What you'll learn in this article
- The six causes of runaway CPC I actually find when I open an account, ranked by how often
- The intervention order I run and why cutting the bid first usually makes things worse
- How I read a CPC increase to tell competitor pressure apart from my own account decaying
- The two-week diagnostic that separates a query problem from a quality problem
- The situations where high cost is correct and lowering it would be the mistake
The message always arrives the same way: "our google keyword cost per click has doubled, can you look". And in almost every account I've opened after that message, the bid was not the thing that changed. Something upstream of the bid changed the query mix, the ad quality, the auction, the match type, or a bid strategy quietly doing exactly what it was told to do.
What follows is the diagnostic order I actually run, the six causes I find in practice, and the concrete edits I make in each case. Not "improve your Quality Score". The specific intervention, in the order I make it, and what I expect the number to do afterwards.
What actually sets your cost per keyword in Google
You have to be precise about which number you're panicking about, because max CPC and actual CPC behave nothing alike and only one of them is a symptom of anything.
Your max CPC is a ceiling you set. What you're charged is the actual CPC, and it's determined at auction time. Google's documentation on actual cost-per-click is explicit about the mechanism: you pay the minimum required to clear the Ad Rank thresholds and beat the advertiser immediately below you, with ad quality, auction competitiveness and search context all feeding the calculation.
Three things move the price, and only one is your bid
The first is what your ads are worth to the auction expected clickthrough rate, ad relevance, landing page experience, and the expected impact of your assets. The second is who else is bidding and how far apart your Ad Ranks sit; the documentation notes that as the gap between two advertisers widens, the higher-ranking one becomes more likely to win but may also pay more for that certainty. The third is your bid, and it is the lever most people pull first and the one with the least durable effect.
Where you show is part of the price
Ad Rank thresholds are higher above the results than below them, so the same click bought at the top costs more than the same click bought lower on the page even with nobody directly beneath you. A cost per keyword Google reports as "rising" is sometimes just an impression-share shift you asked for by changing a bid strategy target.
The two numbers I pull before touching anything
Average CPC segmented by week for the last ninety days, and the search terms report for the same window. If CPC moved but the query mix moved with it, that's a targeting problem. If CPC moved and the queries stayed identical, that's an auction or a quality problem. Everything downstream depends on which of the two it is.
The six causes of a runaway google keyword cost per click I actually find
Ranked by how often I find them, not by how interesting they are.
1. The query mix drifted, not the auction
This is the majority of cases. A phrase or broad keyword started matching a wider set of queries usually more commercially generic ones and generic queries are expensive. The keyword report shows one row and one CPC; the search terms report shows fifteen queries with wildly different costs behind that row. The intervention is not a bid change. It's tightening what the keyword is allowed to match, which almost always means work on the account's negative keyword layer before anything else.
2. Ad quality decayed while nobody was looking
CTR falls slowly a competitor rewrote their headlines, a seasonal angle went stale, the landing page got slower after a template change. Quality inputs are recalculated continuously, so a gradual relevance decline shows up as a gradual CPC climb with no event to point at. If your CPC rose 20% over four months with a flat query mix and flat impression share, this is nearly always it, and I go and read what weak quality signals cost you at auction alongside the account's own CTR trend before doing anything else.
3. A bid strategy target got changed and nobody logged it
Someone raised a tROAS, loosened a tCPA, or switched a campaign to maximise clicks. Change history, ninety days, filtered to bid strategy this takes forty seconds and I find something roughly one time in four. Automated strategies buy exactly what you told them to buy; a target CPA raised by 30% is an instruction to pay more per click.
4. Internal competition between your own keywords
Two ad groups holding near-identical terms don't bid against each other directly, but they do fragment performance data across duplicated rows, which weakens the signal each one accumulates. Consolidating the duplicates into one better-fed ad group is frequently worth more than any bid edit.
5. A new competitor with different economics entered
Auction insights, month over month, is where this shows: a new domain appearing with meaningful impression share, or an existing one jumping their overlap rate. This is the one cause you cannot fix someone with a higher customer lifetime value can rationally outbid you forever. The response is a narrower keyword set, not a bigger budget.
6. Position creep from an impression-share strategy
Target impression share set to absolute top will buy the absolute top, and the top costs more by design. I see this set optimistically at launch and never revisited. If CPC rose while top impression share rose in lockstep, you didn't lose control of the price you bought a more expensive slot on purpose.
The intervention, in the order I run it
Order matters more than any individual edit here. Cutting the bid first is the most common mistake and it usually raises cost per conversion while lowering cost per click, which is a worse outcome dressed up as a better metric.
