How to Find Wasted Ad Spend in Google Ads

Three kinds of waste, and only one is a negative keyword problem

Before you open a report, split the word "waste" into three piles. They have different fixes, and mixing them is why audits produce long lists and small results.

Wrong intent. The person searching was never going to buy. "plumber salary", "how to unblock a sink yourself", "plumbing courses near me", "is plumbing a good career". No landing page, no bid, no ad copy changes the outcome. This pile is what negative keywords are for, and it is the only pile PPCleaner touches.

Right intent, bad economics. "emergency plumber london" converts, but at $180 against a $60 target. That is a real buyer and a real conversion. Adding it as a negative is not a fix, it is amputation. The fix is bids, landing page, call handling, or accepting that this term is expensive because the intent is valuable and everyone else knows it too. If you find yourself writing high-intent commercial terms onto a negative list, stop and check which pile you are in.

Spend with no term attached. Some of your Search cost never appears next to any query in the report. You cannot negative it because you cannot see it. More on this below, because every audit guide quietly ignores it.

Most articles on this keyword collapse all three into "find irrelevant queries and block them". That is fine advice for the first afternoon and useless from the second week onward.

Pull the data account-wide, 90 days, one row per term

The single most common mistake in a wasted spend audit is looking at the search terms report inside one campaign, segmented by ad group. The same query then appears six times at $4 each and never looks like a problem.

What you want is one row per search term for the whole account, with clicks, cost, conversions and conversion value, over 90 days. In the API that is campaign_search_term_view, which covers Search, Shopping and Performance Max. Display, Video and App campaigns report no search terms at all — there is nothing to audit there, and any tool that claims to clean them is describing something else (placements, audiences, content exclusions).

A few practical notes on the pull:

• 90 days is the default, not the rule. If the account spends $2,000 a month, 90 days gives you enough clicks per term to judge anything. If it spends $200 a month, use 12 months, or you will be reading noise. If it is seasonal — Christmas gifting, tax season, summer holidays — pull the equivalent period from last year separately rather than blending it in.
• Include conversion value, not just conversion count. A term with one conversion worth $9 and $140 of cost is a different animal from a term with one conversion worth $900.
• Keep the campaign name as a column, but aggregate on the term. You want to see the total damage first, then decide where the negative goes.
• Sort by cost descending once, then never again. That view is the fat head of the distribution and it is the part your eyes were always going to find. The work is below it.

The threshold: how much must a term spend before zero conversions means anything

Here is the answer most audits skip, because it requires arithmetic rather than judgement.

A term with 6 clicks and no conversions has told you nothing. A term with 75 clicks and no conversions has told you something. The boundary between them is not a feeling, it is a function of your conversion rate.

Step one: how many clicks does one conversion normally take?

Divide 1 by your account conversion rate.

• 10% conversion rate → 10 clicks per conversion
• 4% → 25 clicks
• 2% → 50 clicks
• 1% → 100 clicks

Use the conversion rate of the campaign type you are auditing, not the blended account figure. Shopping and Search rarely convert at the same rate.

Step two: multiply by three

If a term converts at your account average, the chance it produces zero conversions across n clicks is (1 − rate) to the power n. Run that out for a 4% account:

That is plain arithmetic on your own numbers, not a published benchmark. Do it with your rate, not mine.

Read the table as a false-positive rate. Block at 25 clicks and roughly one in three of your exclusions kills a term that was actually fine. Block at 75 clicks and you are wrong about one time in twenty — which, against the cost of leaving dead terms running for another quarter, is a trade most accounts should take.

So the working rule is: three times your clicks-per-conversion figure, with zero conversions, earns an exclusion. At 4% that is 75 clicks. At 1% it is 300 clicks, which is why low-conversion-rate accounts genuinely cannot judge most of their terms statistically and have to lean on the semantic pass in the next section.

Step three: use cost instead of clicks when CPCs vary

If your CPCs range from $0.40 to $14, click counts mislead. Switch to money: a term with zero conversions that has spent two to three times your target CPA is a candidate; at three times, act unless the intent is obviously commercial. Target CPA $60 means $120 is a question and $180 is an answer.

Cost and clicks usually agree. When they disagree — 200 cheap clicks, $30 spent — the term is not urgent. Note it and move on.

The exception that protects you

One conversion cancels the whole calculation. A term that converted, even once, even badly, is no longer a negative keyword question; it is a bid question. This is a hard rule in PPCleaner: we never block a term that converted in the last 90 days, regardless of how bad the cost per conversion looks. The downside is real and worth stating — a term that produced one fluke lead against $400 of spend will keep running until you deal with it yourself. We would rather leave that decision with you than take a conversion off the board automatically.

The long tail is where the money actually is

Run the threshold test and you will get maybe 30 or 40 terms in a mid-sized account. Add them all, and the share of spend going to non-converting queries barely moves. This is the part that surprises people.

The reason is the shape of the data. In the accounts we audit, the terms that individually pass a 3× threshold are usually a minority of the non-converting spend. The rest is spread across hundreds or thousands of queries with one, two or three clicks each. No single one of them will ever be statistically judgeable. Waiting for them to accumulate evidence means waiting forever, because most of them will never be searched twice.

The fix is to change the unit of aggregation. Stop counting queries. Count words.

The n-gram pass

Take your 90-day export. Break every search term into its individual words and its two-word pairs. Sum cost, clicks and conversions against each word and pair. Sort by cost, zero conversions first.

