Reading search terms like an auditor, not a spreadsheet
Intent, n-gram patterns, and account baselines — the signals that separate a costly query from a profitable one, and how they stack into a verdict.
Ask an experienced auditor to review a search terms report and they won’t start by sorting on cost. They’ll read it the way you read a room — looking for what the numbers are made of, not just what they add up to. That distinction is the entire difference between catching waste early and finding it three months later.
What a spreadsheet shows you — and hides
A spreadsheet is built to sort and total. Sorted by cost, it surfaces your single most expensive query — the term that already announced itself. It says nothing about the three hundred queries spending a few dollars each in the exact same wasteful pattern, because no individual row is big enough to rise to the top.
An auditor doesn’t read rows. They read three layers of signal underneath the rows — and stack them together before reaching a verdict.
Signal one: intent
The same keyword can carry entirely different intent depending on the exact phrase a searcher typed. “Project management software” is a transactional query — someone close to buying. “Project management software free” and “project management software jobs” share the root term but carry no buying intent at all. A spreadsheet sorted by cost treats all three the same way. Reading for intent means classifying what the searcher actually wants before judging whether the click was worth paying for.
Signal two: n-gram patterns
A single query rarely tells you enough. An n-gram analysis breaks queries into shared word fragments — “free”, “near me”, a competitor’s name — and looks at how those fragments perform across the entire long tail, not query by query. A fragment appearing in forty low-cost queries that individually look harmless can represent a real structural leak once you see it as one pattern instead of forty unrelated rows.
This is precisely the kind of waste a cost-sorted spreadsheet is built to hide: nothing in it is big enough to be alarming on its own.
What does the exact phrase suggest the searcher wants — transactional, informational, or navigational?
Do shared word fragments repeat across many queries in a way no single query reveals alone?
Is this normal for this account’s own history, or only normal compared to a generic benchmark?
Signal three: account baseline
A generic industry benchmark tells you almost nothing about a specific account. A conversion rate that would be alarming for one business is healthy for another with a longer sales cycle. Reading like an auditor means comparing a term against what’s actually normal for this account’s own history, not a number pulled from an unrelated industry average.
No single signal proves a term is wasteful. The pattern across all three does.
How the signals stack into a verdict
Wrong intent alone doesn’t condemn a term — it might still be new enough to warrant watching. A recurring n-gram pattern alone might be coincidence in a small enough sample. It’s when the three line up — off-intent, part of a repeating pattern, and abnormal against the account’s own baseline — that a verdict becomes defensible instead of a guess.
Intent classification, n-gram analysis, and your account’s own baseline are evaluated together for every term — not sorted by cost and reviewed one row at a time.
See how AdKinex worksA spreadsheet answers “what did this cost?” An auditor answers “does this deserve to keep costing it?” Only one of those questions actually stops the bleed.