CPA Too High in Google Ads? Check Search Terms First

A high CPA is an average, and part of it has no conversions at all

Your CPA is total cost divided by total conversions. Some of that cost goes to search terms that will never convert, so the average is pulled up by spend that has nothing to show for it. The CPA you see is a blend of terms that convert well, terms that convert badly and terms that can't convert at all.

Here is an illustrative account. It spends $10,000 a month on Search and gets 200 conversions. The CPA is $50. Over the same period, its search terms report breaks down like this:

The terms that convert do so at $32.50 each ($6,500 ÷ 200). The account-level CPA of $50 comes from the $3,500 spread over the other two groups. Only the last group is waste in the strict sense. Take out its $1,500 and the same 200 conversions cost $8,500. That's a CPA of $42.50, and no bid, target or ad has changed.

This is why the order of work matters. Most CPA advice treats the $50 as one number to push down. It's really three numbers, and you fix one of them differently from the other two.

Why the usual "lower your CPA" advice is incomplete

The pages that rank for this problem mostly suggest changing your bid strategy, lowering your target CPA, improving Quality Score, and fixing landing pages. All of that is reasonable. What's missing is that these steps adjust how much you pay per auction. They don't change which auctions you're in.

Lowering a target CPA tells Smart Bidding to aim for a lower average. It does that by bidding less, and it bids less across the board. It bids less on "emergency plumber near me", which was already converting at $32.50, and you lose some of those auctions to competitors. It also bids less on "plumber salary", but it still enters those auctions whenever your keywords match. You keep paying for some of those clicks, just a bit less per click.

Smart Bidding does adjust bids auction by auction, and it does learn that some queries convert poorly. But it learns from conversion data, and a term like "plumbing apprenticeship" will never produce any. Each new variant of that intent ("plumber apprentice pay", "plumbing trade school near me") starts with no history. Under broad match and Performance Max, new variants keep appearing. Bidding can make those clicks cheaper. Only a negative keyword stops you buying them at all.

There's a second problem with lowering the target too early. The usual complaint is that CPA went up after raising a target, and that's expected: a higher target lets Google buy more expensive, less likely conversions. The reverse also holds. Lower the target on an account carrying dead spend and you squeeze volume out of good terms before you've removed bad ones. You might reach the target CPA, but on fewer conversions than you could have kept.

So the question to ask first is: how much of my cost is going to terms that were never going to convert? Answer that before deciding whether the target is wrong.

How to measure the share of cost going to terms that never convert

You can do this in under an hour with the search terms report or a spreadsheet export. The goal is one number: the cost of clearly impossible terms as a share of total cost.

1. Pull enough history

Use 90 days. Shorter windows mix "hasn't converted yet" with "won't convert". If your conversions take days or weeks to be reported (lead forms that become sales, offline imports), end the range a little before today so recent clicks have time to convert. Use the conversion column your bidding actually optimises to, not "all conversions".

Behind the report, Google reports search terms for Search, Shopping and Performance Max in campaign_search_term_view. Expect the term rows to add up to less than the campaign's total cost, because Google doesn't show every search term. Note the gap. You can't act on it, but you shouldn't pretend it isn't there.

2. Sort by cost, not clicks or impressions

Cost is where CPA comes from. A term with 400 impressions and two clicks barely matters. A term with 40 clicks at $6 each is $240 that has to be justified. Sort descending and work down the list until the terms are too small to matter. In most accounts a few hundred rows cover nearly all the spend.

3. Put every term in one of three groups

• Converted. At least one conversion in the window. Keep these, whatever their individual CPA. A term with a high CPA is a bidding question, not a negative keyword question.
• Relevant, no conversion yet. Someone searching this could plausibly buy from you. "Water heater replacement cost" is a research query from a homeowner who may well hire a plumber next week. Leave it in.
• Could never convert. The intent doesn't match anything you sell. Job seekers, DIY instructions, free versions, wrong product, wrong location, a different meaning of the same word.

Judge the third group on intent, not on clicks. You don't need 60 clicks on "plumber salary" to know it won't produce a service call. Our guide to job seekers clicking Google Ads covers the most common pattern. The article on blocking "free" and "cheap" searches without losing buyers covers the one people most often get wrong.

4. Calculate the share

Add up the cost of the third group and divide by total cost. In the example above: $1,500 ÷ $10,000 = 15%. Calculate the second group separately: $2,000 ÷ $10,000 = 20%. That figure tells you how much CPA depends on terms that are still unproven.

Be strict with the third group and generous with the second. If you're unsure, the term goes in group two. If you'd like a fuller method, including how to handle the unreported gap, see how to calculate wasted ad spend honestly.

In the accounts we audit, the third group usually takes somewhere between 10% and 25% of search term cost. It tends toward the top of that range, sometimes beyond it, in accounts that lean on broad match or Performance Max and haven't reviewed terms in months. Your number may sit well outside that range. Calculate it rather than assume it.

"Never converts" vs "hasn't converted yet"

This is the part that needs judgement, and it's where both people and tools make mistakes. A term with clicks and no conversions isn't automatically waste. Low volume means you don't have enough evidence.

Run the numbers. Say your account converts 5% of clicks. A relevant term with 20 clicks and no conversions isn't unusual: the chance of 20 clicks in a row without a conversion at a 5% rate is about 36%. At 60 clicks and still nothing, that chance drops to about 5%. Now the zero tells you something. Even then, it might mean the term converts at 2% rather than 5%, which makes it expensive, not impossible.

So treat the two groups differently:

• Impossible intent: block on the first click if you like. The evidence is in the words, not the conversion column. "Plumbing apprenticeship" won't convert however many clicks it gets.
• Relevant but non-converting: leave it, or deal with it through bidding, ad copy or landing page. Block only if the volume is high enough that the zero is convincing and you understand why it doesn't convert. Example: "tankless water heater" might never convert for a plumber who doesn't install tankless units. That's really a wrong-product term.

