Google Ads
Negative keyword mining for Shopping and search
A weekly routine for finding wasted queries: waste patterns, n-gram analysis, match type traps and where each negative belongs in the account.
By CartKernel · Published
Negative keyword mining is a recurring routine, not a setup task. Queries change as the catalog changes, as seasons turn and as the auction shifts, so a list built once and left alone stops protecting the account within a few months. The routine has four steps: pull the queries, classify them, decide at the right level of aggregation, and place each negative where it will do the intended job and nothing more.
Done well it lowers wasted spend without narrowing reach. Done carelessly it removes converting traffic that nobody notices is gone, because blocked queries leave no trace in a report.
What mining is looking for
Not every query with no conversion is waste. The account has three genuinely different populations mixed together in the same report:
- Wrong intent. The shopper wants information, a repair, a job, a manual, a free version, or a product you do not sell. These will never convert regardless of bid or landing page.
- Right intent, wrong economics. The shopper wants what you sell, but the query attracts clicks at a price the product’s margin cannot support.
- Right intent, not yet enough data. The query looks poor because it has had a handful of clicks.
Only the first group deserves a negative on sight. The second group is a bid, budget or segmentation decision, and the third is a patience decision. Treating all three the same way is how accounts get quietly starved.
Set the cadence from spend, not from habit
| Account stage | Review frequency | What to review |
|---|---|---|
| First month of a new campaign | Twice weekly | Every query with a click |
| Steady state, larger spend | Weekly | Queries above a click threshold, plus the n-gram view |
| Steady state, modest spend | Every two weeks | Queries above a click threshold |
| Mature and stable | Monthly | N-gram view and top spenders only |
| After a catalog or seasonal change | Immediately | Everything, since new products attract new query shapes |
Put the review in a calendar with a named owner. Mining is the first task dropped when a week gets busy, and its absence is invisible until a quarterly report.
Classify against a fixed list of waste patterns
Working from a fixed classification is faster and more consistent than judging each query fresh. Seven patterns cover most of what a catalog account sees:
- Information and how-to. Queries asking how something works, what a term means, or how to make it.
- Repair, parts and maintenance, where the store sells whole products rather than components.
- Free, cheap, wholesale, bulk and secondhand, where those do not describe your offer.
- Employment and corporate. Jobs, careers, head office, complaints, contact details for something you are not.
- Adjacent categories. Products that share vocabulary with yours but are a different purchase.
- Wrong audience or region. Queries specifying a market you do not ship to, or a use case you do not serve.
- Named products you do not stock. Model numbers and product lines from ranges you do not carry.
Two more categories deserve their own handling rather than a blanket negative: competitor names, which some stores bid on deliberately, and your own brand terms, which belong in a structural separation rather than a waste list. That separation is covered in brand versus non-brand.
Use n-grams when individual queries are too thin
Most catalog accounts have thousands of queries with one or two clicks each, where no single query justifies a decision. Aggregate instead. Split every search term into its component words and word pairs, then sum cost, clicks and conversions against each fragment.
The output is a short list of words that consistently accompany spend without revenue. A single query containing “manual” might have two clicks. The word “manual” across the account might have a meaningful share of monthly spend and no orders, which is a decision you can make with confidence.
Do the same in reverse to protect reach: fragments that accompany most of your conversions are the ones a careless negative is most likely to block. Check every proposed negative against that list before adding it.
Match types, and the ones that overblock
Negative keywords behave differently from positive keywords, and the differences cause most accidental blocking.
- Negative broad match blocks a query only when every word in the negative appears somewhere in the query, in any order. It does not expand to synonyms, plurals or close variants, so plurals and misspellings must be added separately.
- Negative phrase match blocks queries containing the words in that order.
- Negative exact match blocks only the exact query.
Two practical rules follow. Single-word negative broad terms are powerful and easy to regret: a word like “kit” or “used” may appear in queries you want. Prefer phrase-level negatives for anything ambiguous, and reserve single words for terms that are unambiguous in your category. Second, because negatives do not match close variants, a serious list includes plurals, common misspellings and spacing variants explicitly.
Keep a dated log of every negative added, with the reason. When traffic drops later, the log is the first place to look.
Place each negative at the right level
- Account-level lists for waste that is wrong everywhere: employment queries, obviously unrelated categories, adult terms. Apply once, maintain in one place.
- Campaign-level lists for structural routing, such as keeping brand queries out of acquisition campaigns or keeping category queries out of a tier that should only serve generic terms.
- Ad group level only inside search campaigns where you are steering queries between tightly themed ad groups.
Shared lists are easier to audit than negatives scattered across campaigns. Name each list for its job, and record which campaigns it is applied to.
Shopping needs a different hand than search
Standard Shopping campaigns have no positive keywords, so negatives are the only query-side control available, and they carry more weight. A broad negative in a Shopping campaign can remove a whole family of queries across the entire catalog at once, since every product in the campaign is affected.
Two consequences. Be more conservative with negative breadth in Shopping than in search. And remember that negatives are also the routing mechanism when campaigns are separated by priority, which is a different job from waste elimination and needs its own lists, described in query sculpting across Shopping campaigns and standard Shopping campaigns.
Broad-coverage campaign types expose less query detail than Standard Shopping and search do. Use the account-level and brand controls the campaign offers, check what your account currently supports rather than assuming, and accept that the query view will be partial.
Keep a watch list instead of blocking early
Not every suspicious query should be blocked this week. Maintain a watch list of terms that look wasteful but have too little data, review it at the next cycle, and promote a term to the negative list only once the pattern holds across enough clicks to be believable. This single habit prevents the most common mining error, which is blocking a term during the week it happened to perform poorly.
Set the threshold from economics rather than from a round number. A product with thin margin justifies blocking after fewer wasted clicks than a high-margin one, and the break-even ROAS calculator gives the figure to reason from.
Measure the effect honestly
After each cycle, record wasted spend removed, the change in cost per conversion for the affected campaigns, and the change in impression share. The third number is the safety check: a large drop alongside a modest efficiency gain usually means the list went too far.
Rising costs are not always a waste problem. If cost per click is climbing across well-qualified queries, the cause is more likely competition or feed quality, and Shopping CPCs rising and how do you lower CPC in Google Shopping work through those. Match type strategy on the positive side is in should ecommerce search campaigns use broad match, and the mining routine runs as part of Google Ads.