Google Ads N-Gram Analysis: Find Wasted Spend Without Scripts

Introduction

A Search Terms Report can contain hundreds of low-cost queries with no conversions. Viewed separately, they look too small to worry about. Their combined cost can tell a different story.

Suppose 46 queries contain the word "jobs." Each accounts for €2–€9 in spend, so few appear near the top when the report is sorted by cost. Group those queries by the repeated word, however, and the total is hard to ignore: €284 spent, 71 clicks, and no recorded conversions.

N-gram analysis makes such patterns visible. It splits search terms into recurring words and phrases, then totals their impressions, clicks, cost, and conversions. It does not label traffic as good or bad. It reduces a long query report to a manageable set of patterns that still require review.

If the account needs a broader diagnostic first, use the 30-minute Google Ads account audit. N-grams are more useful after tracking, campaign scope, and the primary conversion action have been checked.

What is n-gram analysis in Google Ads?

An n-gram is a sequence of nwords. In paid search, those words usually come from the Search Terms Report. Take the query "free accounting software for students":

• 1-grams: free, accounting, software, for, students
• 2-grams: free accounting, accounting software, software for, for students
• 3-grams: free accounting software, accounting software for, software for students

The same split is applied to every query, and the metrics are added together for each recurring sequence. If "free" appears in 120 queries, its 1-gram row shows their combined performance. A 2-gram such as "free trial" has its own row. That distinction matters when the business offers a free trial but wants to exclude searches for free templates or downloads.

When to use 1-grams, 2-grams, and 3-grams

Start with 1-grams to find broad intent markers: jobs, free, template, course, used, or wholesale. Single words collect data quickly and often expose the largest concentrations of spend. The missing context also makes them the easiest to misread.

Move to 2-grams when a word has several meanings. "Free consultation" may describe the offer, while "free download" may attract the wrong audience. Three-word groups can be useful in a large account, but the data thins out quickly. A 3-gram seen twice is rarely enough evidence for an account-wide exclusion.

Why query-by-query review misses cumulative waste

The native report is organized around complete searches. That works well when one expensive query needs attention. It works less well when the same irrelevant intent is spread across dozens of inexpensive variations.

Sorting by cost still ranks complete queries, not the words they share. Sorting by conversions has the same limitation: several useful variations may each show one conversion, while their shared phrase performs well in aggregate. N-gram totals add a second way to inspect the data. The original queries remain essential for context.

Prepare the Search Terms data

An n-gram table can look authoritative even when its source data is wrong. Before reading it, check the reporting window, selected campaigns, conversion columns, and currency.

Choose a useful date range

The range should cover the normal conversion delay and contain enough search terms to reveal repeated language. Thirty days may be sufficient for a busy e-commerce campaign. A low-volume B2B account may need 60 or 90 days. Do not combine periods with different offers, markets, or tracking setups unless those periods can be separated later.

Recent clicks with no conversions may still convert. If the account has a seven-day conversion lag, leave the latest seven days out of a waste review or assess that spend separately.

Export the right columns

At minimum, keep these fields in the CSV:

• Search term
• Campaign and ad group
• Impressions and clicks
• Cost
• Conversions and conversion value, where applicable

Campaign and ad group labels preserve context. A term that is irrelevant in a generic campaign may be intentional in a competitor or recruitment campaign. Without those labels, an account-wide total can make a valid search pattern look like a negative keyword candidate.

How to read an n-gram report

Start with a question, not the longest list or the highest impression count. Finding wasted spend, expanding keyword coverage, and checking landing-page relevance each require different filters.

Find waste candidates

For a budget cleanup, filter for n-grams with meaningful cost and no conversions. Sort by cost from highest to lowest, then open the queries behind each candidate. Read them as a group: do they express an intent that the business cannot serve?

A zero-conversion row is not automatically waste. The sample may be too small, conversions may be recorded offline, or one costly query may account for most of the total. A phrase becomes a negative candidate only after its contributing searches have been checked.

Find patterns worth keeping

Some recurring phrases point to useful demand. A phrase with solid conversion volume may deserve its own keyword, ad group, ad copy, or landing-page section. A high-impression phrase with a weak click-through rate may show that the ad does not address the search clearly.

This broader use is covered in the Search Terms Report best-practices guide. Blocking traffic is only one possible outcome of the review.

Set thresholds from account economics

A fixed rule such as "add a negative after five clicks" ignores how much a conversion is worth and how often the account normally converts. Set the review threshold in relation to an acceptable acquisition cost.

If the target CPA is €80, a non-converting phrase with €6 of spend is weak evidence. At €160, it deserves immediate investigation. For e-commerce, compare cost with the margin available from an average order, not revenue alone. A threshold should decide what gets reviewed, never what gets excluded automatically.

