What is an 'almost relevant' search term?

An almost-relevant search term is a query that is topically about your product or service but carries an intent that will never convert — someone researching rather than buying, looking for a job rather than a vendor, wanting a free or DIY version rather than a paid one, or at the wrong stage of the funnel. It reads as relevant because the subject matches, which is exactly why it survives an irrelevance scan, and it often spends steadily without converting. Karooya frames the 2026 shift bluntly: the biggest issue today is not irrelevant queries but almost-relevant ones that waste spend without converting.

Why do zero-conversion and irrelevance scans both miss these terms?

They slip through both filters by design. An irrelevance scan looks for queries that are off-topic, and these are on-topic, so they pass. A zero-conversion scan catches terms with spend and no conversions, but almost-relevant terms often convert occasionally — a stray sale from a researcher who came back, or a low but non-zero rate — which is enough to keep them off a strict zero-conversion list while their blended economics stay poor. The term is relevant enough to look fine and productive enough to look alive, so neither automatic cut reaches it. Finding it takes reading the query as a person would, not sorting a column.

Should I always negate an almost-relevant term?

No. Some wrong-intent-now queries are genuine top-of-funnel demand that converts later through another channel or a return visit, and negating them cuts the top of your funnel to flatter a last-click report. Before you negate, decide whether the term is truly unproductive or merely early. If it spends materially and shows no assisted or downstream value over a fair window, negate it. If it is cheap, on-topic and plausibly a future customer learning the category, consider leaving it or moving it to a lower-bid context rather than blocking it outright.

Where should intent-based negatives live — campaign or account level?

Intent modifiers that are wrong for the whole business — job-seeking, free, DIY, torrent — belong in an account-level shared negative list applied broadly, because they will never convert in any campaign. Intent that is only wrong for a specific campaign or ad group stays local to that scope. Keeping the universal wrong-intent words in one shared list means you fix them once and every campaign inherits the block, and you avoid the drift of the same ten modifiers half-applied across a dozen campaigns.

Almost-Relevant Search Terms: Cut Wrong-Intent Waste

Q: What intent signals mark a term for negation?

Common wrong-intent signals include learning and DIY words (how to, tutorial, guide, template, yourself), job and career words (jobs, salary, career, hiring, vacancy), free and low-value words (free, cheap, crack, torrent, trial), research words (what is, meaning, examples, statistics, reddit), and wrong-tier or wrong-audience words that point at a product you do not sell or a buyer you do not want. The modifier, not the head term, is what you negate — adding the modifier as a negative blocks the wrong-intent variants while leaving the on-intent queries that share the same subject untouched.

The search term that drains the most budget in 2026 is not the obviously irrelevant one. You already catch those: the query that has nothing to do with your product stands out, gets negated, and never bothers you again. The expensive problem is the query that looks right — it is about your product, it uses your words, it would pass any "is this relevant?" check — but carries an intent that will never buy. Karooya, reframing the year's negative-keyword problem, puts it plainly: the biggest issue today is not irrelevant queries, but "almost relevant" queries that waste spend without converting.

These are the terms that survive every automatic cut. A relevance filter waves them through because they are on-topic. A zero-conversion filter waves them through because they occasionally convert, or because their spend is spread thin enough to look harmless line by line. So they sit in the search terms report in plain sight, bleeding budget at a steady drip, and the only way to catch them is to read the query the way the person who typed it meant it. This post is about doing that systematically: naming the intent buckets, finding them in the report, and negating the modifier rather than the subject.

What "almost relevant" actually means

An almost-relevant query matches your topic but not your buyer. Sell project-management software and "project management certification" is about your subject — and about a student, not a customer. Sell running shoes and "how to clean running shoes" is on-topic and worthless. "Plumber salary" is about plumbing; it is a career search, not a service request. In each case the head term is yours and the intent is someone else's. That mismatch is why these queries are corrosive: every signal your eye uses to judge relevance says yes, while every signal that matters for conversion says no.

The reason this has become the dominant form of waste is that match types have loosened. Broad match and close variants now pull in queries that share meaning or context rather than words, so the net catches far more topically-adjacent intent than it used to. The off-topic junk still appears and is still easy to cut; what is new is the volume of on-topic, wrong-intent traffic that the looser matching sweeps in. Negatives have shifted from a filter you run occasionally to a control you use to define the boundary of who a campaign is for — and the boundary you most need to draw now is by intent, not by subject.

Why both standard scans miss them

Start with the irrelevance scan, the one most practitioners run first: scroll the search terms report and cut anything obviously off-topic. Almost-relevant queries are immune to it, because they are not off-topic. "Free CRM," "CRM tutorial," and "CRM jobs" all read as relevant to a CRM advertiser at a glance, so they pass the exact check designed to catch waste. The filter is looking for the wrong property — subject match — when the problem is an intent mismatch that subject match cannot see.

