Google Ads Negative Keywords: Lists, Levels, Scripts and Account Hygiene

by Francis Rozange | Apr 4, 2026 | Google Ads

Introduction: negatives are the highest-ROI hygiene work in any account

Most Google Ads accounts leak budget through the same hole: search queries that should never have triggered an ad. WordStream’s long-running benchmark studies place wasted spend in untended accounts at 20 to 40 percent of monthly budget, with the bulk of that loss attributable to off-target search terms rather than bid mistakes or creative weakness. Tinuiti’s quarterly Digital Ads Benchmark reports tell the same story from a different angle: accounts that mine their search-terms report on a weekly cadence outperform accounts that do not by 22 to 31 percent on cost-per-acquisition.

Picture a paid online course at 600 dollars. Someone types “learn python for free.” Google decides the query is close enough to your bid on “learn python,” your ad shows, the click costs 4.20 dollars, and the user bounces. Tomorrow it happens a thousand more times. The keyword you bought worked exactly as designed. The negative keyword you never added is what failed.

This is the highest-ROI work in the account, and it is also the work that gets postponed first. Bidding strategies, creative tests, and audience layers all compete for attention. Negative keywords sit quietly in a sub-menu, waiting. Search Engine Land has covered this dynamic for years: every consultant audit they publish opens with the same finding, that negative coverage is thin, inconsistent, or absent.

Why negatives matter more in 2026 than they did in 2020

Three forces compound the cost of weak negative discipline. First, broad match has been quietly rebuilt. Since 2021 Google has pushed broad match as the default expansion mechanism, and Smart Bidding is now tuned to thrive on the volume broad delivers. The catch is that the same expansion logic surfaces queries you would never have bid on by hand. Without negatives, broad match is a budget vacuum.

Second, Performance Max changed the rules. From its 2021 launch through late 2023, Pmax accepted no negative keywords at all unless you opened a support ticket. Google quietly enabled account-level negative keyword lists for Pmax in 2023, and in March 2025 raised the per-list ceiling from 100 to 10000 entries, a change Search Engine Land documented as part of a broader concession to advertiser feedback. The capability exists. Most accounts have not used it.

Third, close variants have widened. Google removed exact match’s literal-only behavior in 2017, then loosened phrase match in 2021, and has continued to broaden semantic interpretation since. A bid on “running shoes” now matches “best sneakers for jogging” with no warning. Negatives are how you re-impose precision on a system that has been deliberately built to ignore your intent on the keyword side.

Optmyzr’s 2025 PPC State of Search report frames the practical impact in clear numbers. Accounts they audit show median wasted-spend recovery of 18 percent in the first 60 days after a structured negative-keyword overhaul, with no reduction in conversion volume. The mechanism is simple: spend that was producing zero gets redirected to keywords that already produced positive ROAS.

The three levels: account, campaign, ad group

Google Ads applies negatives at three structural levels. Where you place an exclusion determines its blast radius and the cost of getting it wrong.

Account-level negatives

Account-level negative keyword lists, accessible from Tools > Shared library > Exclusion lists, apply across every Search and Shopping campaign in the account simultaneously. The lists also extend to Performance Max since the 2023 update. Each list holds up to 5000 keywords, and you can apply one list per campaign at this level.

Use the account level for exclusions that have no exception anywhere in your business: terms that no campaign, present or future, would ever want to bid on. The classic universal blockers are job-seeking terms (“careers,” “salary,” “vacancy”), tutorial-seeker terms (“how to,” “tutorial,” “learn for free”) for premium-service businesses, and wholesale-grade negative noise like “free download,” “crack,” “torrent,” “wikipedia.”

The risk at the account level is asymmetric. A bad keyword on a single ad group costs that ad group reach. A bad keyword in an account-level list silently strangles every campaign you run. Ship account-level negatives slowly, with one-week observation windows between additions of more than ten terms.

Campaign-level negatives

Campaign-level negatives apply to every ad group in a single campaign. This is the right level for thematic exclusions: terms irrelevant to the campaign’s product or audience but potentially valuable in another campaign in the same account.

