How to Find Negative Keywords and Wasted Spend With N-gram Analysis

The short answer

The usual way to find negative keywords is to scroll the search terms report and pick out bad searches one by one. That works for a few hundred rows. It fails when the report has thousands of rows that each cost a dollar or two, because no single row looks worth worrying about.

N-gram analysis fixes that. It breaks every search term into words and short phrases, then adds up the cost, clicks and conversions for each one. Suddenly “jobs” isn’t three cheap searches. It is one word that spent $98 and never brought a customer. In a large account, the same word can sit inside dozens of searches.

This guide covers Google Ads and Amazon Ads: where to get the data, how to set limits from your own numbers, and how to add negatives without blocking searches that make you money.

What n-gram analysis is

An n-gram is a run of words inside a longer text: one word is a 1-gram, two words in a row a 2-gram, and so on. Each n-gram inherits the numbers of the search term it came from. Add up the totals across every row, and you can see which words and phrases cost money across the whole report.

Here is a made-up example for a plumbing company, to show why it works:

No single “jobs” search spent more than $40. Together they spent $98 on job seekers, not customers.

The same view also shows winners. A phrase with many conversions at a low cost per conversion is a candidate for its own keyword.

Get the right search term data

N-gram analysis is only as good as the report you feed it.

Google Ads search terms report

Open Campaigns, then Insights and reports, then Search terms, and download the report. Google’s About the search terms report page explains two limits worth knowing:

• Not every search is listed. The report shows search terms that a significant number of people used. Rarer searches are left out for privacy. So your phrase totals will be lower than your real spend, and some waste stays invisible.
• The Match type column describes the search, not your setting. It shows how closely each search matched your keyword, which can be narrower than the match type you chose for that keyword.

Add the Conversions column, and also the Keyword column, which Google hides by default. You need conversions to judge waste, and keywords to check that a new negative won’t block one of them.

Pick a date range that has had time to convert. Google lets you set a conversion window of 1 to 90 days (About conversion windows), and says conversions can be reported up to 90 days after the click, depending on the window you chose (Find out how long it takes for your customers to convert). A search clicked last week may still bring a lead next month. So end your date range at least as long ago as your customers usually take to convert. The Days to conversion segment in your campaign or keyword report shows that delay, and Google’s own example for checking it uses a date range that ends at least 30 days ago. Our guide to how long your sales cycle is explains how to read it. Until a phrase’s clicks are older than your usual delay, “no conversions yet” doesn’t mean waste.

Performance Max campaigns (Google’s all-in-one campaign type) have their own search terms view. Negatives there work differently: Google says negative keywords in Performance Max apply to Search and Shopping inventory only.

Amazon Ads search term report

Amazon’s search term report for Sponsored Products has a few rules of its own:

• Only search terms with at least one click are included, so impressions won’t match campaign manager.
• The lookback is 65 days, with summary or daily time units.
• Sales are counted within 7 days of the click for sellers, and 14 days for vendors and authors (Ad campaign attribution). In Amazon’s newer Reporting section the columns are simply called Sales and Purchases, without the window in the name (Conversion metrics), so check which account type you have.
• ASINs can appear as search terms. An ASIN is Amazon’s product ID number. Entries like b00ipgvvz4 are product pages where your ad showed, in automatic campaigns and product-targeted ad groups. Those are blocked with negative product targeting, not negative keywords.

Leave out the most recent week or two. Amazon reports a sale on the date of the ad click, and says the numbers for a date stay incomplete until the lookback window has ended (Ad campaign attribution). So end your date range at least 7 days ago if you are a seller, or 14 days ago if you are a vendor. Amazon also says sales data can take up to 48 hours to populate, and cost can change for up to 3 days while clicks are validated (Performance metrics). A search term with clicks but no sales from last week may simply not have its sales yet.

Note the reporting change. Amazon’s migration page says its Sponsored ads reports page shuts down on December 31, 2026, and that saved reports, scheduled reports and historical data still on that page will be permanently deleted. Amazon offers a migration tool that moves saved and scheduled reports to the newer Reporting section. Build new search term reports there, and download any old reports you want to keep before that date.

Set your own limits

Every n-gram tool needs a rule for “this phrase is wasting money.” Neither Google nor Amazon publishes one, and any article that gives a universal number of clicks is guessing. The right limit depends on your conversion rate and on what a sale or lead is worth to you. Here is how to work it out from your own account.

1. Start with relevance, not numbers. Some words are wrong no matter what the data says: jobs for a plumber, free for a paid course, a competitor’s product you don’t sell. If a word can’t lead to a customer, you don’t need to wait for it to spend more.

