- Negative keywords do not match close variants. Casing and misspellings are handled for you, but plurals and synonyms are not, so "house" does not block "houses".
- Negative match types share names with positive ones and behave differently. Negative broad requires every term to be present, which makes it far narrower than people expect.
- Stop reading the search terms report term by term. N-gram analysis finds the pattern behind the waste in a fraction of the time.
- Build shared lists by intent theme, not by campaign, so one addition protects the whole account.
- Performance Max now supports campaign negative keywords at scale and exposes full search terms reporting, so the old excuse is gone.
The rule that costs accounts the most money
Positive keywords expand aggressively. Exact match now covers misspellings, plurals, stemming, reordered words and paraphrases with the same intent. Phrase match absorbed what broad match modifier used to do. Broad match interprets meaning rather than words.
Negative keywords do almost none of that. Google's own documentation is explicit that negative keywords do not match to close variants. Casing is ignored and misspellings are generally handled, but singular and plural forms, synonyms and related meanings are not. Adding course as a negative does not block courses. Adding job does not block jobs.
The asymmetry, stated plainly
Your positive keywords are matched by meaning. Your negative keywords are matched close to literally. That gap is where the waste lives, and it is why a list of eleven negatives has never protected anyone.
The practical consequence is that a negative list is not a set-and-forget artefact. It is a living document that grows every week, with plurals and synonyms added deliberately, driven by what the search terms report actually shows rather than by a list downloaded from a blog.
How negative match types actually behave
All three types exist, and each behaves differently from its positive counterpart. This matrix is worth internalising, because most over-blocking and under-blocking traces back to it.
| Type | Blocks when | Use it for |
|---|---|---|
| Negative broad | Every term appears in the query, in any order | Multi-word intent patterns like free quote template |
| Negative phrase | The exact word sequence appears somewhere in the query | Most intent negatives: "how to", "second hand" |
| Negative exact | The query is exactly that, and nothing more | Surgical exclusions where the broader term is valuable |
Single-word negatives behave the same across all three types, since there is nothing to reorder or contain. The distinction only matters from two words up.
Read the search terms report properly
Most people open the report, sort by cost, skim the top twenty, add four negatives and close it. That finds the expensive individual terms and misses the pattern that generated them.
Set it up once, properly:
- Ninety days, not thirty. Individual junk terms rarely spend enough in a month to look alarming. Over a quarter they aggregate into something obvious.
- Filter to zero conversions, then sort by cost. This is the waste list.
- Add the search term match type column. It tells you whether broad match is doing the damage or your phrase keywords are drifting.
- Include Performance Max. Full search terms reporting is available now, and it is often where the worst waste is hiding.
- Download it. The next step needs a spreadsheet.
You cannot see everything
Low-volume search terms are withheld for privacy reasons, so a meaningful share of queries never appears in the report at all. That is another argument for blocking patterns rather than individual terms: the pattern catches what the report does not show you.
Find the pattern with n-gram analysis
An n-gram is just a word or short phrase extracted from a longer query. Instead of asking which search terms wasted money, you ask which words wasted money across every query they appeared in. One hundred and forty three junk terms usually collapse into six or seven recurring words.
Doing it in a spreadsheet
You do not need a tool. Export the search terms report, then split each query into words and total the cost against each one. In Google Sheets, this formula gives you the total cost of every query containing a given word:
// A: search term B: cost C: conversions
// E: the word you are testing
=SUMIF(A:A, "*"&E2&"*", B:B) // total cost of queries containing the word
=SUMIF(A:A, "*"&E2&"*", C:C) // total conversions from those queries
=COUNTIF(A:A, "*"&E2&"*") // how many distinct terms contain it
// build the candidate word list from the queries themselves:
=UNIQUE(FLATTEN(SPLIT(A2:A500, " ")))Sort by cost descending, and the account tells you where it is bleeding in about five minutes. Anything with meaningful spend and zero conversions across a decent number of distinct terms is a negative keyword candidate.
Build the list architecture
Negatives applied campaign by campaign create an account nobody can maintain. Organise by intent theme in shared lists instead, so a single addition protects everything it should.
Four or five lists cover most businesses. The universal junk list travels with you between accounts; everything else should be built from the search terms report of the account it belongs to. Downloaded negative lists are scaffolding at best, since one advertiser's junk term is another's core product.
Apply them, then check they are still applied
A shared list attached to eight campaigns in 2024 is not attached to the four campaigns built since. This is one of the most common findings in an audit: good lists, quietly protecting less than half the account.
