Why Google Ads Shows Irrelevant Search Terms

Google is matching meaning, not words

Since 2021, broad match does not look for your keyword inside the search. It looks for a search that means roughly the same thing. Google's own documentation says broad match considers the meaning of the search term, the meaning of the other keywords in the ad group, the content of your landing page, and the user's recent search activity.

Read that list again and the irrelevant terms stop being mysterious. A plumber bidding on plumber has a landing page full of the word plumber. The ad group is full of plumbing keywords. So when someone searches "plumber salary uk" or "plumbing apprenticeship near me", the matching model sees a query about plumbers, an advertiser about plumbers, and a landing page about plumbers. Three out of three. It serves the ad.

Nothing in that chain encodes the one fact that matters: the person typing "plumber salary uk" will never buy a boiler repair. Commercial intent is not a property of the words. It is a property of the person, and the model is inferring it from signals that are often too weak to carry the distinction.

This is why the usual framing — "Google is wasting my money on purpose" — leads people to the wrong fix. The matching is not random and it is not malicious. It is systematically biased toward topical similarity, because topical similarity is the thing it can measure reliably. You are the only party in the transaction who knows that job seekers are worthless to you.

Exact match is not exact either

If your response to broad match is "fine, I'll use phrase and exact only", you will cut the volume of strange terms but not the mechanism. Close variants apply to all three match types. Exact match matches searches with the same meaning as your keyword, not the same string.

So [running shoes] picks up "shoes for running", which is fine, and also picks up plurals, misspellings, reorderings and rewordings that Google judges to have identical intent. Phrase match is looser again: the meaning of your keyword has to be included in the search, but extra words can flip the intent entirely. "office chair" matches "office chair repair near me" and "how to fix office chair gas lift" — both contain the meaning, both are people who already own a chair and want to keep it.

The practical consequence: moving to exact match reduces the number of distinct wasted terms, at the cost of reach, and does not remove the need for negatives. Accounts running pure exact still accumulate junk. It just arrives slower.

Semantic closeness is not commercial intent

The terms that waste money cluster into a small number of families, and the same families show up in almost every account we audit. Once you can name them, you can predict them before they cost you anything.

People who want to do it themselves. "how to unblock a sink", "fix office chair gas lift", "repair leather sofa tear". Topically perfect. Commercially the opposite of your customer: they searched precisely because they decided not to pay anyone.

People who want a job at a company like yours. "plumber salary", "electrician apprenticeship", "dental nurse vacancies", "HGV driver jobs". For any service business that employs the trade it sells, this family is quietly large.

People studying the thing. "plumbing courses", "what is cloud ERP", "crm definition", "types of scaffolding". Informational intent wearing your vocabulary.

People who want it free or second hand. "free legal advice", "used office chairs", "crack", "torrent", "open source alternative to". They found you because you sell the thing. They will not buy the thing.

People looking for someone else. Competitor brand names, marketplace names, and the retailer you also sell through. Sometimes profitable, usually not, and worth a decision rather than a default.

People in the wrong place. Geographic modifiers your service area does not cover. A London-only locksmith gets "locksmith manchester" because the meaning is a locksmith, and location targeting governs where the searcher is, not what they typed.

People at the wrong stage or the wrong size. "enterprise" queries hitting an SMB product, "wholesale" and "bulk" hitting a retail store, "jobs for under 16s" hitting a recruitment site.

None of these are matching errors in the model's own terms. Every one of them is a matching error in yours. The cost of the gap is exactly what a wasted spend audit is designed to measure.

Smart Bidding does not fix this — it changes its shape

The standard reassurance is that Smart Bidding will simply bid low on the bad matches, so loose matching costs you nothing. That is half true, and the half that is false is the expensive half.

Smart Bidding does bid lower where it predicts a low conversion rate, so clicks from loose matches are usually cheap. What it also does is make loose matching safe to attempt at scale. When each speculative match costs $0.60 instead of $4, the system can afford to try far more of them. The waste stops arriving as ten obvious terms at $40 each and starts arriving as eight hundred terms at $2 each. Same money. Much harder to see in a report sorted by cost.

There is a second mechanism worth understanding. Smart Bidding optimises toward the conversion signal you gave it. If your conversion action is a newsletter signup, or a phone call counted at 15 seconds, or a form fill from a job applicant, then the terms that produce those events look good to the algorithm — and it will find more of them. A meaningful share of the "why is Google sending me rubbish" cases we see are really "Google is sending me exactly what my conversion action told it to value".

And a third, the one that matters most. Google has no negative signal from you unless you provide one. A click that does not convert is weak evidence: most clicks do not convert, including good ones. At the volume a single search term generates — often one or two clicks, ever — the system cannot distinguish "this term is wrong" from "this term was unlucky". Only you can, by looking at the words and knowing your business. A negative keyword is the only channel through which that knowledge reaches the auction.

