Negative Keywords: N-Gram Analysis of Search Terms, Step by Step

Your account pays for queries nobody has looked at. In the search terms report that money is scattered across hundreds of rows at a few dollars each, and no single row stands out. You see it only after breaking every query into single words and word pairs and adding up the cost of each fragment.

The method does not need any particular tool. It works the same in a spreadsheet, in a script, or when a person does all the counting. The recording above walks through the whole process on one account, from raw queries to two exclusion lists, and below we lay it out step by step.

1. The full set of queries

Pull every query from the analysis period, ideally six months. The trap: an export capped at a few thousand of the most expensive rows looks complete. The account in the recording had 8,807 queries. In the top 3,000, two phrases came out with zero conversions in the target campaign, while the full data showed conversions there. Excluding them would have cut leads.

2. Context before the table

Before you count anything, pin down the account goal: what a lead is allowed to cost or what return sales have to deliver, and which campaigns actually spent in the period. Without a goal, a table of fragments is arithmetic and tells you nothing about what to exclude.

3. Splitting and the threshold

Break each query into one-, two- and three-word fragments and sum cost, clicks and conversions for each fragment. Set the threshold before you read the table, or your eye will stop at whatever happens to look odd. In the recording the threshold was a tenth of a percent of search term spend and at least thirty clicks. 473 fragments cleared it, and 78 of them brought no leads at all.

4. Zero conversions on your own products

Those 78 fragments are a list to check first and to exclude later. Most of them are the products themselves: the sizes, the formats and the words the business makes its money on. Zero conversions on a core phrase tells you to check the landing page and measurement first. Excluding such a phrase shuts out demand the business wants to serve.

5. Competitors in the queries

Competitor names are tempting to cut in bulk. In the recording, thirteen competing businesses showed up in the queries and most of them brought leads. Only three cost money and returned nothing. Only those three go on the list.

6. The level of the exclusion

Match every phrase against the campaigns it appeared in. A phrase that costs money and returns nothing in one campaign while bringing leads in another gets excluded in the first campaign only. An account-wide exclusion would cut the leads in the second.

A shared negative keyword list acts in every campaign it is attached to, immediately. In the recording the list was attached to thirteen campaigns, so one new entry took effect in thirteen places at once. Before adding a phrase to a list, check where that list is attached.

7. A preview before anything is sent

Before anything reaches the account, lay the change out in one table: campaign or list, phrase, match type and the number of campaigns in reach. That is where you see whether the exclusion lands where it should. For a shared list, the last column tells you the most.

Order matters

Steps one and six decide whether the rest makes sense. A truncated set of queries suggests excluding phrases that convert elsewhere, and an exclusion at the wrong level cuts leads from campaigns where the phrase works. Only after those two steps does the fragment table tell the truth about what eats the budget.