A law firm running separate campaigns for employment law, real estate law, and business law illustrates the case. “Salary” is a hard negative for the employment-law campaign (someone searching salary data is researching, not litigating) but neutral for real estate. “Rent” is a hard negative for real-estate-sale campaigns but central to a tenant-rights campaign. “Templates” is a hard negative for business-law transactional services but might be acceptable copy in a top-of-funnel content campaign.

Campaign-level negative keyword lists, like account-level lists, can hold up to 5000 entries. Most healthy campaigns sit between 30 and 200 specific terms at this level, in addition to the universal lists they inherit.

Ad-group-level negatives

Ad-group-level negatives are surgical. They block a term for one ad group only, leaving every other ad group in the campaign untouched. This level exists to resolve internal cannibalization and intent splits within a single campaign.

Example: a SaaS company runs an “Enterprise Demo” ad group and a “Self-Serve Free Trial” ad group inside one campaign. Both bid on “project management software.” The Enterprise ad group adds “free” as an ad-group negative because Enterprise sales does not want freemium-tier inquiries. The Self-Serve ad group does the opposite, adding “enterprise” and “for teams of 500” as negatives because those queries belong to the Enterprise group’s funnel.

Ad-group-level negatives should be rare. If you find yourself adding the same negative to four ad groups in the same campaign, promote it to a campaign-level negative. If you find yourself adding it across multiple campaigns, promote it to a shared list.

Match types for negatives: the rules differ from positives

Negative match types invert the logic of positive match types in ways that catch even experienced practitioners off guard.

Negative broad match

A negative broad match keyword (entered without quotes or brackets) blocks any search that contains all of the negative’s terms, in any order, with any other words present. Negative broad match does not expand to synonyms, plurals, or close variants. This is the single most important detail to remember about negatives.

If you add “-free shipping” as a negative broad match, your ad will not show for “free shipping,” “shipping free of charge,” or “the shipping is free.” Your ad will still show for “free returns,” “free delivery,” or “shipping costs,” because none of those queries contain both terms. The negative will also still show for “freedom shipping company” if such a query existed, because the negative blocks at the token level.

The non-expansion rule is what makes negative broad practical. You can add “free” as a negative broad and it will block “free trial,” “free download,” “free version,” and any other compound containing the literal word “free,” but it will not silently expand to “freedom,” “freely,” or “freemium.” You have to add those plurals and variants explicitly.

Negative phrase match

Negative phrase match (in quotes: “free trial”) blocks searches containing the exact phrase in the exact order, with any words allowed before or after but not in the middle. “free trial” blocks “best free trial,” “free trial software,” and “30 day free trial,” but not “free month-long trial” because a word interrupts the phrase.

Phrase is the workhorse type for compound concepts. “free shipping,” “do it yourself,” “open source,” “next to me” are all natural phrase-match negatives because the meaning lives in the word order and any wrapped variant is also irrelevant.

Negative exact match

Negative exact match (in brackets: [free trial]) blocks only the literal query, with no other words. [free trial] blocks “free trial” and effectively nothing else. Use it when you need to eliminate one specific high-cost zero-conversion query without disturbing related traffic. Common case: a single search term has accumulated 40 clicks at 6 dollars each with zero conversions, but the same noun appears in 12 converting variants. Block the exact, leave everything else alone.

The match-type hierarchy in practice

Mature accounts mix all three match types deliberately. Universal lists tend toward broad match for category-level exclusions (“free,” “job,” “salary”). Campaign-level lists lean phrase for compound concepts (“free trial,” “open source”). Ad-group-level lists use exact match for individual zero-conversion queries identified through search-terms mining.

One non-obvious rule: negative phrase or negative exact does not match queries with extra words inside the phrase. If you block “[shoe repair]” exact, “shoe repair shop near me” still triggers your ad because that query has extra terms. Use phrase, not exact, when in doubt.

Building a universal baseline list (50 to 200 entries)

Every account should ship its first campaign with a pre-populated universal negatives list. The list does not need to be perfect, only present. WordStream’s frequently cited starter list contains around 75 entries; Tinuiti’s published recommendation for B2B accounts hits 150; Optmyzr’s template runs to roughly 200. The right size depends on the business, but the categories are standard.