Cover the variants yourself
Because negatives do not expand, every exclusion needs its variants added explicitly. This is tedious and it is exactly the discipline that separates an account that leaks from one that does not.
| Variant type | Handled automatically? | What to do |
|---|---|---|
| Casing | Yes | Nothing. Jobs and jobs are the same negative |
| Misspellings | Generally yes | No value in adding typo variants |
| Singular and plural | No | Add both forms, every time |
| Synonyms | No | Add each one: course, class, training |
| Reordered or stemmed | No | Use negative broad where word order should not matter |
| Accents | No | Add accented and unaccented forms where relevant |
A practical habit: when you add a negative, immediately write its plural or singular partner underneath. It costs two seconds and closes the most common gap in the account.
Protect Performance Max and Shopping
Performance Max used to be the place negatives could not reach, which made it the place waste accumulated. That is no longer true. Campaign-level negative keywords are supported at substantial scale, negative keyword lists can be shared across Performance Max campaigns, and the full search terms report is available in the interface and through the API.
Two controls, two jobs:
- Negative keywords for query patterns. Free, jobs, DIY, second hand, and whatever else your search terms report reveals.
- Brand exclusions for brands. Google's brand exclusion control is built to recognise misspellings, variants, subsidiary brands and non-Latin script versions, which a literal negative keyword will not. Use it for your own brand and for competitor brands you have decided not to pay for.
The Performance Max guide covers how those exclusions fit alongside asset group structure and channel reporting.
Set a cadence you will actually keep
Negative keyword work fails through neglect rather than through error. A schedule proportional to spend beats an ambitious plan you abandon in March.
| Frequency | Task | Time |
|---|---|---|
| Weekly | Scan the last seven days of search terms, filtered to zero conversions, sorted by cost | 10 minutes |
| Monthly | Full n-gram analysis across 90 days, all campaign types including PMax | 45 minutes |
| Quarterly | Review the negatives themselves for over-blocking; confirm lists are applied to every campaign | 30 minutes |
The quarterly review matters more than it sounds. Negative lists accumulate, and a term that was junk two years ago may be a service you now offer. Blocked impressions are invisible, so nothing in the interface will ever tell you that a negative is costing you money.
What not to block
It is possible to negate an account into stagnation, and it happens more often than people expect. Broad match with automated bidding needs some exploratory volume to find pockets of demand you would never have written a keyword for.
- Do not block a word on the strength of one bad query. Check the word's total performance across every term containing it before excluding it.
- Be careful with informational language. "How to" is usually junk for a plumber and genuinely valuable for a software product with a self-serve trial.
- Watch for accidental breadth. A negative broad on
servicewill block a great deal you wanted, and you will never see what you lost. - Leave discovery room. If your search terms report contains nothing you did not expect, your targeting is probably too tight to grow.
The test before you add
Ask one question: could a customer who would genuinely buy from us type this? If the answer is yes, even occasionally, use a narrower match type or a longer phrase rather than blocking the word outright.
Frequently asked questions
Do negative keywords match plurals and misspellings?+
Misspellings and casing are generally handled automatically. Plurals, singulars and synonyms are not. Google's documentation states that negative keywords do not match close variants, so excluding "house" will not block "houses". Add both forms explicitly.
How does negative broad match actually work?+
It blocks a query only when every term in the negative appears somewhere in it, in any order. A negative broad of "running shoes" blocks "best running shoes" and "shoes for running", but not "running shoe" or "running trainers". It is much narrower than positive broad match.
Can I add negative keywords to Performance Max?+
Yes. Campaign-level negative keywords are supported at substantial scale, negative keyword lists can be shared across Performance Max campaigns, and full search terms reporting is available. For brands specifically, use brand exclusions instead, since they recognise misspellings and variants that a literal negative will not.
How many negative keywords should an account have?+
There is no target number, and chasing one leads to over-blocking. What matters is that the list is built from your own search terms report, organised into shared lists by intent theme, and reviewed on a schedule. Most accounts we audit have far too few; a small number have too many and are quietly suppressing real demand.
What is n-gram analysis?+
Aggregating performance by individual words or short phrases across every search term that contains them, rather than reading queries one at a time. It collapses a list of a hundred junk terms into the handful of recurring words that caused them, which is a far faster route to the right negatives.
Can adding negative keywords hurt performance?+
Yes, if you over-block. Blocked impressions are invisible, so an over-broad negative quietly suppresses demand with nothing in the interface to warn you. Check a word's total performance across all terms containing it before excluding it, and review your lists quarterly.
Where this fits
Cutting waste is usually the fastest measurable win available in an account, and it is almost always where an engagement should start. It is also the work that never finishes: search behaviour changes, match types drift, and every new campaign is a fresh surface for junk to accumulate on.
If you would rather have someone maintain this properly, our Google Ads consulting engagements include it as standard, and the free audit will quantify exactly how much your account is currently spending on queries that will never convert.