Where wasted Google Ads spend actually goes

A Phoenix HVAC operator was spending $920 a month on the search term "free ac repair." Nine hundred and twenty dollars, every month, to attract people who explicitly did not want to pay. The campaign was set up correctly. The bid was reasonable. The leak had been there fourteen months.

That number is not unusual. Across the small-business accounts we audit, somewhere between ten and forty percent of paid clicks come from queries the account owner would not have chosen to bid on, if they had been asked term-by-term. Most of that waste sits in four recognizable patterns. Once you know the patterns, the fix is mechanical.

This piece is the working version of those four patterns and what each one is costing you.

Pattern one: irrelevant clicks

The first leak is the simplest. Someone searches for something adjacent to your service, the matcher decides it is close enough, your ad shows, the click costs five dollars, the user lands on your page, realizes it is the wrong service, and bounces.

A residential plumber bidding on plumber gets clicks from "plumber salary," "plumber jobs near me," and "wholesale plumbing supply." None of those searchers will hire a residential plumber. The clicks still cost the same.

The fix lives in the negative keywords pillar: block the words that mark a non-buyer (free, salary, jobs, wholesale, schools) and most of the irrelevant pool disappears in one sweep.

Pattern two: junk traffic from broad-match expansion

The second leak is structural. Google's broad match expands keywords aggressively to "related" queries. Bid on plumber phoenix as broad and the matcher serves your ad against "plumbing services in arizona," "best plumber tempe," and "emergency drain repair scottsdale." Some of those are good. Some are not. Broad match does not let you choose.

Junk traffic from broad expansion is harder to find than pattern-one irrelevance because the queries look almost-relevant. The user did intend something plumbing-related; they just did not intend to call your specific business. Conversion rate on these terms is much lower than on the literal keyword.

The fix is two-part: tighten match types where the data justifies it (move high-volume broad-match keywords to phrase or exact when their conversion rate trails the account average), and add phrase-level negatives for the recurring junk patterns.

Pattern three: low-quality search terms with high cost

The third leak is the one most operators never look at, because the queries technically converted at some point. A term gets one conversion, the auto-bidder gives it more budget, the term spends two thousand dollars over six months, generates four conversions total, and quietly drains the campaign.

These are low-quality search terms in the cost-per-acquisition sense. They convert, but at a rate too high to justify the spend. The auto-bidder cannot tell the difference between a query that converts at three percent and one that converts at zero point three percent until a lot of money has been spent.

Finding low-quality terms requires sorting the search terms report by cost-per-conversion instead of by raw cost. The terms you want to flag are high-cost, low-conversion-rate, and consistently spending. Once flagged, the move is usually a phrase-match negative rather than a full block, because the term does sometimes work; you just want to push the matcher away from it.

Pattern four: bot traffic and click fraud

The fourth leak is rarer and worth less time than the first three, but it is real. Click fraud, scraper bots, and competitor sabotage clicks all show up in Google Ads and all cost real money before Google's filters catch them. Google does refund a portion of invalid clicks, but the refund is incomplete and slow.

The signal: a sudden spike in clicks on one term with no corresponding conversion lift, often from a single geography or device class. Genuine traffic spikes have a conversion echo. Bot spikes do not.

For most accounts, the cheapest fix is geographic targeting (exclude geographies you do not serve), audience exclusions (block known low-quality segments), and patience. Click fraud tools exist; they help on accounts spending five figures a month. Below that, the time investment is rarely worth it.

The longer treatment is in click fraud vs irrelevant clicks. For a starting-point operator, the rule is simple: do not chase pattern four until patterns one through three are clean.

What each pattern costs, in rough numbers

A composite from accounts we have audited over the last two years. Numbers are typical estimates:

How to triage your own account in fifteen minutes

The first pass on an unfamiliar account is faster than people expect. Three questions, fifteen minutes, decisions about where to sweep first.

Why operators do not do this

Most operators know that wasted spend exists. The reason it persists is not ignorance; it is friction. The search terms report is buried. The decision per row takes thirty to sixty seconds. The cumulative time to clean an account from scratch can be ninety minutes or more.

That friction is a real problem because the work has high payoff. A ninety-minute sweep that recovers $700 a month works out to roughly $470 per hour, every month, for the life of the account. Few operator activities pay back at that rate.

The friction is the reason we built Sensei Ads. The product reads the search terms report alongside a written business description and returns a categorized negative list with reasons. The per-row decision drops from sixty seconds to about ten. The work is the same; the time per account is roughly fifteen minutes instead of ninety.

Whether you use a tool or a spreadsheet, the patterns are the same. Find them in your own account and the math takes care of itself.