Day 1 Strip the queries, not the bid
Search terms report, last sixty days, sorted by cost descending. Every query above your target CPA with zero conversions gets excluded as an exact-match negative. Every recurring irrelevant pattern gets a phrase negative. On a neglected account this alone takes 15–25% off average CPC within two weeks, because you stop buying the expensive tail entirely rather than buying it more cheaply.
Day 1 Tighten the match types on the worst offenders
The three or four keywords carrying the most wasted spend get moved from broad to phrase, or phrase to exact. I don't do this account-wide; I do it surgically, on the rows the cost data indicts. Broad match with a well-fed smart bidding strategy is fine. Broad match with thin conversion data is an expensive experiment.
Day 2 Fix the ad, because it's the only permanent discount
Better ad quality earns lower actual CPC at the same bid that's the mechanism, not a growth-hack claim. So I rewrite headlines to mirror the top three converting queries verbatim, pin the primary term into headline position one where relevance is weak, and check landing page load time. This is the slowest lever and the only one whose effect compounds.
Day 3 Then, and only then, adjust bidding
If I'm on manual or eCPC, I cut max CPC on the high-cost low-conversion rows by 15%, never more, and never across the board. If I'm on smart bidding, I don't touch the bids at all I tighten the target by 10% and wait a full learning cycle. Aggressive target cuts starve the strategy and you lose the conversion volume that was keeping the CPC honest in the first place. The full logic of how to set the bid itself is a separate decision from what I've done above, and I keep them separate on purpose.
Day 14 Measure, and attribute the movement
I compare average CPC, conversion rate and cost per conversion against the previous fourteen days, and I write down which of the four edits I believe moved which number. If CPC fell and conversions fell proportionally, I bought less traffic rather than cheaper traffic and I've solved nothing. That distinction is the entire point of the exercise.
What I infer from how the cost moved
CPC up, impression share flat, queries unchanged: the auction got more expensive. Someone is paying more for the same slot. Auction insights will name them. There is no account-side fix, only a decision about whether these terms are still worth buying.
CPC up and top impression share up together: you're buying a better position. Check the bid strategy target before treating this as a problem. Frequently it's a setting someone changed deliberately and forgot to mention.
CPC up with CTR down: this is quality, and it's your side of the ledger. The most fixable cause on this list, and the slowest to fix. Expect four to six weeks before the auction fully re-prices you.
CPC up, conversion rate up, CPA flat or down: nothing is wrong. You're buying more expensive but more qualified clicks. I've watched people optimise this back to a cheaper, worse account. Cost per click is a diagnostic metric, not a goal.
CPC spiked on a single day then normalised: check the calendar, not the account. A competitor promotion, a news event, a seasonal spike. One day is never a trend and I don't act on it.
CPC below forecast but volume collapsed: you've fallen out of the top slots. Cheap clicks at the bottom of the page are cheap for a reason. This is the failure mode of an over-aggressive bid cut, and it's why I limit cuts to 15%.
CPC identical but total cost up: your click volume grew. A budget question, not a pricing question. Worth naming clearly, because these two get conflated in almost every meeting I've sat in.
What I stopped doing
Cutting bids as the first move. It treats the symptom, costs you position, and hides the underlying cause for another month. Bids are the last edit in my sequence for a reason.
Chasing a target CPC number. A cost per keyword Google charges you is only meaningful next to what the click produces. I've never once been asked to explain a high CPC on an account that was profitable.
Pausing every expensive keyword. Expensive frequently means high-intent. I pause on cost per conversion, never on cost per click, and the two lists overlap far less than people assume.
Rebuilding an account structure to solve a pricing problem. I did this twice early on. Both times the actual cause was query drift, and the rebuild cost six weeks of learning data to fix something a negative keyword list would have solved in an afternoon.
Reacting inside a smart bidding learning period. Two weeks of elevated CPC after a strategy change is the system working, not failing. Intervening mid-learning resets it and guarantees a third expensive week.
Benchmarking against industry CPC averages. They're an average of accounts with different quality, geographies and offers. Your own trailing ninety days is the only benchmark that knows anything about you.
The practical takeaway
When the cost per click climbs, resist the bid. Pull the search terms report first: if the queries changed, it's a targeting problem and negatives fix it in two weeks. If the queries held steady and CTR fell, it's quality, and only the ad and landing page fix it. If both held steady, it's the auction, and the honest answer is a narrower keyword set rather than a bigger budget.
The order is queries, then match types, then ad quality, then bids. Reversing that order is how a cost problem becomes a volume problem on top of a cost problem.
And measure the outcome on cost per conversion, not on the click price. Halving what you pay per click while halving your conversion rate is a report that looks like a win and isn't one.