First, confirm what matched
Open the search terms report for a period long enough to be representative and add the columns that show which keyword triggered each query and its match type. Sort by cost, not by impressions, because a few expensive mismatches matter more than many cheap ones. For each irrelevant query, note three things: the triggering keyword, that keyword's match type, and the campaign. Queries without a triggering keyword in a search campaign, or appearing in Performance Max search term insights, point to a different route than those matched by a named keyword.
Decide what "irrelevant" means for this business before acting. A research query from a buyer early in their decision can look irrelevant and still convert later; a query for a competitor's support line never will. Mark each term as wrong product, wrong audience, wrong intent, or adjacent but useful. If the real worry is that clicks arrive but nobody buys, read the clicks without conversions diagnosis as well, since query quality is only one of its causes.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Loosely related queries all trace to broad-match keywords | Match type column in search terms; bid strategy in use | Broad match with weak signals reaching loosely related queries | Pair broad with Smart Bidding and strong signals, or tighten to phrase or exact |
| The same unwanted theme recurs across campaigns | Negative keyword lists, levels and match types | Negatives missing, at the wrong level, or with too narrow a match type | Add the theme to a shared list at the right level and match type |
| One ad group catches queries for two different needs | Keyword list and ads of the ad group | Ad group mixes intents, so the signals it sends are blurred | Split by intent; give each group its own ads and landing page |
| Search terms you never targeted appear under Performance Max | Search term insights; overlap with search campaigns | Performance Max or AI Max search coverage overlapping search campaigns | Use exact-match search campaigns for priority terms; add exclusions where supported |
| Matches drift toward cheap, low-intent queries after a goal change | Primary conversion actions and their values | Conversion goals rewarding cheap low-intent actions | Make the real business outcome primary; demote soft actions |
Fix, verify and prevent each cause
Broad match with weak signals
The help centre describes broad match as reaching searches related to the keyword and using signals, including the landing page and other keywords in the ad group, to judge relevance. When those signals are thin, for example a generic landing page or a single vague keyword, the system has little to anchor on, and "related" becomes very loose. Broad match also depends on the bid strategy: without automated bidding it has no conversion feedback to tell good matches from bad.
Fix: strengthen the signals by giving the ad group a focused landing page and a coherent keyword set, run broad only alongside a conversion-based bid strategy, and fall back to phrase or exact where volume allows.
Verify: the share of cost on off-topic queries from those keywords falls in the next review cycle.
Prevent: introduce broad match one ad group at a time with a review date, following the rules in the match types guide.
Missing or misplaced negatives
A negative at the ad group level will not protect a sibling ad group; a phrase-match negative will not block a reordered variant; a list built for one campaign may never have been attached to the new one. The help centre is explicit that negative keywords do not automatically include close variants such as plurals and misspellings.
Fix: add recurring unwanted themes to a shared list, choose the match type that blocks the full theme without catching wanted queries, and attach it to every campaign where the theme should never appear.
Verify: the theme stops appearing and no wanted query disappears with it.
Prevent: run the review workflow in negative keywords, which also covers how to detect over-blocking.
Ad groups mixing intents
If one ad group holds keywords for repairs and for new installations, its ads and landing page speak to neither precisely, and broad or phrase keywords borrow each other's context. Queries for one need get matched by keywords written for the other.
Fix: split the ad group by intent, so each has keywords, ads and a page about one thing.
Verify: each new ad group's search terms stay on its own theme, and relevance ratings improve where they were low.
Prevent: design groups around the customer's need, not around a list of words, as set out in the account structure guide.
Performance Max overlapping search
Performance Max can serve on search inventory, and the help centre documents precedence rules: an exact-match keyword identical to the query takes priority, and otherwise the campaign with the higher Ad Rank serves. A Performance Max campaign with broad asset groups can therefore pick up queries your search campaigns were not built for, and AI Max features in search campaigns widen matching in a similar way.
Fix: cover priority queries with exact-match keywords in search campaigns, apply brand exclusions and account-level negatives where supported, and tighten asset group themes and final URL settings.
Verify: search term insights show fewer unwanted categories, and the search campaigns regain their core queries.
Prevent: check overlap whenever a Performance Max campaign launches or changes, using the guidance in Performance Max structure.
Goals rewarding low-intent actions
Automated bidding pursues whatever is marked as a primary conversion. If a page view, a brochure download or a short visit counts, the system learns that cheap, curious queries are a success and seeks more of them; the traffic gets worse while reported conversions look healthy.
Fix: make the outcome the business actually values primary, give soft actions a secondary role or a low value, and let the strategy relearn.
Verify: the conversion mix shifts back toward real leads or sales, and the query mix follows within a few weeks.
Prevent: review primary actions before any bidding change, as described in the measurement hub.
Prevention: a search terms review routine
Irrelevant matches never stop entirely, so prevention means catching them while they are cheap and spotting when their volume signals a structural cause. Review search terms on a fixed rhythm: weekly for new or broad-match campaigns, less often for mature exact-match ones. Sort by cost and read the top of the list, not all of it. Classify each unwanted term by theme and ask, each time, whether it points to one of the causes above; three similar mistakes in a month usually mean a structural fix is due, not three more negatives. Similar mistakes spread thinly across many cheap queries are easier to see in aggregate, which the n-gram analysis of search terms provides by totalling cost and conversions for each recurring word and phrase. Keep shared negative lists named by theme so anyone can see what they block. After every launch, bid strategy switch or goal change, read the search terms within days rather than at the next monthly review. Matching quality also affects price, because low relevance weighs on Quality ScoreGoogle's 1–10 keyword diagnostic. and so on what each CPCAverage cost of each click. costs, which makes this routine as much a cost control as a cleanliness one in any SEMPaid search advertising (sometimes used to include SEO). account.
A garden centre paying for houseplant care questions
A garden centre runs broad-match keywords such as "buy indoor plants" with target CPA bidding. A review finds that 22% of the month's cost of 2,400 went on queries such as "why are my monstera leaves yellow". The matched keywords sit in one ad group alongside "plant care kits", and the primary conversion includes a care-guide download. The centre splits the ad group in two, demotes the download to secondary, and adds a phrase-match negative list for "why", "yellow leaves" and "how to". The next month, care questions take 6% of cost and plant orders rise from 41 to 53.
If irrelevant searches trace back to broad match running without the conditions it needs, the comparison of broad match and exact match explains when each match type should be the default.
Common mistakes
Adding negatives one query at a time without asking why the query matched. Switching all keywords to exact match and losing the volume the account needed. Counting soft actions as primary goals and blaming broad match for the result. Ignoring Performance Max when reviewing search coverage.
