Google Ads n-gram analysis: find the words that waste spend

A search terms report shows you what people typed. N-gram analysis shows you which words inside those searches cost money and never convert. Here's how to do it properly, including the checks that stop you switching off good traffic.

What n-gram analysis is

An n-gram is a run of n consecutive words. The search "emergency plumber near me" contains four 1-grams (emergency, plumber, near, me), three 2-grams (emergency plumber, plumber near, near me) and two 3-grams (emergency plumber near, plumber near me).

N-gram analysis splits every search term in your report this way, then adds up impressions, clicks, cost and conversions for each n-gram across all the searches it appears in. Instead of 4,000 rows of individual searches you get a list of words and phrases ranked by what they cost you and what they returned.

Why reviewing search terms one at a time misses waste

Most search terms in a report have tiny numbers: one or two clicks, a few dollars, no conversions. Individually none of them looks worth acting on, and with a few thousand rows nobody has time to read them all anyway. So the usual review sorts by cost, fixes the top 20 and stops.

The problem is that waste is usually spread thin. A word like jobs might appear in 40 different searches ("plumber jobs", "plumbing jobs near me", "apprentice plumber jobs"), each spending $3 or $4. None of them reaches the top of a cost-sorted list, but together they're a real line item. N-grams add those fragments up so the pattern is visible, and one negative keyword covers all 40 searches plus the ones that haven't happened yet.

It works the other way too. N-grams show which words reliably convert, which is useful for new keywords, ad copy and landing pages.

Step by step

1. Export the search terms report

1. In Google Ads, open Insights and reports, then Search terms. You can do this for the whole account or from inside one campaign.
2. Pick a date range. 60 to 90 days is a sensible start for most accounts; low-volume accounts may need longer to get enough data per word.
3. Make sure the columns include Search term, Clicks, Cost and Conversions. Campaign, Impressions and Conv. value are worth adding.
4. Click the download icon and choose .csv.

Two things to keep in mind. First, Google doesn't list every search: some low-volume terms are left out of the report, so its totals will usually be lower than your campaign totals. Second, conversions can be recorded days after the click, so if your range ends yesterday the most recent searches will look worse than they are. Ending the range a week or so back avoids that, or at least keep it in mind.

2. Split the terms and total each n-gram

You can do this with a spreadsheet and some formulas, a script, or a tool like TermSweep, which does it in your browser without uploading the file. Whatever you use, count each n-gram once per search term, then sum the metrics of every search term it appears in.

3. Read 1, 2 and 3-grams differently

• 1-grams find the broad themes: free, jobs, diy, course, a competitor's name, a product you don't sell. They give you the biggest, cleanest negatives, but also the riskiest, because one word can sit in good and bad searches.
• 2-grams add the context a single word lacks. how on its own might be mixed; how to is usually someone wanting to do it themselves, while how much is someone asking for a price.
• 3-grams are the most specific. They're useful for confirming what a 1 or 2-gram is really picking up and for finding narrow negatives when a shorter one would block too much. Their numbers are smaller, so they need more data before you trust them.

A good habit: when a 1-gram looks bad, look at the 2 and 3-grams that contain it before deciding. Use the shortest n-gram that captures the waste without catching the searches that convert.

4. Set thresholds before you start

Without thresholds every n-gram with one unlucky click looks like waste. Two numbers do most of the work:

• Minimum spend. Don't judge a zero-conversion n-gram until it has spent a meaningful amount. A reasonable floor is somewhere around your account's average cost per conversion: if a word has spent one CPA's worth and converted nothing, it's a candidate. If your conversion rate is low or conversions are expensive, set it higher. TermSweep defaults to $20, which you should change to suit the account.
• CPA multiple. For n-grams that do convert, flag the ones whose cost per conversion is well above the account's, for example 2× or more. These usually aren't negatives straight away; they're candidates for a closer look, lower bids, or splitting into their own ad group.

If you track conversion value rather than conversions, the same idea applies with ROAS: flag n-grams whose return is far below the account's.

5. Apply guardrails, then add negatives

This is where most of the damage is done, so take it slowly.

• Don't negative a word carrying more than about 10% of spend. If a single n-gram is in a large share of your searches, it's almost certainly part of your core traffic. Even if its numbers look poor, blocking it can take out a big piece of the account. Investigate it with 2 and 3-grams instead.
• Don't block your own keywords. Words from your keywords, brand and product names will appear everywhere. Keep a protect list and never add anything on it as a negative. Google Ads can flag negatives that conflict with your keywords, but don't rely on that.
• Check close variants. Your positive keywords match close variants such as plurals and misspellings; negative keywords don't. If you add job as a negative, jobs still gets through. Look down the n-gram list for singular and plural forms, abbreviations and common misspellings, and add each one you need.
• Use per-campaign negatives where it matters. A word that wastes money in one campaign can be what another campaign sells. commercial might be waste for a residential service campaign and the whole point of a commercial one. Run the analysis per campaign, or at least check the campaign split, before adding anything account-wide.

Then pick a match type. Phrase match negatives are a sensible default for n-grams, since they block any search containing those words in that order. The negative keyword list guide covers match types and how to add lists in Google Ads and Google Ads Editor.

6. Repeat on a schedule

New searches keep arriving. Running the analysis monthly, or every couple of weeks on bigger accounts, catches new patterns before they add up. Keep a note of what you added and when, so you can check later that nothing important dropped off.

A worked example

Here's a made-up local plumbing account to show the process. Over 90 days the search terms report shows $6,000 of spend and 150 conversions, so the account CPA is $40. The minimum spend threshold is set to $40 (one CPA) and the CPA multiple to 2× ($80).

Figures are invented for illustration. Shares add up to more than 100% because one search contains several n-grams.

Reading it:

• plumber and emergency carry far more than 10% of spend. plumber is the account's own keyword and emergency converts below the account CPA. Neither is touched.
• how converts, but at nearly four times the account CPA. Blocking it would also block "how much does a plumber cost", which converts well. The 2-grams split it cleanly: how to has spent $262 with nothing back, so "how to" goes in as a phrase negative and how much stays.
• jobs and diy are over the $40 floor with no conversions, so they're added. job is under the floor on its own, but because negatives don't cover close variants, it's added alongside jobs: it's clearly the same intent.

Those four negatives cover $531 of the 90 days' spend that produced no conversions. That isn't all saved money, since some of those clicks might have gone to better searches, but it's budget that was clearly going to the wrong people.

Common mistakes

• Too little data. A word with three clicks and no conversions tells you very little. Wait for the spend threshold.
• Judging by clicks instead of cost. Ten cheap clicks matter less than three expensive ones.
• Adding everything account-wide. Shared lists are convenient, but check that a word is bad everywhere before it goes on one.
• Never reviewing old negatives. Offers and services change. A negative added two years ago might now block something you sell.

This guide is about Search campaigns. Performance Max and Shopping campaigns report search terms and handle negatives differently, so check what's currently available for those before applying the same process.