2. For Google Ads, use your cost per conversion. Decide what you are willing to pay for a conversion, or use your account’s average cost per conversion. A phrase that has spent more than that with no conversions has already used up a full conversion’s budget. In the example above, at a $60 target, “jobs” spent $98, well past the line.

3. Or use your conversion rate. Multiply the phrase’s clicks by your usual conversion rate to see how many conversions you’d expect. Say your account converts 5% of clicks: 29 clicks should bring about 1.5. Zero is a warning. Zero from 4 clicks is not, because at 5% you wouldn’t expect one yet.

4. For Amazon, use your target ACOS. Amazon defines ACOS as ad spend divided by sales, shown as a percentage (Amazon Ads guide to ACOS). Multiply your average order value by your target ACOS, and you get the spend you can afford per sale. With a $40 average order and a 25% target, that is $10. A phrase that spent more than $10 with no sales has already missed your target. Cutting that spend is how n-gram negatives help lower ACOS: the sales stay, the wasted spend goes.

5. Be careful with small numbers. A phrase with three clicks tells you very little. In a small account, widen the date range instead of lowering your limit.

The n-gram analysis tool for Google Ads lets you type your own spend limit and falls back to your report’s average cost per conversion when you leave it empty. The n-gram analysis tool for Amazon Ads takes your target ACOS and a spend limit in the same way, and lists ASIN search terms separately for negative product targeting. Either way, you check the evidence before you act.

Turn phrases into negative keywords

The match type decides how much a negative blocks, and Google and Amazon work differently.

Google Ads

Google’s About negative keywords page defines three negative match types:

• Negative broad (the default, written plumber jobs): blocks searches that contain all the words, in any order.
• Negative phrase (written "plumber jobs"): blocks searches that contain the words in the same order, even with other words around them.
• Negative exact (written [plumber jobs]): blocks only that exact search, with no extra words.

One rule surprises people: negative keywords don’t match close variants. A negative for job doesn’t block jobs. If both forms waste money, add both. When you add negatives straight from the search terms report, Google sets them to exact match by default (Get negative keyword ideas), so add the symbols yourself if you want a phrase.

Where to put them matters too. A negative keyword list can hold up to 5,000 negatives, and an account can have up to 20 lists (About negative keyword lists). Lists suit words like jobs or free that are wrong for every campaign. Words that are only wrong for one campaign belong in that campaign.

Amazon Ads

Amazon’s negative keywords page offers only two types:

• Negative phrase: blocks shopping queries that contain the complete phrase or close variations. At most 4 words and 80 characters.
• Negative exact: in Amazon’s words, ads don’t show on queries that contain the exact phrase or close variation. At most 10 words and 80 characters.

Two differences from Google: Amazon’s negatives cover close variations, so you usually don’t need every plural, and Amazon says new negatives take up to 72 hours to take effect. You also can’t change the match type of an existing negative.

Single words or phrases? A one-word negative is the most powerful and the most dangerous. Use it only when the word is never relevant; otherwise use a longer phrase or exact negatives.

Check before you add them

Google puts it plainly on its negative keywords page: choose your negative keywords carefully, because too many can mean your ads reach fewer customers. Two checks catch most mistakes.

Look behind every phrase. An n-gram total can hide a good search. “Near me” may be wasteful in one ad group and your best converter in another. Before you add a phrase, read the search terms that make up its total. If any of them converted, block the bad searches only.

Test against your own keywords. Google warns that negatives which overlap your regular keywords stop your ads from showing (Add negative keywords). This is easy to do by accident with a single word: a negative repair added to a list would block your own keyword water heater repair. Shared lists and account-level negatives make it easier still, because they reach other campaigns. The negative keyword conflicts checker for Google Ads compares your negatives with all your keywords, using Google’s broad, phrase and exact rules, and lists every keyword a negative would block.

A monthly routine

New searches appear all the time, so treat this as a habit:

1. Download fresh reports once a month, or every two weeks for busy accounts. End the date range at least as long ago as your customers usually take to convert (7 or 14 days ago on Amazon).
2. Run the n-gram analysis with the limits you set from your own cost per conversion or target ACOS.
3. Add the obvious ones first: irrelevant words that will never convert.
4. Review the data-driven ones: read the search terms behind each phrase and pick the narrowest match type that stops the waste.
5. Check for conflicts with your keywords, then add the negatives to the right campaign or list.
6. Write down what you added and why. If conversions drop next month, you will know where to look.

Done every month, this keeps wasted ad spend from quietly building up again, without blocking the searches that are still on their way to becoming customers.