Google Ads
Negative keywords with AI: it drafts the list, you approve it
Negative keywords with AI: it reads your search terms, groups them by intent and drafts the list; you review and approve. Plus the mistakes that cut sales.
Also available in: Português · Español
It is Monday, the search terms report has 1,400 rows, and you have 20 minutes before the client call. Building negative keywords with AI turns that into a four-step job: the AI reads the period's search terms, groups them by intent, drafts a negative list with a match type for every entry, and you review it before anything is applied. The win is time: 1,400 rows become two dozen proposals in six buckets. The risk is the old one, only faster: a negative that is too broad cuts queries that were selling, and nothing in the account flags it. That is why the review stays with you, line by line.
The basics of reading the report by hand are in Google Ads search terms. This post is about what changes when an AI sits in the middle.
The workflow in four steps
1. The AI reads the search terms, with a threshold
"Search terms for the campaign Search – Plumbing over the last 30 days with cost above US$ 20, most expensive first. For each one, show the keyword that matched it and whether the term is already a keyword or already excluded."
The cost threshold drops the long tail of one-click queries, which eats your review time and saves pennies. The matched keyword matters because the junk usually comes from one or two broad matches. Knowing what is already excluded keeps the AI from proposing duplicates.
One limit that is not the AI's fault: as of October 2026, Google Ads Help says search terms without enough query activity are left out of the report for privacy reasons. Part of your spend has no visible query behind it, and the AI cannot see what Google does not show.
2. It groups the terms by intent
Grouping is where the AI saves the most time. Ask for each bucket with its total spend. Say, as an example, a plumbing company that only serves the Phoenix metro:
| Bucket | Example term | Monthly spend (example) | Proposal |
|---|---|---|---|
| Jobs and training | "plumber apprenticeship program" | US$ 260 | Exclude |
| Do it yourself | "how to unclog a drain without a plumber" | US$ 210 | Exclude |
| Product shoppers | "tankless water heater for sale" | US$ 170 | Exclude |
| Outside the service area | "emergency plumber tucson" | US$ 140 | Exclude |
| Parts | "toilet flapper replacement part" | US$ 80 | Exclude |
| Right intent, no conversion yet | "emergency plumber near me" | US$ 110 | Keep and watch |
The last row matters most. That is a buyer who simply has not converted yet, and excluding it cuts future revenue. Always ask the AI to separate "wrong intent" from "no conversions so far".
3. It drafts the list, with a match type for each entry
A useful list is not a column of terms. It is term, match type and reason. As of October 2026, Google Ads Help describes the three negative match types like this:
- Broad: blocks any search that contains all the negative's words, in any order.
- Phrase: blocks searches that contain the words in the same order, even with other words before or after.
- Exact: blocks only that exact query, with no extra words.
In the example, "apprenticeship" and "tucson" go in as broad, because they signal the wrong intent anywhere; "how to unclog" goes in as phrase; one specific bad query goes in as exact. The point the AI sometimes skips: negatives do not match close variants. Plurals and synonyms have to be added on their own ("apprenticeship" and "apprenticeships"). Misspellings and capitalization Google already handles.
4. You review, approve, and then it applies
The AI shows the final list, the campaign receiving it and the match type of each line. You strike what you disagree with, it redraws the list, and only after your yes does it apply. A week later, the follow-up question: "spend, conversions and CPA for the campaign in the 7 days before and the 7 days after the negatives added on the 6th."
The classic mistakes, and the request that prevents each one
AI makes the same mistakes as a rushed media buyer. The difference is that you can ask for the check before anything goes live.
| Mistake | How it happens | Request that prevents it |
|---|---|---|
| Excluding your own brand | The company or product name contains a common word, and that word becomes a broad negative | "Before proposing, check the list against the brand, the product names and the campaign's active keywords" |
| Broad match that is too broad | A single word like "cheap" also blocks "cheap emergency plumber phoenix" | "For every one-word broad negative, list the search terms that converted in the last 90 days and would be blocked" |
| Excluding on performance, not intent | A relevant term with little volume looks like waste | "Separate wrong intent from relevant terms with no conversions yet" |
| Forgetting plurals and synonyms | Excluding "class" lets "classes" through | "Add plurals and obvious synonyms for every broad negative" |
| Wrong level | A negative meant to split two ad groups lands on the whole campaign | Decide the level before asking; the AI applies at campaign level |
The first two are the expensive ones. Say the company is called Simple Plumbing and the AI proposes "simple" as a broad negative because of "simple way to fix a leaky faucet". Brand searches, the cheapest and most reliable clicks in the account, stop showing your ad, and you notice only when brand volume disappears. Overly broad negatives do the same thing to good queries, silently: the account spends less, sells less, and it looks like Google got worse.
Cross-checking against converted terms is the most valuable request on the list. The AI has the data to do in seconds what nobody does by hand.
Performance Max belongs in the cleanup too
As of October 2026, Google Ads Help says you can add negative keywords to a specific Performance Max campaign or at account level, and they apply to Search and Shopping inventory only. In an account that runs both Search and Performance Max, cleaning only the Search campaign often pushes the bad spend into the other one. Ask the AI for cost per conversion by campaign type, side by side, before and after the cleanup.
What the AI does through Vazante, and what stays with you
The connection itself is covered on the Claude connector page.
How often to clean
Weekly on a new account or one leaning on broad match; every two weeks or monthly on a stable one. Search term cleanup is one of the three audits in optimize Google Ads with AI; the other two cover budget caps and zero-conversion keywords. Ready-made prompts for the rest of the account are in Google Ads prompts, and the full method, from data to guardrails, is in optimize ad campaigns with AI.
To read search terms and apply negatives by talking to your account, see the plans: starting at US$ 39 per month, with the first 7 days for US$ 2.99.
Frequently asked questions
Can AI add negative keywords on its own?
It shouldn't. It reads the search terms and drafts the list; you review every line and approve. A bad negative quietly cuts sales, and nothing in the account warns you.
Which match type should a negative keyword use?
Exact to block one specific query, phrase to block a word sequence, broad for a word that signals the wrong intent in any context. When in doubt, go narrower.
Do negative keywords block plurals and synonyms?
No. As of October 2026, Google Ads Help says negatives do not match close variants, so plurals and synonyms must be added separately. Misspellings and capitalization are already covered.
Can I add negative keywords to Performance Max?
Yes. As of October 2026, Google Ads Help says you can add negatives to a specific Performance Max campaign or at account level, and they apply to Search and Shopping inventory only.