Tutorials and DIY

For premium-service or paid-product accounts, every tutorial-seeker is wasted spend. The standard block list:

tutorial
how to
guide
diy
do it yourself
step by step
youtube
video
walkthrough
instructions

Free, cheap, and freebie hunters

The single highest-impact category for paid-product accounts:

free
freeware
cheap
discount
coupon
promo code
cost-free
no cost
open source
crack
keygen
torrent

Job and career seekers

If you are not actively recruiting, every job-search query is wasted spend. This category is universal across almost every B2B and consumer account:

jobs
job
career
careers
hiring
recruitment
vacancy
salary
resume
cv
linkedin
indeed
glassdoor

Reviews and comparisons (case-by-case)

This category requires judgment. For high-consideration B2B SaaS where comparison shoppers convert, blocking “reviews” and “vs” is a mistake. For impulse-purchase consumer products where comparison traffic almost never converts, these queries waste budget. Test before deploying.

reviews
review
vs
versus
comparison
alternatives
alternative
compared to

Login, support, and existing-customer queries

Almost universal. People searching “[brand] login” are existing customers, not prospects. Bidding on these terms is fine for brand-defensive campaigns; the rest of the account should block them:

login
log in
sign in
account
password reset
support
customer service
contact
help

Adult, illegal, and brand-safety

Self-explanatory but routinely missed:

porn
xxx
nude
illegal
scam
fraud
complaint

Built from these six categories, a universal baseline list lands at 60 to 90 entries before any business-specific additions. Add 20 to 50 industry-specific terms and you reach the standard 100 to 150 entry universal list that most agencies ship with new accounts.

Automating discovery via the search-terms report

The universal list catches the obvious. The search-terms report catches the rest. Google Ads Help documents the report at Campaigns > Insights and Reports > Search Terms; it shows every actual search query that triggered an ad, with click, cost, and conversion data attached. This is the single most important screen in the account for negative-keyword work.

The weekly extraction process

The standard cadence runs four steps. Filter the search-terms report to the last 7 to 14 days, sort by cost descending, and apply two filters: clicks at or above 3 and conversions equal to zero. Review the resulting list, typically 40 to 200 rows in a mid-sized account.

For each row, ask three questions in order. Is this query relevant to my offer? If yes, leave it. Would blocking this query also block converting traffic? If yes, do not add the negative; refine ad copy or landing page instead. Can I improve copy to convert this traffic? If yes, test before blocking. If all three answers are no, add to the negative list.

The third step is pattern detection. Do not add 12 individual negatives if you can add one. If you see “free trial software,” “free trial download,” “free trial 30 days,” “free trial pricing,” and “free trial review,” the pattern is “free trial” as phrase match, not five exact-match negatives. Optmyzr’s n-gram analysis tooling formalizes this as a frequency-weighted token analysis: the highest-cost recurring tokens across your zero-conversion queries are your highest-leverage phrase-match negatives.

Threshold tuning

The threshold (clicks ≥ N, conversions = 0) shifts as the account matures. Search Engine Journal’s PPC management guides recommend 3 clicks as the new-account threshold and 5 to 10 clicks as the mature-account threshold. The principle: in a new account, you cannot afford to wait for statistical certainty before blocking; in a mature account, premature blocking risks killing high-value low-frequency queries.

Bulk import via CSV

For lists over 30 entries, manual entry is wasted time. Export the search-terms report, filter and clean in a spreadsheet, save as CSV, and bulk-upload to a shared negative list via Tools > Shared Library > Exclusion Lists > Add. The Google Ads Editor desktop tool handles bulk negative-list operations even faster, including paste-from-clipboard import.

Performance Max negative keyword lists

Performance Max is the campaign type most likely to leak budget on irrelevant queries because it is the campaign type with the least manual targeting control. Until 2023, account-level negative keyword lists did not apply to Pmax at all. Until March 2025, the per-list cap on Pmax negative lists was 100 entries.

Both restrictions are gone. Search Engine Land documented the March 2025 expansion to 10000 entries per list, framing it as Google’s response to consistent advertiser complaints. The capability is in production. The work to use it is not.