Some terms sit on the line. Competitor names are the usual case: they rarely convert well, but sometimes they're the cheapest new customers you have. We've written separately on whether to block or keep competitor searches. For CPA purposes, keep them in group two unless the data clearly says otherwise.

One rule we apply without exception: a term that converted in the last 90 days doesn't get blocked, however bad its CPA looks. PPCleaner follows the same rule. A converting term with a high CPA is something for your bids to handle. A negative keyword cuts it off completely.

Fix the search terms, then wait, then decide on targets

The order is: block impossible terms, give bidding time to settle, recalculate CPA, and only then decide whether your target needs to move. If you change negatives and targets on the same day, you can't tell which change did what.

Block with the right match type

Use exact negatives for single terms you've seen and want gone: [plumber salary]. Use phrase negatives for patterns that are never relevant in any combination: "apprenticeship", "salary", "jobs" for most service businesses. Avoid broad negatives unless you know exactly what they'll catch. They behave unpredictably across match types, and a broad negative on a common word can quietly block buyers.

Before you add a phrase negative, check it against your own keywords and converting terms. A negative "cost" would block "water heater replacement cost" and probably a good share of your buyers. Negative keyword conflicts Google Ads won't flag lists the collisions to look for. Remember too that your positive keywords match close variants, but negatives don't expand the same way. The reasons are covered in close variants and why exact isn't exact.

Expect the freed budget to be spent again

This is the cost of the advice, and few people mention it. If your campaign is limited by budget, removing $1,500 of dead spend doesn't save $1,500. Google spends it on other auctions. Some of those will convert, some will be group-two terms, and some will be new junk variants you haven't seen yet.

The result shows up differently. Instead of the same conversions for less money, you get more conversions for the same money. You won't buy them at $32.50 each, because the extra clicks are marginal ones. Even at $75 per conversion, $1,500 buys 20 more. That's 220 conversions for $10,000, a CPA of about $45.45. It's an improvement, but a smaller one than the $42.50 you'd see if the budget weren't limited. Measure the result the way it will actually show up.

It also means one cleanup isn't enough. The respent budget finds new terms, and the process starts again. How often to look depends on spend and match types. See how often to check search terms.

Give bidding one full conversion cycle

After a batch of negatives, Smart Bidding is working with a different mix of auctions. Leave targets alone for at least as long as it normally takes a click to become a reported conversion in your account. For an online store that might be a few days. For a B2B lead that becomes a sale weeks later, it's longer. Then recalculate CPA over a clean window that starts after the change.

Then, and only then, look at the target

If CPA is now at or near your goal, you may not need a lower target. If it's still too high, you now know why. The remaining cost goes to terms that are relevant but convert poorly, or to converting terms that cost too much. That's what bidding, landing pages and offers are for, and it's when the advice on the ranking pages becomes useful. Lower targets in steps rather than one big cut, and watch conversion volume as well as CPA.

Performance Max and Shopping: the same idea, fewer levers

The method works across Search, Shopping and Performance Max. The tools for acting on it are different, and in Performance Max some of the cost sits where search terms can't reach it.

Standard Shopping accepts negative keywords and shared negative lists, just like Search. Shopping search terms tend to be messier because there are no keywords, only your feed. The cleanup matters more there, but it uses the same three groups.

Performance Max accepts up to 10,000 negative keywords at campaign level. It doesn't accept shared negative keyword lists, so a list you maintain for Search has to be added to each PMax campaign separately. Its search terms appear in campaign_search_term_view, but they only cover the Search and Shopping parts of its spend.

That's the hard limit. Display, Video and App inventory report no search terms. If a Performance Max campaign spends part of its budget on placements, that cost can't be split into three groups at all. A high PMax CPA could come mostly from somewhere search term cleanup can't touch. Before blaming search terms for a PMax CPA, read where Performance Max wasted spend hides and what the PMax search terms report actually shows. Placements need different tools, and they're more limited than most people expect.

When search terms aren't why your CPA is high

Sometimes the third group is small. If impossible terms take only a few percent of cost, negatives won't fix your CPA. You've ruled out one cause in under an hour, and that's still useful.

The usual suspects after that:

• Conversion tracking changed. A tag broke, a conversion action was removed from the primary goal, or a form started firing twice and then stopped. CPA can rise overnight without a single search term changing. Check this first if the jump was sudden.
• The target is unrealistic for the market. If competitors' CPCs rose, a CPA that worked last year may not be available this year at the same volume.
• Conversion lag. Recent days always look worse before late conversions come in. Don't judge last week's CPA on Monday.
• Conversion rate dropped on the site. A slower page, a price rise or a sold-out best seller all raise CPA without any change in Google Ads.
• Spend moved somewhere else. Device, location or time of day can move cost without changing search terms. Wasted spend by device is one place to look.

A quick way to work through these in order is the Google Ads audit checklist you can run in an hour.

Where PPCleaner fits, and where it doesn't

PPCleaner automates the third group. Every night it reads the search terms your account paid for across Search, Shopping and Performance Max and judges which had no plausible path to conversion. It adds those as negative keywords. It never blocks a term that converted in the last 90 days. It never touches budgets, bids, ads or keywords, so your target CPA stays your decision. Every block can be undone in one click for 90 days. If you'd rather do it by hand, here's how to undo negative keywords in Google Ads.

It won't fix a CPA problem caused by tracking, landing pages, an unrealistic target, or Performance Max spend on Display and Video inventory. Those have no search terms to read. What it does is keep the impossible-term share near zero night after night. When you sit down to change a target, you're making that decision on a clean number.