Worked example: scattered job-seeker traffic

Consider a lead-generation account for commercial cleaning services. Its report contains these searches:

• office cleaning jobs near me
• night shift commercial cleaner jobs
• cleaning supervisor jobs
• part time office cleaner vacancy

No complete query appears often enough to stand out. The 1-gram "jobs" combines 38 searches with €214 in cost and no conversions. "Vacancy" accounts for another €47 across nine searches. The underlying rows are clearly from job seekers, while the landing page sells cleaning contracts to businesses.

Blocking every complete query as an exact negative would deal only with searches that have already occurred. A reviewed list of recurring job-seeker terms will also cover new variations. First, check whether the company runs recruitment campaigns. If it does, apply the list only to customer-acquisition campaigns.

Turn a pattern into a safe negative keyword

An n-gram row identifies a pattern, but the row itself is not always the right keyword to add. Read the matched searches and find the word or phrase that represents the unwanted intent without also covering useful searches.

Identify the unwanted intent

If every search containing "jobs" is irrelevant, that single word may be sufficient. If "free" occurs in both "free template" and the valuable phrase "free consultation," the negative should be more specific, such as "free template."

Check singulars, plurals, misspellings, and word order. Google Ads does not expand negative keywords to close variants in the same way it expands positive keywords. Relevant variants may need separate entries.

Choose the negative match type

• Negative broad: blocks a search when it contains every word in the negative keyword, even in a different order. The search may contain additional words.
• Negative phrase: blocks searches containing the complete phrase in the same order. Words may appear before or after it.
• Negative exact: blocks the search only when it contains the same words in the same order, without extra words.

Over-blocking is harder to notice than continued waste because excluded searches disappear from future reports. Preview the affected queries and keep a change log. The weekly search-term checklist provides a cadence for reviewing those decisions.

Run the analysis in MirachSEM

Spreadsheet formulas and Google Ads scripts can produce n-gram tables, but the review and negative-keyword list usually live in separate places. MirachSEM puts them in the same panel.

1. Open the Search Terms Report and enable the MirachSEM panel.
2. Open the N-grams tab. Visible report rows are available for a quick directional check.
3. For a full-report analysis, export the report from Google Ads and upload the CSV in the Source section.
4. Choose 1-, 2-, or 3-grams, set the match mode, and filter by spend, conversions, or the Waste preset.
5. Open a candidate and read the search terms and metrics behind it.
6. Edit the proposed negative when the n-gram is broader than the unwanted intent, choose a match type, and preview the affected queries.
7. Add the reviewed keyword to the active list. Visual Picker, Bulk Editing, and Ctrl+M remain available for the surrounding term-by-term cleanup.

Simulate list removal recalculates candidate metrics as though the active negative list had already been applied. This keeps traffic covered by an existing negative from being counted again as a new saving. The panel also shows which search terms the active list would have removed, along with their spend, clicks, conversions, and CPA.

See the MirachSEM n-gram analysis guide for the controls and source-data details. MirachSEM works best in the Search Terms Report and can also be used on the Keyword Planner page when preparing lists before launch.

Where n-gram analysis goes wrong

• Using too little data. Two clicks and no conversions say little about a phrase, regardless of how prominent the row looks.
• Treating the report as a record of every search. Google does not include every query in the Search Terms Report. N-gram totals cover the terms in the source file, not all campaign traffic.
• Ignoring conversion lag.Yesterday's clicks may not have had time to produce leads or sales.
• Mixing incompatible campaigns. Recruitment, brand, competitor, and acquisition campaigns can assign different value to the same word.
• Counting overlapping n-grams as separate savings.The rows for "jobs," "cleaning jobs," and "office cleaning jobs" may contain many of the same queries.
• Adding the aggregate row without reading its queries. Weak overall performance can hide a profitable sub-theme.
• Optimizing against broken tracking. No-conversion filters cannot distinguish irrelevant traffic from a missing conversion event.

A repeatable n-gram routine

Review high-spend campaigns weekly and run a full-report CSV analysis once a month. Keep the reporting window, filters, reviewed patterns, added negatives, and affected historical spend in the change log. Return to recent exclusions after the account's normal conversion delay has passed.

Report impact without double-counting

Time savings depend on the process being replaced. For a team that spends 15 hours a month exporting reports, maintaining formulas, and copying negatives, a 3× faster workflow would reduce that work to about five hours and free roughly 10. Account size, data quality, and review standards all affect the result.

Apply the same caution to estimates of wasted spend. An analysis that flags 20% of historical spend has not proved that the whole 20% could have been saved. Some clicks may convert later, and overlapping n-grams may include the same queries. Remove that overlap and report the spend covered by the final negative list, not the sum of every candidate row.

MirachSEM includes a 5-day trial with no credit card required.