Why your search term report hides the waste it is supposed to show
A search term report has thousands of rows and almost no repeats. The pattern costing you money never appears as a single line item — which is exactly why n-gram analysis exists.
Open the search term report on any account spending more than a few thousand a month and you will see the same thing: thousands of rows, sorted by cost, where the top row accounts for perhaps 0.4% of spend. Nothing looks like a problem, because no individual query is big enough to be one.
This is the structural flaw in reading the report row by row. Search queries have a very long tail. The waste is real, but it is distributed across hundreds of rows that each look too small to bother with.
The pattern is the unit, not the query
Consider an outdoor equipment retailer. Their search term report contains:
free camping gear— $612, 0 conversionsfree hiking boots— $443, 0 conversionsfree tent giveaway— $388, 1 conversion- ...and fifteen more queries containing the word "free", each between $30 and $200
No single row justifies a decision. Together they are $2,148 a month buying one conversion.
An n-gram breaks every query into its component one, two and three-word sequences and aggregates performance across all of them. The word "free" stops being fifteen unremarkable rows and becomes a single line reading $2,148 spent, 1 conversion, CPA $2,148.
That is a decision you can make in four seconds.
Where n-gram tools go wrong
Two failure modes are common enough to be worth naming.
Reporting patterns that are not significant. If a two-word n-gram appears in three queries with $40 of combined spend and no conversions, that is not evidence of anything. Three clicks with no conversion is a completely ordinary outcome for a keyword with a 4% conversion rate. A tool that lists it as a finding is generating work, not saving it.
Ignoring match type when you apply the negative. Finding the pattern is half the job. Adding it as a negative in a way that blocks what you intended — and nothing else — is the other half, and it depends on getting exact, phrase and broad negative semantics right, including plurals, misspellings and word order.
What to do with it
The practical loop is short:
- Sort n-grams by wasted spend, not by impressions or frequency.
- Ignore anything the tool has flagged as below significance.
- For each pattern worth acting on, decide whether it is universally bad (shared negative list) or bad in this context only (ad group negative).
- Check the negative you are about to add is not going to block a keyword you are actively bidding on.
Step four is the one people skip, and it is why negative conflicts exist as a category of problem at all.
Read next
The most expensive distinction in paid search
Conversions fell. Is that a performance problem or did your tag stop firing? Getting this wrong costs more than any bidding mistake, and most dashboards do not help you tell.
Your ad test is probably reporting winners that do not exist
Comparing the best of five ad variants against a threshold designed for two comparisons inflates false positives dramatically. Here is the arithmetic, and what to do instead.
Make your next change an informed one.
Tell us about your accounts. We’ll help you explore the findings, the workflow and the results with a 14-day trial.
14-day trial · No credit card · Approval required by default