
Every recommendation was matched against the negatives in place: 185 of the 227 exist nowhere in the accounts, and 42 exist in other accounts but not where the search runs. The same check found 3,014 searches excluded in one campaign and still bought in another, and 86 existing negatives blocking searches that had converted 147 times. Everything lands in a sheet, with 781 rows ready for Google Ads Editor.
The waste was real. Negatives were the wrong tool for most of it.
Dead spend here means searches with no conversion in five years that no negative blocks today. Over the five years it came to about €800k. The searches that all 2,002 proposed negatives would block cost about €66k of it, and the 227 confident ones about €10k. The bars show how much of the last twelve months of dead spend each list would stop, with every search counted once even when several negatives match it.
Even with every doubtful candidate approved, the lists stop less than a twentieth of the money that buys nothing this year. The rest goes to people searching for the product itself, and 963 of the phrases they use convert at under half the rate of comparable traffic. That is a bid and structure problem, and an exclusion list would only hide it.
Your search terms will split differently. The order of the questions will not.
Score first, prove second, check third. Every suspect gets a score instead of a glance at the top of a report, nothing is excluded until its clicks can prove it converts worse than comparable traffic, and nothing is recommended before checking whether it is already live somewhere else in your accounts. The model does the reading at a scale no one on your team can, and the statistics stop it from excluding a search that simply has not had enough clicks yet.
Do not expect the same split, and do not expect a refund. Negatives were not where the money was in this account: the confident list stops half a percent of this year’s dead spend, and the useful result was knowing that before spending weeks on exclusions. Yours may divide the other way, and you only find out once the scoring is done. As for the refund, a negative stops future spend, and the budget it frees has to earn its place somewhere else before it counts as saved.
This work was delivered by MTA Group, which Zero Fluff Digital is part of, by the same specialists who would work on your account. The figures on this page are rounded amounts, shares and counts the team computed from its own pull of the account’s search terms and negative lists through the Google Ads API, scored by a decision model and tested for significance. They are not an export anyone else published. The client is not named and the figures are given as changes rather than amounts, at their scale rather than in their currency. There is no third-party study to link to here, so treat this as weaker evidence than the cases that carry one, and ask us for the workings on a call. The reviews behind our work are public and verified on Clutch, where clients rate the work rather than the agency describing itself.
Measured from September 2021 to September 2026.
Your search spend leaks somewhere. Scoring tells you where.
30 minutes, no deck. You leave knowing how your search terms would be scored and what the first pull from your accounts would need.
Book a callNext case studyS&G Plus: prospecting only on Google Ads, 2x return by week three →