Shopping and Performance Max have no keywords to drift from

In Search, at least there is a keyword to compare the query against. In Shopping there is none. Google matches queries to your products using the product data: title, description, product type, brand, GTIN, category. The feed is the keyword.

This is why the Shopping complaint is so consistent — the matching surface is text you wrote for a different purpose. A product titled "Dell XPS 15 Laptop Charger 130W USB-C" will be pulled into "dell xps 15" searches, because the title genuinely contains a laptop model. The searcher wants a $1,600 laptop. You sell a $49 charger. The match is textually defensible and commercially useless, and no keyword edit fixes it because there is no keyword. Only negatives and feed edits do, and the rules for Shopping negatives are their own subject.

Performance Max goes further. It has no keywords, no match types, and no query-level bidding controls. It matches from the asset group, the final URL, the feed and the audience signals, across Search, Shopping, Display, YouTube, Discover, Gmail and Maps. When a PMax campaign starts showing up on brand searches for a competitor or on "how to" queries, that is the model reading your landing page and deciding the topic fits.

You can still block queries: PMax accepts campaign-level negative keywords, up to 10,000 of them. What it does not accept is shared negative keyword lists, which is the single biggest operational difference between PMax and everything else, and the reason the 10,000 cap becomes a real budgeting problem in accounts with many campaigns. Every PMax campaign needs its own copy of your exclusions, and every copy counts against its own ceiling.

One more limit worth stating plainly. Search terms are reported for Search, Shopping and Performance Max — they come from campaign_search_term_view. Display, Video and App campaigns report no search terms at all. If your Display campaign is spending on the wrong audience, no amount of search term work will show it to you, because there is nothing to show.

Why "check the search terms report weekly" is incomplete advice

Every article on this keyword ends the same way: open the search terms report, sort by cost, add negatives, repeat weekly. That advice is correct and it is not sufficient, for three reasons that have nothing to do with discipline.

The waste is in the tail, not the top. Sorting by cost surfaces the terms that spent the most. Under modern matching, the expensive terms are usually the relevant ones — that is what Smart Bidding is for. The damage sits in the long tail of terms that each spent $1.80 and never appeared again. In the accounts we audit, the majority of distinct search terms over a 90-day window appear exactly once. You cannot catch a term that never comes back by reviewing it after it has already cost you the click.

Adding the exact term you saw does nothing. Blocking "plumber salary uk 2024" as an exact negative protects you from precisely one string. Next week it is "plumber salary london", "average plumber wage", "how much do plumbers earn". The unit of waste is the pattern, not the term. Blocking patterns means phrase negatives on the pivot word — "salary", "jobs", "apprenticeship" — and that requires understanding that negative match types do not behave like their positive counterparts. Broad negatives in particular behave unpredictably across match types, which is why most careful accounts default to phrase.

Some terms are never shown to you at all. Google withholds search terms that were searched by very few people, for privacy reasons. There is always a slice of spend attributed to a row you cannot read, and it is not small in low-volume accounts. We covered the size and the workarounds in what the search terms report hides. The practical takeaway: the report is a sample of your waste, not an inventory of it.

Add the fourth reason, which is human. A weekly review of a 4,000-row report is a task that gets done properly twice and skimmed thereafter. The terms that need blocking are boring, individually small, and identical every week. This is work that survives automation well and attention badly.

What actually closes the gap

Three things, in order of how much they matter.

Block families, not instances. Build a small set of phrase negatives around the intent words that disqualify a searcher in your business: jobs, salary, career, course, training, DIY, free, used, second hand, template, meaning, definition, wholesale, and the competitor names you have decided against. A well-chosen list of 200 phrase negatives blocks more waste than 3,000 exact negatives copied from last month's report.

Keep it clean enough to audit. Negative keyword lists grow forever and nobody prunes them. Past a certain size you cannot tell whether a campaign is underperforming or suffocating, and the cost of an over-broad negative added two years ago is invisible by construction — you never see the traffic you blocked. There is a real limit here, and how many negative keywords are too many is worth reading before you build a 10,000-row list you will never review.

Match the cadence to the mechanism. Terms arrive nightly, once each, and never return. A weekly review is one to seven days late on every one of them. This is the specific problem PPCleaner exists to solve: it reads the search terms your account paid for, judges which ones were never going to convert, and adds them as campaign or list negatives every night, across Search, Shopping and Performance Max. It never touches budgets, bids, ads or keywords.

The honest caveats, because they are the part that decides whether this is right for you. Any automated blocking will occasionally exclude a term that would have converted eventually — that is the cost of blocking anything, manual or not. We reduce it two ways: no term that converted in the last 90 days is ever blocked, and every block can be reversed in one click for 90 days. Neither of those is the same as never being wrong. And nothing here helps with Display, Video or App campaigns, because they report no search terms to act on.

What you get in exchange is that the gap between Google's definition of relevance and yours stops widening every night while you are not looking.