The zero-conversion scan fails differently. Sort by spend, find terms with conversions of zero, negate them: a reasonable routine that misses almost-relevant terms for two reasons. First, many of them convert occasionally — a researcher who came back and bought, a one-off sale — which keeps them off a strict zero-conversion list while their true economics stay underwater. Second, their spend is often distributed across many low-cost queries that share one intent word, so no single line looks bad enough to flag. This is the same concentration problem that an n-gram analysis solves for ordinary words, and the same reason hidden wasted spend hides so well: waste that is spread thin reads as noise, not signal.

The intent buckets to look for

Reading for intent gets faster when you know the buckets. The recurring ones are learning and DIY ("how to," "tutorial," "guide," "template," "yourself"), jobs and careers ("jobs," "salary," "career," "hiring," "vacancy"), free and low-value ("free," "cheap," "crack," "torrent," sometimes "trial"), research and definition ("what is," "meaning," "examples," "statistics," "reddit"), and wrong-tier or wrong-audience terms that name a product you do not sell or a buyer you do not serve. Each bucket is a pattern of modifier words, and the modifier is the tell — not the subject it attaches to.

The buckets matter because they are reusable. The specific almost-relevant queries in your report are account-specific, but the intent words that create them are nearly universal: "jobs" and "salary" signal a career search whatever you sell, "free" signals a non-buyer in most paid accounts, "how to" signals a DIY-er who wants to avoid paying anyone. Once you have mapped a query to a bucket, you are not negating that one query — you are negating the modifier that will keep producing its siblings next week and the week after.

Classifying a query by intent in the report

The method is a change of reading, not a change of tool. Open the search terms report, set a window with enough data to be fair, and sort by cost so your attention goes where the money is. Then, instead of asking of each top-spending term "is this relevant?", ask "what did the person who typed this want?" A query is a sentence a stranger wrote about their need; read it as one. "Best X for small business" wants to buy; "what is the best X" might; "X meaning" does not; "X jobs" definitely does not. The relevance view flattens all four into "about X." The intent view separates the buyer from the student, the researcher and the applicant.

As you read, tag each wrong-intent term with its bucket rather than negating it one at a time. Tagging turns a scroll into a list of modifiers, and a list of modifiers is what you act on. It also stops you treating a single query as the unit of work when the real unit is the pattern — the difference between grouping by pattern and negating single terms. A term that is a single ambiguous word rather than a clear wrong-intent phrase is a different case: send it to a review queue rather than negating on a guess, because an ambiguous token can hide good traffic you cannot yet see.

Building intent-based negatives — and where they go

Negate the modifier, not the subject. If "free crm" is the problem, a phrase-match negative "free" blocks the whole wrong-intent family while leaving "crm software" and "crm pricing" untouched; negating the exact query "free crm" alone leaves "free crm tool," "free crm download" and a dozen variants still spending. Choosing the match type is the usual trade-off between reach and safety — a phrase-match intent negative catches more variants but risks a word that is sometimes legitimate, so weigh it the way you would any negative match-type decision. And remember negatives do not match close variants the way positive keywords do, so plurals and misspellings of an intent word may each need their own entry unless Google's automatic misspelling coverage handles it.

Placement follows scope. Intent words that are wrong for the entire business — jobs, salary, free, torrent, DIY — belong in an account-level shared negative list applied across campaigns, because they will never convert anywhere. Intent that is only wrong in a particular campaign stays local to it. The danger to respect is the mirror image of under-filtering: over-negating a modifier that sometimes rides along with a buyer. "Cheap" is a clean non-buyer signal for a premium brand and a core term for a budget one; "trial" is a tyre-kicker for some and a high-intent SaaS signal for others. The bucket is a prompt to look, not a licence to block on sight.

When an almost-relevant term is worth keeping

Not every wrong-intent-now query is waste. Some are the top of your funnel: a researcher reading about the category today is a buyer comparing vendors next quarter, and a last-click report that shows them converting nothing is describing the wrong moment, not the wrong person. If you negate every informational and research query, you can hand your competitors the entire early funnel and wonder later why your branded demand stopped growing. The question is not "did this term convert on last click?" but "is this term plausibly a future customer, and is it cheap enough to tolerate while it learns the category?"

So apply the same discipline you would to any marginal-conversion trade-off: over a fair window, does the term carry assisted or downstream value, or is it a steady drip that goes nowhere? Negate the ones that spend materially and show no sign of ever paying back. Keep — or move to a lower-bid context — the cheap, on-topic ones that might be early rather than wrong. The point of the almost-relevant lens is precision: cut the queries whose intent is genuinely incompatible with buying, and stop cutting the ones that merely have not bought yet.

For single ambiguous words that need judgement rather than a blanket negative, see the review-queue approach to high-spend tokens. For weighing terms that do convert but barely, see the ROAS trade-off on marginal conversions, and for finding waste spread thin across many queries, see n-gram search-term analysis.