How to fix: Search terms triggering ads on irrelevant queries

TL;DR

Your Search campaigns are paying for clicks on queries that share words with your keywords but carry a different intent — "running shoes" picking up "shoe repair", "crm software" picking up "criminal record management". The fix is a Search terms report cleanup: tag the mismatched intent stems, push them into negative keyword lists, and tighten match types on the keywords that produced the bleed. Each cleared mismatch removes a direct cost leak and one bad training signal from Smart Bidding.

This rule is narrower than [[fix-wasted-spend]] (which targets all zero-conversion terms regardless of intent) and broader than [[fix-negative-keywords-missing]] (which targets gaps in your negative keyword strategy as a tactic). Irrelevant search terms are the symptom both other rules try to prevent.

Why it matters

The mechanism is straightforward and well-documented in the Google Ads Search terms report itself: keywords (especially broad and loose phrase match) are matched against semantic neighbors, related concepts, and reformulations of the user's query, not just the literal keyword text [1]. When a keyword's semantic neighborhood overlaps with a different industry, profession, or use case, the auction system happily serves the ad on those queries — they look "related" to the model even when they are unrelated to your business.

The magnitude has two layers. The visible layer is direct spend leakage: on audited Search accounts, irrelevant queries typically account for 10-25% of broad-match spend and 3-8% of phrase-match spend over a 90-day window [2]. At a $20,000/month Search budget, the lower end of that range is still $2,000-4,000 of monthly spend on queries that cannot convert because the user is shopping for a different product entirely.

The hidden layer is signal corruption. Smart Bidding strategies (tCPA, tROAS, Maximize Conversions, Maximize Conversion Value) learn from every click-and-no-conversion event, not just from conversions [3]. Each irrelevant click teaches the model that the auction context that produced it — device, time, geo, audience signals, query embedding — is low value. The model then bids down on similar contexts in the future, including legitimate ones that happen to share a feature with the noise. Smart Bidding cannot tell the difference between "this query is irrelevant to your business" and "this query is relevant but the user didn't convert today"; both look like negative training examples. Cleaning irrelevant terms therefore compounds: you stop the bleed and you let the model re-learn on a cleaner signal.

The reason it persists in real accounts is two structural patterns. First, Google's 2024-2025 push toward broad match plus Smart Bidding [4] surfaces a much wider semantic neighborhood per keyword than legacy match types — accounts that migrated to broad match without a negative-keyword review process now have far more surface area to leak. Second, Search terms reports are weekly hygiene tasks that get skipped when the account looks profitable in aggregate; aggregate ROAS hides the 10-20% of spend that is structurally lost.

How to fix

1. Open the Search terms report (Campaign → Insights and reports → Search terms). Set the date range to the last 30-90 days depending on traffic volume — go wider on accounts with low daily query counts so the sample is meaningful.
2. Add the right columns. Default columns hide the diagnostic signal: add Conversions, Conv. value, Cost / conv., and Match type. Sort by Cost descending — the top 20-50 rows usually contain 60-80% of the bleed.
3. Tag mismatched intent rows. Walk the top rows and mark each as either relevant (correct intent, just didn't convert today — leave alone, this is [[fix-wasted-spend]] territory if it persists), partially relevant (right industry, wrong product — refine match type or add a more specific negative), or irrelevant (different industry, profession, or audience entirely — negative keyword now).
4. Add negatives at the correct level. For a stem isolated to one ad group, add as ad-group negative. For stems that recur across ad groups inside the same campaign, add as campaign negative. For stems that recur across campaigns with the same product scope, add to a shared Negative keyword list (Tools → Shared library → Negative keyword lists) [5] and attach to all relevant campaigns.
5. Choose the right negative match type. Use negative exact when only the specific query is wrong ("[shoe repair]"). Use negative phrase when a stem is universally wrong ("repair"). Use negative broad only sparingly — it can over-block legitimate close variants. The default in the UI is negative broad; switch to phrase or exact deliberately.
6. Tighten the source keyword if needed. If a broad-match keyword is producing repeated mismatches across many ad groups, that keyword is the wrong tool. Either tighten to phrase or exact match, or pause it and rebuild with the highest-converting query as a new exact-match seed inside its own ad group. Broad match without aggressive negative hygiene plus value-based Smart Bidding (tROAS with revenue tracked) is not a stable configuration.
7. Re-pull the report 14 days later. Irrelevant cost share should drop below 5% on the next audit. If it does not, the negatives were too narrow — repeat with deeper stems, or restrict match types further. Allow Smart Bidding 7-14 days to re-learn if the cleanup shifted more than 20% of historical impressions.

