How to Find Wasted Google Ads Spend Without Blocking the Searches That Pay

A sample account that paid more for legal aid than for injury claims

Harbour & Finch is a Toronto employment and injury law firm that does not exist. It is the sample client behind our free tools, and its four-week search terms export (August 25 to September 21, 2026, 28 terms, CA$13,627.47 of spend) is built to look like a real small-firm account. In it, the firm paid CA$49.21 a click for the search "legal aid ontario". It paid CA$46.00 a click for "slip and fall lawyer toronto". The injury search brought three conversions. The legal aid search brought 13 clicks, CA$639.67 of spend and none.

Thirteen of the 28 terms recorded no conversion. Together they cost CA$2,917.57, about 21% of the window's spend. That number is what most audits lead with, and it is the least interesting thing in the file.

The harder questions come after the list. How much of that CA$2,917 is waste, and how much is research traffic that converts later or by phone? And which negative removes the waste without also blocking the searches that brought the firm its 45 conversions? The obvious broad fixes in this account would have switched off terms responsible for most of those conversions. What follows is the method, run on the sample, with the export quirks and matching rules that decide whether it works.

The search terms export is built for Excel, not for arithmetic

The report lives under Campaigns, then Insights and reports, then Search terms, as Google Ads Help describes it. Before downloading, add the columns the analysis needs: Campaign, Ad group, Clicks, Cost, Conversions and Added/Excluded. That last column says whether a term is already a keyword (Added), already blocked (Excluded), both, or neither (None). Terms marked Excluded have been dealt with; skip them.

The file itself is laid out for a person reading a spreadsheet, and a script that treats it as a clean table gets the numbers wrong. Four things to handle:

• Two lines sit above the header: the report name, then the date range. The date range appears nowhere else in the file, so read it from there and keep it with every figure you quote.
• Total rows sit at the bottom. The sample ends with "Total: Search terms" and "Total: Account", each CA$13,627.47. Sum the whole Cost column and you triple the account's spend.
• Undefined values are written as --. In the sample, every term with zero conversions shows -- for Cost / conv., because dividing by zero has no answer. A dash is not a zero, and a dash in a Conversions or Cost cell means that value was not measured, so that term cannot be judged.
• Google offers several download formats, including a plain .csv and an Excel-flavored one. If the numbers turn to garbage when a tool opens the file, it is reading the wrong encoding, not reading bad data.

The export also leaves things out on purpose. Google says some search terms "that don't have enough query activity are omitted" for privacy. When Google tightened that threshold in September 2020, Seer Interactive measured its clients' accounts and found the share of spend with visible search terms fell from 98.7% to 71.0%. Your own share will differ, and you can measure it the way Seer did: compare the cost in the search terms export with the campaign-level cost for the same dates. Whatever the gap is, the waste you find in the export is a floor. Spend you cannot see can be wasted too.

Check the conversions before you trust the zeros

Every candidate for waste is defined by a zero in the Conversions column, so the first check is whether that column can be trusted. Add up conversions across the whole export. If the account spent money and recorded no conversions at all, the problem is conversion tracking, and blocking terms on that evidence switches off the ones that were working. The sample passes: 45 conversions on 543 clicks.

A second trap is timing. Google's primary conversion columns are "calculated based on the time of the click, not the time of the conversion". A searcher who clicks on September 20 and books a consultation on October 2 adds a conversion to September 20, after you downloaded the report. For a law firm, where people research, compare and call back, the last week or two of any export understates conversions. Either leave a gap between the end of the window and the download, or treat terms whose clicks fall late in the window with more caution.

The opposite failure matters too. If one inquiry fires two conversion actions, or an action counts a page view instead of a lead, conversions are inflated and real waste looks productive. Fix the counting first, then judge the terms.

Zero conversions on 13 clicks is weak evidence

The mechanical definition of a candidate is simple: cost above zero, zero conversions, not already excluded, and above some minimum of clicks or spend. The trouble is that in a small account the zero carries less information than it seems to.

The sample account converts 8.3% of its clicks (45 of 543). Suppose a term is perfectly good and converts at exactly that rate. The chance it records no conversion on 5 clicks is about 65%. On 10 clicks it is 42%. On 13 clicks, the number "legal aid ontario" received, it is 32%. A term as good as the account average comes up empty on 13 clicks roughly one time in three. None of the 13 non-converting terms in the sample had more than 14 clicks.

So at this scale the count cannot tell you which terms are waste. What the search means can. Read each candidate and put it in one of three piles. Here is the sample, sorted by that reading, which is a judgment and not an output of any tool:

The off-topic pile is about job seekers, students and people looking for help they do not pay for. Seven terms, CA$1,652.17, and they are safe to block because of what they mean, whatever their click counts. The research pile is people working out whether they have a case. Two of those terms already earn organic clicks from the firm's own pages in the sample Search Console data ("accident benefits checklist" at an average position of 4.4, "what is a contingency fee" at 5.1), which makes blocking the ad a reasonable call; the others deserve a better landing page or a lower bid before a block. The unclear pile depends on whether the firm offers a free first consultation. If it does, those are its buyers.