The Pmax negative-keyword setup

Account-level negative keyword lists are applied to Pmax campaigns through the same shared library used for Search and Shopping. Open Tools > Shared Library > Exclusion Lists, create or open a list, and apply it to the Pmax campaign in the campaign’s settings panel. Brand exclusions, the older Pmax-specific feature, sit alongside but do not replace negative keyword lists.

For Pmax, the highest-impact negatives are the universal categories already covered (jobs, free, tutorials, login) plus a tightly maintained brand-defensive list. Pmax’s lack of search-terms granularity makes prevention more important than reactive mining, because the query data you need to mine is partially obscured.

Pmax search-terms insights

Pmax does expose search-term data, but in aggregated “category” form via the Insights tab. The data is less granular than Search campaign reports and arrives with delay. Treat Pmax negative-list maintenance as quarterly work, not weekly: extract the search categories report, identify recurring irrelevant categories, and add aggressive phrase-match negatives. Re-evaluate every 90 days.

Negative scripts: automating auto-negation

Manual maintenance scales to roughly 100 to 200 negatives per week per analyst before it becomes the analyst’s full-time job. Beyond that volume, scripts are required. Google Ads Scripts is a JavaScript runtime built into the Google Ads UI; it executes against your account data without external infrastructure and supports scheduled execution.

Auto-negate non-converting search terms

The canonical script pattern: pull the search-terms report for the trailing N days, filter to clicks above a threshold and conversions of zero, add each surviving term as a negative to a designated shared list. The Google Ads developer documentation publishes a maintained reference implementation under “Common Negative List.”

The skeleton:

Define the campaign or campaign-list in scope
Define the time window (commonly 14 to 30 days)
Define the click threshold (commonly 5 to 10)
Pull SEARCH_TERMS_REPORT for the window
Filter rows where Clicks ≥ threshold AND Conversions = 0
Append each filtered query to a designated NegativeKeywordList
Email a summary of additions to the account owner

The script runs on a daily or weekly schedule. The email summary is non-negotiable: silent automation that mutates the account is how good campaigns die.

N-gram pattern detection

Brainlabs and Nils Rooijmans both publish open-source n-gram scripts that go one step further: instead of adding individual queries, they tokenize the zero-conversion queries, count token frequency, and identify high-frequency tokens whose presence correlates with non-conversion. The output is a recommendations list (“the token ‘free’ appears in 34 percent of your non-converting clicks”) rather than direct mutations.

This is the right pattern for accounts where a human still wants final approval on negatives. Run the n-gram script weekly, review the report, manually approve the top 5 to 10 patterns, deploy as phrase-match negatives.

Multi-account sync

Agencies and in-house teams running multiple accounts under one MCC use sync scripts to push universal negative lists from a master account to all child accounts. The mechanic: a single shared list lives in the master account, the script reads it daily, diffs against each child account’s mirror list, and applies adds and removes. Optmyzr ships this as a managed feature; Karooya and Adalysis offer similar tooling.

Common conflicts: when negatives hurt you

Aggressive negative-keyword work has known failure modes. Each one is recoverable, but only if you watch for it.

Blocking your own brand

The classic mistake: a negative meant to block competitors accidentally blocks your own brand. Add “-shopify” as a competitor block, but your campaign is for an “Acme Shopify Plugin” and now half your converting queries are blocked. The fix: bracket competitor blocks at the appropriate level (campaign, not account) and audit ad-copy keywords against the negative list before deploying.

Blocking translations and variants

“Free” blocks the English token. It does not block “gratis,” “gratuit,” “kostenlos,” or any other localized variant. International accounts must build localized negative lists per market. Search Engine Land has covered this gap repeatedly: agencies localize their ads, their landing pages, and their bids, then ship a single English-language negative list across 12 markets.

Over-negation killing reach

Symptom: impressions on your top campaigns drop 40 percent week-over-week with no obvious campaign change. Cause: an aggressive broad-match negative was added that blocks more than intended. “Training” added as broad-match negative kills “training wheels” for a kids-products account, kills “endurance training” for a fitness account, kills “training data” for an ML-tooling account.

The discipline: never add a broad-match negative without a 7-day observation window on impression volume. If volume drops more than 15 percent, the negative is doing damage and needs to be moved to phrase or exact.

