N-Gram Analysis for Google Ads Search Terms
N-gram analysis splits every search term into one and two word sequences and sums cost against the sequences instead of the terms. It turns a long tail of individually small rows into a short list of patterns worth negating.
A search term report in a working account has hundreds of rows, each with a few clicks and a small amount of spend. Individually none of them looks like a problem, so nothing gets negated, and the long tail keeps costing money every month. N-gram analysis is the arithmetic that turns that list into ten rows you can act on.
The method
An n-gram is a sequence of n words. Split every search term into its one-word and two-word sequences, then sum cost and conversions against each sequence rather than against the term.
The term free roof repair quote contributes to the unigrams free, roof, repair, quote and the bigrams free roof, roof repair, repair quote. Do that across two thousand terms and a word like free stops being invisible: it appears in eighty terms, has spent a real share of the month, and has converted nothing.
That is the whole idea. The value is not sophistication, it is aggregation — seeing a pattern where the interface only shows you instances.
Reading the output correctly
Three columns matter: cost, clicks, and the number of distinct search terms containing the n-gram.
High cost with a high term count is the strongest signal. It means a pattern, not one unlucky query. A single expensive term that happens to contain a common word is noise.
Zero conversions is not the same as no value. An n-gram with four clicks and no conversions has told you nothing yet. Set a cost floor high enough that everything you look at has had a fair chance to convert, and set it relative to your cost per conversion, not to a round number.
N-grams overlap by construction. The same click is counted under every gram its term contains, so the total of the column is several times the account spend. Read the rows, never the sum. Any tool that reports “total wasted spend” from an n-gram table is adding up the same money repeatedly.
Going from the report to a negative
Aggregation tells you where to look. It does not tell you what to exclude, and the step between the two is judgement.
- Read the actual terms behind the gram. A word that looks disqualifying in isolation is often fine in context. Cheap in a market where you compete on price is not the same as cheap in a market where you do not.
- Decide the match type deliberately. Phrase negatives are the usual right answer for a bigram; broad negatives on a single common word can remove far more than intended.
- Decide the level. A gram that is bad everywhere belongs in a shared negative list applied account-wide. A gram that is bad in one campaign only belongs in that campaign.
- Check for conflicts. A negative that overlaps an active keyword blocks it entirely, and the interface will not stop you. See negative keyword conflicts.
Frequency and window
Weekly, over thirty days. Daily volume is too thin to draw conclusions from, and anything longer than a quarter mixes in a version of the account that no longer exists.
Run it after any significant structural change too. A new campaign, a match type loosened, a Performance Max campaign added — each of those changes what the account matches against, and the previous month’s conclusions expire.
Where the method runs out
Performance Max does not report raw search terms. It reports grouped categories, so n-gram analysis on Performance Max traffic is working with a summary someone else has already made. The technique still helps on Search campaigns, which is where most of the recoverable waste sits anyway.
The script that does this is the n-gram waste report. It is read-only: it produces the list, and the decision stays with you.
Questions
- What is an n-gram in Google Ads?
- A sequence of n consecutive words taken from a search term. Unigrams are single words, bigrams are word pairs. Summing spend against these sequences reveals patterns that individual search term rows hide.
- Can I add up the cost column to get total wasted spend?
- No. N-grams overlap, so the same click is counted under every sequence its search term contains. The column total is several times account spend and is meaningless.
- How often should I run n-gram analysis?
- Weekly over a thirty day window is the usual balance. Shorter windows lack the volume to draw conclusions from, and longer ones mix in account structures that no longer exist.
- Does n-gram analysis work on Performance Max?
- Not properly. Performance Max reports grouped search categories rather than raw search terms, so the input is already a summary. The technique is most useful on Search campaigns.