A careful reading cut the sample's "wasted spend" from CA$2,917.57 to CA$1,652.17 that is plainly off topic. That is the figure worth acting on without further evidence.

Negative keywords match literally, and that cuts both ways

Positive keywords in Google Ads stretch. Google's close variants rules let them match misspellings, plurals, stemmings, reordered words and synonyms, and there is no way to opt out. Negatives do not stretch. Google Ads Help says "negative keywords don't match to close variants", and gives the rules for each negative match type. Applied to searches from the sample:

Literal matching has two consequences. The first is that an exact negative is narrow. Blocking [free lawyer consultation] leaves "free lawyer consultations" open, and because the positive keyword does match close variants, that plural can keep showing your ad. Expect exact negatives to need follow-up as variants surface in later reports.

The second is about checking whether a term is already blocked. A string comparison between a proposed negative and the account's existing list misses phrase and broad coverage: a phrase negative "legal aid" already blocks "legal aid ontario", and a naive check would propose it again. The check has to apply each existing negative's match type, on text normalized the same way on both sides (lowercase, accents folded, punctuation as spaces) and never stemmed, because Google does not stem negatives either.

For a single observed term, the default should be an exact negative on that term. It removes exactly what you saw and nothing you did not.

The broad fix that would have cost the firm its buyers

Scan the sample's non-converting terms and words repeat. "Lawyer" appears in six of them, "employment" in five, "legal" in three, "free" in two. One phrase negative per word looks tidier than thirteen exact ones. Here is what each would also have blocked in the same four weeks:

• "lawyer" would block 17 terms, including 11 that converted. Those 11 account for 32 of the account's 45 conversions.
• "employment" would block "employment lawyer toronto": CA$1,388.20 and 5 conversions, the biggest term in the employment campaign.
• "free" would block "personal injury lawyer free consultation": CA$816.30 and 1 conversion, along with the two unclear terms.
• "legal" blocks three non-converting terms worth CA$1,225.97 and nothing that converted in this window.

"Legal" looks clean, and it is clean only for the searches this window happened to contain. A phrase negative on a single word also blocks every future search containing it, such as "employment legal advice toronto", which the firm would probably want. "legal aid" covers the two terms that justified the idea, CA$951.07, with a far smaller blast radius.

The rule that falls out: every proposed negative broader than exact should travel with its collateral, the list of other terms in the window it would block and how many conversions they carried. An empty list means the negative is safe on the evidence you have. A non-empty one means a person decides, looking at the actual searches.

Put the negative where the term was bought

A negative can live in an ad group, a campaign, a shared negative keyword list attached to chosen campaigns, or at the account level. The reach differs a lot. Account-level negatives are capped at 1,000 per account and apply to Search, Performance Max, App, Shopping, Smart and Local campaigns. Performance Max campaigns now have their own search terms report and accept campaign-level negatives for Search and Shopping inventory, so the same method applies there.

The search terms report shows the campaign and ad group each term was bought through, so the smallest safe scope is that ad group. Both legal aid terms in the sample came through Employment | Non-brand, Employment general. Go wider only when the reason for blocking holds everywhere. Nobody searching for legal aid is a prospect for the injury campaign either, so a shared list across both non-brand campaigns is defensible. That is a judgment about meaning, and it should be written down as one.

Ad group scope has its own failure. If another ad group carries a broad keyword that also matches the search, blocking it in one ad group can simply move it to the next. After the negatives go live, rerun the report and look for the same terms under a different ad group or campaign.

The same care applies when you check whether a term is already blocked. Say which scopes you checked. "Checked the ad group and campaign; shared lists and account negatives not checked" is an honest line. "Not already blocked", when half the places a negative can live were never read, is not.

Call it spend with no recorded conversion, and say what you could not see

A client report on this work should give the window, the currency and the scopes checked. It should call the number what it is: in the sample, CA$1,652.17 in four weeks on searches that were off topic, and CA$2,917.57 on all terms that recorded no conversion. Neither is a saving. Blocked budget gets spent on other searches, and some of the blocked clicks might have converted. Because privacy thresholds hide part of the spend, both are floors for the window, not totals.

Measuring the effect means comparing like with like: the account's cost per conversion in the weeks after the negatives went live against the same weeks before, and against campaigns you did not touch, so a seasonal swing is not credited to the list.

The free Ads leak check runs the mechanical half of this on your own export in the browser, with nothing uploaded: it reads the preamble and Total rows, skips Excluded and unmeasured terms, proposes an exact negative at the ad group for each non-converting term, and shows which converting searches a broader phrase negative would also block. It cannot see the account's existing negative lists, and it says so. The reading, which terms are research and which are job seekers, stays with you. In AIO Copilot the same method runs on connected accounts, with negatives written for a person to review and apply.

Before you upload any list, sort it by spend and read the top twenty searches out loud. If one of them sounds like a person who might hire you, take it off.

Frequently Asked Questions