Negative Keyword Builder

Adding negatives is easy. Adding negatives that block waste without silently killing converting traffic is the actual skill, and it's where most negative lists go wrong — someone blocks cheap because it appears in junk queries, not noticing that "cheap flights to lisbon" was their best converting query all quarter.

Two things make a good negative list: the right match type (so one entry does the work of fifty), and a safety check against converting queries (so it doesn't cost more than it saves).

Getting the data

You need the full search terms report — not just the bad queries. The converting queries are what makes the safety check possible, so if someone pastes only the junk, ask for the complete export.

1. Connected account — pull search terms for the last 30–90 days. Longer is better here; a query that converted twice in 90 days still matters.
2. File or paste — CSV/TSV/Excel from Google Ads → Campaigns → Insights & Reports → Search terms → download. Skip Google's title rows and find the header row containing Search term.
3. Already have verdicts — if the search-term-miner skill (or the person) has already labeled terms to exclude, take that list but still ask for the full report to run the safety check.

Also establish: which campaigns these apply to, and whether they use a shared negative list already. Adding to the wrong level is a common and confusing mistake.

How to build the list

1. Find the themes, not the queries

Tokenize every wasted query — zero conversions, spend above the volume floor — into words and 2–3 word phrases (n-grams). Count how much spend each token appears in across all zero-converting queries. A token appearing in many wasted queries with meaningful spend is a theme — one phrase negative kills the whole theme, now and in future.

This is the difference between a list of 60 exact negatives that goes stale next week and a list of 6 phrase negatives that keeps working.

Quantify "meaningful" before you trust a theme. A token clears the bar when its zero-converting queries total at least 1× target CPA in spend, or ~15 clicks if no target is known — the same floor search-term-miner uses, applied to the token's aggregate rather than any single query. Below that, zero conversions means no data, not bad intent: list it as watch-later and re-check next month.

Also discard structural tokens that carry no intent — articles, prepositions, and the head term itself. "for", "the", "best", and your own product noun will top any spend-weighted n-gram count and negating them would blank the account.

2. Safety-check every candidate — this is the step that matters

For each candidate negative, scan the entire report (including converting queries) for terms it would also block. A negative blocks any query containing that token, forever, silently.

• If it would block queries with zero conversions only → safe, recommend it.
• If it would block any query with conversions or revenue → do not recommend it as-is. Either narrow it to an exact negative on the specific junk query, or flag it to the person with the exact converting query and revenue at risk and let them decide.

Report the at-risk revenue explicitly. "free would block 'free consultation' which drove 12 conversions and $4,100" is the sentence that prevents an expensive mistake.

3. Choose the match type deliberately

Default to phrase for themes and exact for one-offs. Broad negatives are the most likely to cause collateral damage; use them only when the token is unambiguous (like a competitor brand or jobs).

Match types are not universal. Broad negatives exist only in Google Ads — Microsoft Ads and Amazon Ads support negative exact and negative phrase only, so never propose a broad negative there. Amazon also has no shared negative list: negatives live at campaign or ad group level, so ask which campaigns to apply to and output per-campaign blocks instead.

4. Expand every negative into its word forms

On Google Ads and Microsoft Ads, negative keywords do not match close variants. A negative blocks only the literal word forms you type — no plurals, no singulars, no misspellings, no stemming. Negative phrase "salary" does not block "salaries". Negative exact [running shoe] does not block "running shoes". This is the opposite of how positive keywords behave, and it's the single most common reason a list looks complete and still leaks spend. (Amazon differs — it does apply close variants to negatives, so don't expand there.)

For every negative you propose, add the singular and the plural as separate lines, plus any misspelling that actually appears in the report with spend against it. Don't invent misspellings — take them from the data. Each variant is its own line, and each still goes through the safety check in step 2.

5. Pick the level

• Ad group — the term is fine for the account but wrong for this ad group (classic for tightly themed groups stealing each other's traffic).
• Campaign — wrong for this campaign's goal.
• Shared list — universally wrong for the business (job seekers, DIY, competitor brands, other product categories). Most negatives belong here, and a shared list is what stops the same waste reappearing in the next campaign someone builds.

What to produce

A paste-ready block plus the reasoning. Ad platforms accept one entry per line, so the person can copy the block straight into the negative keyword box.

NEGATIVE KEYWORD LIST — [account], built from [n] terms / [date range]
Blocks [total wasted spend] across [n] queries · [projected monthly saving]
Safety check: [n] candidates rejected for touching converting traffic

── SHARED LIST: "Universal exclusions" ──
"salary"
"jobs"
"how to"
[…]

── CAMPAIGN: [name] ──
"free"
[…]

── AD GROUP: [name] ──
[exact query]
[…]

Then, briefly:

• Why each theme — one line per entry with the spend it blocks.
• Rejected candidates — what you did not recommend and the converting query that saved it. This section builds more trust than the list itself.
• How to apply — Tools → Shared library → Negative keyword lists → + → name it → paste the block → Save, then tick the list and choose Apply to campaigns, selecting every campaign it should cover. A saved list that isn't applied to a campaign blocks nothing, and this is the most common way a negative list ends up doing no work at all. Campaign- and ad-group-level entries go somewhere else: Campaigns → [campaign] → Audiences, keywords and content → Search keywords → Negative keywords. Changes take effect going forward only.

Judgment worth applying

The projected saving is a forecast, not a promise. Blocked spend usually doesn't vanish — it reallocates to other queries. Say "roughly $X/month of clearly wasted spend stops" rather than promising a saving that lands in the bank.

Over-negating is a real and invisible failure. An account with thousands of negatives, accumulated over years by different people, will starve. If the list is already large, say so and recommend auditing existing negatives before adding more.

Some "waste" is upper funnel. Informational queries can be genuinely bad, or they can be a top-of-funnel play someone is running deliberately. Ask if there's a content or awareness goal before excluding the whole research-intent category.

Never apply to a live account without approval, even with write access. Present the list, let the person review it, then apply only what they explicitly approve. A wrong negative removes traffic silently — there is no error message, just a slow decline nobody attributes to this change three months later.