Conflicts between positive and negative

Google Ads surfaces obvious conflicts in the UI under Tools > Diagnose Performance. The non-obvious conflicts slip through. A positive keyword in exact match for [project management software] coexists fine with a negative phrase “project management” because exact-match positives override phrase-match negatives only at exact-query level. Run a conflict audit monthly: download positive and negative keyword lists, compare for token overlap, investigate every overlap.

Negatives blocking close variants you wanted

You added “-free” as broad-match negative. It blocks “free trial,” “free shipping,” and “free version,” exactly as designed. It does not block “freely,” “freedom,” or “freemium” because negative broad match does not expand to close variants. This is the source of many “but I thought I blocked that” complaints. If you want to block close variants, add them explicitly.

Hygiene cadence by account maturity

The right cadence depends on account age and spend level. Tinuiti’s agency cadence guidelines map cleanly to three tiers.

New accounts (0 to 3 months)

Weekly search-terms review, mandatory. New accounts have no historical data, so every search-term insight is high-value. Spend 30 to 45 minutes per week. Click threshold of 3 (low statistical bar, but the cost of missed signals exceeds the cost of false-positive negatives). Add 20 to 50 negatives per week as a normal pace.

Mature accounts (3 to 24 months)

Bi-weekly to monthly review. The universal lists have been built, the obvious patterns are blocked, and the marginal negative is adding less value. Click threshold rises to 5 or 10. Spend 30 minutes monthly, plus quarterly deep audits.

Established accounts (24+ months)

Monthly review with quarterly deep audits. The deep audit reviews the existing negative list for stale entries: a “training” negative added in 2023 may now be blocking a paid training product launched in 2026. Schedule a 1-hour audit every 6 months that walks through every negative and asks “does this still apply?”

Annual deep audit

Once a year, every account benefits from a full negative-list audit by someone who did not build the lists. The audit asks four questions for every active negative. Is the negative still relevant? Is it at the right level (account vs campaign vs ad group)? Is it the right match type? Does it conflict with any current positive keyword? Most audits surface 5 to 15 percent of negatives that should be removed or moved.

Common mistakes

The mistakes pattern across accounts. Five recur often enough to call out explicitly.

Adding negatives only after launch, never before. Pre-launch research catches 60 percent of needed negatives before a single dollar is spent. Skipping pre-launch research means the first month is wasted at 30 to 40 percent loss rates instead of 10 to 15 percent.

Adding everything at exact match. Exact-match negatives are surgical and have low blast radius. They are also low-leverage. Adding “free trial,” “free trials,” “free-trial,” and “trial free” as four exact-match negatives is wrong. The right answer is one phrase-match negative for “free trial,” which catches all four variants and any other compound containing the phrase.

Never reviewing old negatives. Negative lists accumulate. Five years in, a list contains terms that were relevant in year one and irrelevant in year five. The list now blocks queries you would happily convert. The semi-annual audit cycle exists to catch this.

Treating Performance Max as exempt. Pmax accepts negative keyword lists since 2023. The accounts not using this feature are leaving 15 to 25 percent of Pmax spend on the table per Optmyzr’s quarterly analysis.

Skipping the email-summary on automated scripts. A script that adds negatives daily without notifying anyone is how good campaigns silently die. Always wire summary emails. Always review them, even if the review is cursory.

Conclusion: precision is leverage

Negative keywords look like maintenance work. They are leverage. The accounts running structured negative discipline outperform on every cost metric that matters, not because they spend more, but because they spend less on queries that were never going to convert. The savings recompound: budget redirected from zero-conversion queries to converting queries amplifies the return on every other optimization in the account.

The work fits in 30 minutes a week for a mature account, 60 minutes a week for a new one. The five steps are constant: build the universal list, mine the search-terms report, automate via scripts where volume justifies it, audit semi-annually for stale entries, and watch for the four common conflicts (own brand, translations, over-negation, positive-negative collisions).

If your account does not have a documented negative-keyword strategy, that is the single highest-ROI half-day of work you can ship this quarter. Start with a universal list, apply it to one test campaign, mine the search-terms report after 14 days, measure the lift on cost-per-conversion. The data will make the case for the rest.