For fifteen years negative keywords were treated as maintenance: an occasional pass over the search terms report, a few exclusions, move on. That era is over. In an account where broad match is the default and AI Max expands queries further, the keyword bounds nothing: it steers. Exclusions are one of the few remaining instruments for telling the system where not to go.
The three negative match types, used as intended
Most accounts use one type, usually exact, then wonder why waste returns every month. The three do three different jobs:
| Type | What it is for | Example |
|---|---|---|
| Negative exact | Removing a single variant without touching its neighbours | [free crm] excluded while "free crm for 30 days" stays live |
| Negative phrase | Removing families of queries tied by a modifier | "how to" to cut informational intent |
| Negative broad | Disqualifying words that change intent wherever they appear | free, DIY, salary, jobs, reviews |
The technical trap. Negative match types do not behave like positive ones: they do not match close variants. Excluding "insurance" does not exclude "insurances", nor a misspelled version. Lists have to be built with plurals, singulars, abbreviations and the most common typos spelled out one by one.
List architecture: three, not thirty
Account-level shared negative lists are the only structure that survives growth. The split that works almost everywhere:
- Global disqualifiers. Words that rule out commercial intent in any campaign: free, DIY, course, jobs, salary, pdf, torrent. Applied account-wide, always.
- Competitors. Applied or removed per campaign, deliberately. On a conquesting campaign you want them in; on a generic campaign almost never.
- Brand. Your brand and its variants, excluded from generic campaigns so broad match does not cannibalise the brand campaign at higher CPCs.
Above those three sit campaign- and ad-group-specific exclusions: those come from the data and change weekly.
The brand boundary problem under AI Max
Campaign-level brand inclusion and exclusion settings are a step forward, but people analysing search terms under AI Max keep finding the same leakage: brand queries matching non-brand keywords, non-brand keywords matching competitors, and brand misspellings escaping the filters. The practical conclusion from the field is that brand controls need solid negative lists alongside them, not instead of them.
Cadence: where this is actually won
The difference between a clean account and one bleeding budget is not list size. It is how often somebody looks at the search terms that spent money.
| Frequency | What you look at | Time required |
|---|---|---|
| Weekly | Terms above a spend threshold with zero conversions | 15 minutes |
| Monthly | Emerging themes, shared list updates | 1 hour |
| Quarterly | Conflict review: negatives blocking good queries | 2 hours |
The third pass is the one nobody runs and the one that costs most. A negative list accumulated over three years always contains exclusions that now block traffic which would convert. Google flags conflicts between negative keywords and active keywords: worth reading before you add more.
Three mistakes that cost more than the waste they prevent
- Excluding everything that has not converted yet. A query with twelve clicks and no conversions is not statistical evidence, it is noise. Set a threshold — twice your target CPA in spend, say — and act only above it.
- Using broad negatives on ambiguous words. "cheap" disqualifies intent in many categories and is exactly your customer's search in others. Lists must be read in your market's vocabulary, not copied from a template.
- Handing the whole job to an automation. Tools that suggest negatives are useful for ranking candidates; human approval is where you avoid excluding a profitable family of queries. The hybrid — AI proposes, a person approves — is what holds up.
When negatives are not the answer
If the same wrong theme returns every week despite exclusions, the problem is not the list: it is the structure. Three signals and their fixes:
- An ad group collecting queries from different intents → split the themes. The same work improves ad relevance, as we explain in the piece on Quality Score.
- A landing page that is too generic → the system infers scope from page content too, especially in campaigns with automatic expansion.
- A campaign covering too many services at once → splitting by commercial intent beats any list.
Negative keywords, in 2026, are less housekeeping and more a statement of scope: the one place an account gets to tell the system what its market is not.
If you want to know how much of your spend lands on out-of-scope queries, the free audit measures it against your last 90 days of search terms. You can also see how we work and which services cover this ground.