For a plumbing account, that produces rows like:

• salary — 41 terms, 68 clicks, $310, 0 conversions
• jobs — 33 terms, 51 clicks, $244, 0 conversions
• how to — 96 terms, 180 clicks, $520, 1 conversion
• apprenticeship — 12 terms, 19 clicks, $91, 0 conversions
• wikipedia — 8 terms, 11 clicks, $48, 0 conversions

Not one of those individual queries would have crossed a click threshold. Together, salary alone is five hours of a plumber's revenue burnt on people researching a career change.

This is the pass that separates a real audit from a tidy-up, and it is the pass that is genuinely tedious by hand — which is most of why PPCleaner exists. It runs the same aggregation every night across the whole account and only surfaces patterns with enough cost behind them to be worth a block.

The patterns worth looking for by name

When you scan the n-gram list, these categories account for most of it:

• Job seekers: salary, jobs, vacancies, apprenticeship, career, hiring, cv, "how much do X earn"
• DIY and research: how to, diy, yourself, tutorial, youtube, forum, reddit, wikipedia, meaning, definition
• Free and cheap: free, freeware, gratis, cheapest, second hand, used, crack, torrent
• Wrong geography: city and country names you do not serve, even when the campaign is geo-targeted, because location targeting matches location of interest as well as presence
• Wrong product tier: "enterprise" when you sell to SMBs, "single user" when you sell 50-seat licences
• Competitor names: sometimes valuable, sometimes not. This is a business decision, not an audit finding. Decide the policy once and apply it.
• Adjacent products you do not sell: the boiler company catching "radiator valve", the accountant catching "bookkeeping software"

Where the n-gram method bites back

It over-blocks on ambiguous words, and the ambiguity is account-specific.

free is dead spend for a $5,000 software product and essential for a retailer whose best query is "free shipping over 50". cheap is a disqualifier for a luxury brand and the core proposition of a discounter. used kills a manufacturer and feeds a parts dealer.

So never block an n-gram without first filtering your export for terms containing that word that did convert. If there are any, the word is not the problem — a narrower pattern is. If there are none across 90 days and meaningful spend, you are safe.

The spend you cannot see in the search terms report

Add up the cost column of your search terms export and compare it to the campaign cost for the same dates. It will not match. In the accounts we audit, the search terms report typically accounts for somewhere between 60% and 85% of Search spend, and the gap is usually wider in Shopping and Performance Max.

Google withholds queries that too few people have searched, for privacy reasons. Those queries still cost money. This has one honest consequence for anyone selling or performing a wasted spend audit: there is a floor under how much of your waste you can find, and it is not zero. No tool, ours included, can add a negative for a query it was never shown.

What you can do about the invisible portion:

• Block patterns, not queries. A phrase negative on salary catches every unseen query containing the word, including the ones the report never showed you. This is the strongest argument for the n-gram pass — it is the only method that reaches into the hidden portion.
• Tighten match types on the keyword side. Broad match is where most unattributed spend originates. That is a keyword decision, not a negatives decision, and PPCleaner deliberately does not make it for you.
• Watch the gap as a metric. If the unattributed share is climbing month over month, something upstream changed — usually a match type loosening or a new campaign type.

Turning candidates into negatives without breaking the account

A list of bad terms is not the job. Where and how you exclude them decides whether you save money or quietly strangle a campaign.

Match type. Use exact negatives for specific queries you want dead and nothing else. Use phrase negatives for patterns — a phrase negative on salary stops "plumber salary uk" and "average plumber salary" without touching anything else. Avoid broad negatives. They behave unpredictably across match types, and the failure mode is silent: traffic disappears and nothing in the interface tells you which negative did it. The three or four extra phrase negatives you write instead are cheaper than one afternoon of debugging.

Level. Ad group negatives for terms that belong in a different ad group. Campaign negatives for terms that should not enter this campaign. Shared lists for account-wide disqualifiers — job seekers, competitor names if that is your policy, obviously unrelated products. Shared lists work on Search and Standard Shopping.

Performance Max is different. It accepts up to 10,000 campaign-level negative keywords and does not accept shared negative keyword lists. If you run several PMax campaigns, your account-wide exclusions have to be written into each of them separately, and the cap is real enough to plan around once your patterns run into the thousands. We covered the mechanics and how to stay under the ceiling in Negative Keywords in Performance Max and the 10,000 Cap.

Keep a way back. Every exclusion is a bet, and at a 3× threshold roughly one in twenty is wrong. Log what you added, where, and why. PPCleaner keeps every block reversible in one click for 90 days for exactly this reason — the value of an automated cleaner is not that it is never wrong, it is that being wrong is cheap and visible.

How to tell whether it worked — and what will not change

Say this out loud before you report on the work: negative keywords do not reduce your spend. Your daily budget is unchanged. Google will spend it on the queries that remain. If you blocked $600 of quarterly waste and did not touch the budget, that $600 still leaves your account — it just buys different clicks.

So "saved $600" is a fiction, and any tool or agency reporting it that way is selling you a number. The honest claims are:

• Cost per conversion at constant budget. Same spend, more conversions, or the same conversions cheaper. This is the real outcome and it usually takes 4 to 6 weeks to read cleanly.
• Share of spend on zero-conversion terms. Track it monthly. In accounts we audit it commonly sits between 10% and 30% of Search spend over a 90-day window before any cleaning. Watching that number fall and stay down is the most direct evidence the process is working.
• Time to first exclusion. How long a new bad query lives before it is blocked. Monthly audits mean a bad term runs for up to 30 days. Nightly means it runs for one.

And watch one guard rail. If total conversions fall while cost per conversion improves, you have cut into real demand — usually via a broad negative or an over-eager n-gram. Pull the log, find the exclusion that landed just before the drop, and reverse it. That is a two-minute job if you kept records and a two-week job if you did not.