The essentials
- Its role: stop an ad triggering on off-target queries, which eat the budget without ever converting.
- What is at stake in 2026: with broad match and automatic query expansion, 15 to 40% of the clicks in an account that is never cleaned are off-target.
- Three levels: account, campaign, ad group, plus the shared lists that can be reused across campaigns.
- Routine: read the search terms report every week for the first month, then every month; 15 minutes is enough.
A negative keyword is a term you exclude to stop your ads triggering on the queries that contain it. It is the main tool for controlling spend in Google Ads, and the most neglected: with broad match and the automatic expansion Google practises, an account that is never cleaned devotes 15 to 40% of its budget to clicks that will never convert. How match types work is detailed in keyword match types.
The three exclusion levels
| Level | Reach | Use |
|---|---|---|
| Account | All campaigns | Universally unwanted terms: “free”, “job”, “pdf”, insults |
| Campaign | A badly run | Products or services you do not sell, areas outside your patch |
| Ad group | One group | Partition groups from one another so they do not compete |
| Shared list | Several campaigns of your choosing | Reusable thematic lists: jobs, free, competitors, out of area |
Shared lists are the most practical tool for a small business: a “job hunting” list, a “free and second-hand” list, a “competitors” list, applied in one click to all the campaigns concerned, and enriched month after month.
The match types for negatives
- Negative broad match (a bare word, no symbol): excludes the queries containing all the words, in any order. “used shoe” blocks “used shoe men” but not “shoes used” if one word differs.
- Negative phrase match (in quotation marks): excludes the queries containing the exact phrase in that order, with words before or after.
- Negative exact match (in square brackets): excludes only the strictly identical query.
Three important differences from positive keywords: negatives do not take close variants into account, so singular and plural, spelling mistakes and accents must be excluded separately; a query is only blocked if it matches the rule exactly as set; and an exclusion that is too broad can block profitable queries, which makes a check compulsory after every bulk addition.
Standard lists by type of business
| Business | Families to exclude | Examples |
|---|---|---|
| All | Jobs, free, training, definitions, second-hand | job, recruitment, internship, salary, free, for free, pdf, definition, what is, second-hand, tutorial |
| B2B service | Consumers, self-teaching, competitors | private individual, personal, course, training, freelance if you are not one |
| E-commerce | Repair, parts, second-hand, hire | repair, spare part, manual, classified ads, hire, bad review |
| Local service | Towns outside the area, equipment, DIY | names of towns you do not cover, “how to”, “yourself” |
| Software | Free, piracy, irrelevant alternatives | crack, free licence, open source if you are not, download |
| Training | Qualifications and schemes you do not offer | individual training accounts if not eligible, apprenticeship, master's, university |
These lists are a starting point, never a replacement for the search terms report: the queries that really cost money are specific to each account.
The cleaning routine
- Open the search terms report: the Insights and reports menu, over the last 7 or 30 days depending on volume.
- Sort by descending cost: the expensive queries with no conversion are the priority target, not the most frequent ones; the metrics to watch are listed in the Google Ads KPIs.
- Decide on each one: exclude it if the query matches no offer; add it as a positive keyword if it converts and is not targeted; do nothing if the volume is too low to conclude.
- Choose the right level: the account for universally unwanted terms, the campaign for anything outside your offer, a shared list for recurring families.
- Check the effect: a sharp drop in impressions after an addition signals an exclusion that is too broad.
- Pace: every week during a campaign's first month, then every month. Fifteen minutes is enough once the routine is in place.
The precautions
- Do not exclude what converts: check the conversions before excluding a query that is expensive but profitable.
- Watch out for automated campaigns: Performance Max only accepts exclusions at account level or through a specific request; see Performance Max for a small business.
- Do not over-partition: too many cross-exclusions between groups end up preventing any serving at all.
- Document: keep the list of exclusions and the reason for each, otherwise no one will know why a query no longer serves.
- Revise: a new offer sometimes makes a query excluded two years earlier relevant again.
How GreenRed helps
Rather than juggling several tools, GreenRed's SEA tracking and optimisation brings these metrics together in a single dashboard, compares them over time and tells you which actions come first. You can try it free, with no card, from the Pricing.
Frequently asked questions
What are negative keywords for?
Stopping ads triggering on off-target queries that eat the budget without converting: searches for jobs, for free things, for definitions, for products you do not sell. In an account that is never cleaned, these queries account for 15 to 40% of clicks.
Where should exclusions be placed?
At account level for universally unwanted terms, at campaign level for offers you do not sell, at ad group level to partition, and in thematic shared lists reusable across campaigns, which are the most practical tool for a small business.
Do negative keywords handle variants?
No, unlike positive keywords. Singulars, plurals, spelling mistakes and accented variants must be excluded separately. An exclusion only blocks the queries that match the rule as set and its match type exactly.
How often should search terms be cleaned?
Every week during a campaign's first month, then once a month. Sorting is done by descending cost, not by number of impressions: it is the expensive queries with no conversion that cost money, not the frequent ones.