Why Your CPC Keeps Rising — and the Quality Score Fix That Works

Rising cost per click provokes one of two reactions. The first is fatalism: the auction is more competitive, costs go up, nothing to be done. The second is panic bidding — raise bids to hold position, watch efficiency collapse, then cut budget and lose volume. Both skip diagnosis.

The auction charges you based on the ad rank of the competitor immediately below you, divided by your own quality-adjusted rank. That formula means your cost is directly discounted by your own relevance. Two advertisers bidding identically can pay materially different prices for the same click. The gap is not luck — it is measurable and fixable.

The five reports that locate the leak

Before changing a single bid, run these in order. Each isolates a different cause, and the sequence matters — fixing search terms before fixing structure just moves the waste around.

1. Search terms report, last 90 daysSort by cost descending, filter to zero conversions. Every line is a query you paid for that produced nothing. This alone usually surfaces the first tranche of negatives.
2. Quality score weighted by spendAn average QS of 7 means nothing if your top-spending keywords sit at 4. Weight by cost, not by keyword count.
3. Auction insights, month over monthDistinguishes genuine new competition from your own impression-share decay. If overlap rate is flat and your CPC rose, the problem is internal.
4. Landing page reportConversion rate and bounce by destination URL. One weak page dragging down an entire campaign is extremely common.
5. Match type distribution by spendIf broad match consumes more than half the budget without a strong conversion signal steering it, you have found the leak.

Quality score, decomposed

The 1–10 number is a diagnostic, not a target. It is built from three components, each reported as below average, average or above average. Treat those three labels as your actual worklist.

Landing page experience deserves particular attention because advertisers most often treat it as somebody else's job. An ad promising "same-day commercial installation" that lands on a generic homepage will score below average no matter how good the copy is. The fix is not a redesign; it is a page that repeats the promise in the first screen and removes every path that is not the conversion.

Match types and negative keyword discipline

Broad match is not inherently wasteful — it is wasteful without a strong conversion signal to steer it. Modern smart bidding uses broad match as a discovery mechanism, which works when the model has enough conversions to learn from and fails expensively when it does not.

• Broad match with under 30 conversions a month — the model cannot learn; you are funding exploration you cannot afford.
• A negative list under 200 terms on a mature account — almost certainly incomplete.
• No negatives for "free", "jobs", "salary", "diy", "template", "reddit" — research intent draining commercial budget.
• Brand and non-brand in one campaign — brand's cheap conversions mask non-brand's real performance.
• No shared negative list across campaigns — you will fix the same waste five times.

Model profit, not ROAS

Return on ad spend is a revenue ratio. It says nothing about whether the business made money. Run your own numbers — the breakeven conversion rate output is usually the most sobering figure in any paid account review.

The bidding sequence that holds

1. Derive target CPA from marginGross profit per order × the share you are willing to spend on acquisition. That number, not a CPC ceiling, is the constraint.
2. Fix tracking before automatingSmart bidding optimises toward whatever you call a conversion. Counting newsletter signups as sales teaches it to buy the wrong traffic.
3. Consolidate for signalFifteen thin ad groups starve the algorithm. Fewer, better-fed campaigns learn faster.
4. Change one variable, then waitLearning periods run one to two weeks. Stacked changes make attribution impossible.
5. Re-audit search terms quarterlyQuery patterns drift. A negative list is a living document, not a project.

Limits, errors and borders of the standard advice

Critical review

Paid search guidance comes from the platform itself, whose revenue rises with your spend, and from agencies paid a percentage of that spend. Both are structurally biased toward more budget and less transparency.

"Improve your quality score to lower your costs."

The claimRaising the quality metric reduces cost per click, so it should be the optimisation target.

Where it breaksIt is a diagnostic that has been widely mistaken for a lever. The reported figure is a rounded, lagging summary computed against competing advertisers, so it moves for reasons unrelated to your changes, and chasing it produces activity without outcomes. The deeper error is optimising cost per click at all — a cheaper click on a worse-matched query is not progress. Accounts routinely improve the metric while conversions fall, because the tactics that raise it, like extremely narrow ad groups, can restrict the volume that made the campaign work.

The borderTreat a low score as a signal to inspect relevance between query, ad and landing page. Then optimise on cost per acquisition and incremental profit, and accept a higher click cost whenever the resulting customers are worth more.

"Use broad match and let the algorithm learn."

The claimMachine learning finds converting queries better than manual keyword selection.

Where it breaksThe automation optimises against the signal you give it, and most accounts give it a poor one — a conversion event that is not the actual thing of value, like a form fill rather than a paid customer. It then efficiently buys more of the wrong outcome. Broad matching also expands into loosely related queries where the platform earns revenue and you do not, and the reporting available has become progressively less granular, making the expansion harder to audit. Learning periods also consume real budget, and small accounts may never reach the volume where the models perform well.

The borderAutomation works well with high conversion volume, an accurate value signal fed back to the platform, and negative keyword discipline. On low volume or with a proxy conversion event, tighter match types and manual control usually outperform.

"Bid on your own brand terms — it's cheap traffic."

The claimLow-cost brand clicks protect against competitors and produce excellent returns.

Where it breaksMost brand clicks are incremental only if the person would otherwise not have reached you — and for a searcher typing your name, the organic result usually captures them at no cost. Reported returns on brand campaigns are therefore inflated by cannibalisation, which is why they always look like the best-performing campaign in the account. It is also the easiest way for an agency to demonstrate strong returns without creating value. The defensive argument has genuine force in some markets and is frequently overstated in others.

The borderBrand bidding is justified when competitors actively bid on your name, when the results page pushes your organic listing down, or when you need to control the message during a promotion. Test it by pausing and measuring total conversions, not campaign-reported conversions — the incremental figure is often far smaller than it appears.

"Scale what works — increase budget on winning campaigns."

The claimA profitable campaign should receive more budget until it stops being profitable.

Where it breaksPerformance is not linear in spend. The initial budget buys the highest-intent, cheapest-to-convert audience; additional budget necessarily reaches progressively less qualified users, so average metrics degrade as volume grows and the marginal customer can be unprofitable while the average still looks fine. Audience saturation and frequency effects compound this in smaller markets. Scaling also resets learning and can destabilise bidding, so a sharp increase sometimes damages a campaign that was working.

The borderScale in increments and judge on marginal return, not average — the question is what the last increment of spend produced. Expect a ceiling in any finite market, and treat a degrading average as the signal that you have found it.

Where our own position is weakest

We give no benchmark costs or rates, because they vary by industry, geography and auction dynamics — so this cannot tell you whether your numbers are good. Platform interfaces, match-type behaviour and reporting granularity also change frequently enough that specific tactics date quickly. And our scepticism about automation is increasingly contestable: the models have improved materially, and an advertiser with clean value signals may well do better trusting them further than we suggest.

Frequently asked questions

The bottom line

Auction inflation explains part of your rising CPC. Structural neglect explains most of it. Audit search terms, weight quality score by spend, tighten match types behind a real conversion signal, and make the landing page repeat the promise the ad made. Then bid to a target CPA derived from gross margin. Pair paid with an owned content engine so your acquisition cost is not permanently rented.

---

Revision log. Jul 2026: added profit model with breakeven conversion rate and refreshed match-type guidance. Reviewed quarterly.