Common mistakes

• Adding only negative exact when the stem is universally wrong. "[shoe repair]" stops only that exact query; "shoe repair near me", "shoe repair shop", "shoe repair cost" continue to bleed. Use negative phrase for stems that are universally off-topic.
• Adding negatives at the account level reflexively. Account-level negatives propagate everywhere — including campaigns where the stem is legitimate (a real shoe repair shop running ads). Default to campaign or shared list; account level is for site-wide brand protection and obvious off-topic stems only.
• Reviewing only zero-conversion terms. Some irrelevant terms convert occasionally by accident (wrong-product user mis-clicks add-to-cart). They are still irrelevant to your business goal and still teach Smart Bidding the wrong context. Filter on intent mismatch, not just on Conversions = 0.
• Negativing the converting query while keeping the broad-match keyword that found it. Counter-intuitive but common: a broad keyword finds a high-converting variant; the merchant adds the variant as a positive exact-match keyword and negatives it elsewhere — but leaves the broad keyword unchanged, so the same broad keyword now competes with the new exact-match keyword and dilutes its share. Either tighten the broad keyword or remove it.
• Treating the Search terms report as a one-time cleanup. Query distributions drift weekly as seasonality, news, and competitor activity shift demand. Schedule a 15-minute recurring review at the campaign level.

FAQ

Is this the same as adding more negative keywords? Partially. [[fix-negative-keywords-missing]] is the tactic — building and maintaining a negative keyword strategy. This rule is one of the signals that the negative strategy is currently failing: irrelevant queries are slipping through. Fix this rule's findings by applying the negative keyword tactics described in that sibling article.

Should I use Google's auto-recommendation to add negatives? Selectively. The Recommendations tab can surface obvious mismatches, but never enable "Auto-apply" for negative keyword recommendations [6] — the system will sometimes block close variants of converting queries. Review individually before applying.

My broad-match campaign uses Smart Bidding. Doesn't the algorithm handle this? Smart Bidding optimizes for the conversion goal you set; it does not know whether a query is relevant to your business — only whether the auction context historically converted. If irrelevant queries occasionally convert by accident (wrong-product clicks, attribution errors, cross-product bundles), Smart Bidding will keep bidding on them. Manual negative hygiene is still required.

Will adding negatives trigger a learning phase reset? A change shifting <20% of historical impressions typically does not trigger a full reset, only a partial re-learn over 3-7 days. Larger cleanups (>20% impression shift) can trigger a 7-14 day learning phase. Plan cleanups in stages if the account is sensitive to volatility.

What about Performance Max — does this rule apply? PMax has its own Search terms insights (limited compared to Search campaigns) and brand exclusions. Irrelevant queries in PMax are handled via Brand exclusions and Account-level negative keyword lists (rolled out 2023-2024); the workflow in this article applies to Search campaigns specifically. PMax-specific findings are flagged separately.

Sources

1. Google Ads Help — Search terms report. Official documentation explaining how the Search terms report surfaces queries that triggered ads and the recommended workflow for adding negative keywords.
2. Search Engine Land — PPC library. Industry coverage of PPC trends, including reporting on broad-match query expansion and irrelevant-term share in audited accounts.
3. Google Ads Help — About Smart Bidding. Explains how Smart Bidding uses auction-time signals and conversion data to set bids, which is the mechanism by which irrelevant clicks corrupt the model.
4. Google Ads Help — About keyword matching options. Reference for broad, phrase, and exact match behavior and the trade-offs between reach and precision.
5. Google Ads Help — About negative keywords. Official guidance on negative keywords, match types for negatives, and account-level vs campaign-level application.
6. Optmyzr — Negative keywords for Google Ads. Industry guide to negative keyword strategy, match-type selection, and the risks of auto-applying recommendations.