Knowing what QS is, how it is calculated and what it is worth, what remains is the method. It is not new: everything in this lesson was known in 2014. What has changed is that almost nobody applies it, because it demands persistence rather than knowledge. Here it is in full: the golden rules and the recipe for a specific keyword.
The golden rules
Distilled from twenty years of accounts, with CTR as the goal of every one of them:
- Small, thematic ad groups — sibling keywords, not flatmates. If an ad group needs two different ads to cover its keywords, it is two ad groups.
- Avoid the generic in every link of the chain: general, uncontrolled broad keywords, catch-all ads, a "one size fits all" landing page, a bland display URL (remember: the display URL has its own CTR history).
- Review the search terms religiously: add the good stuff (to its own ad group), exclude the bad stuff with negatives. Every junk term that keeps coming in erodes the history.
- Test ads properly, looking at the CTR × conversion rate pairing, not at CTR alone (see below).
- Pause or restructure anything carrying a hopeless bad history: a low CTR with no conversions does not deserve to keep writing history against you.
- Assets always: they improve the CTR (which means improving the QS) and they go into the ranking formula.
The ad test that actually matters
With 1,000 impressions:
| Ad | CTR | Conversion rate | Sales |
|---|---|---|---|
| A | 8% | 4% | 80 clicks → 3.2 |
| B | 6% | 7% | 60 clicks → 4.2 |
The "worse" CTR wins the business. And the QS? It is offset by the volume of good history across the rest of the chain: an ad that brings in the right buyer sustains the landing page, the conversion rate and, in the medium term, the account's own CTR. The rule: optimise the pairing, never CTR in isolation. A clickbait ad raises the QS and sinks the business.
The classic recipe for lifting a specific keyword
For a keyword with traffic that is worth rescuing:
- Pull it out into exact match in its own ad group. It stops sharing an ad and a history with its neighbours.
- Write an ad and assets aimed only at it: headline 1 with the keyword, a specific benefit, its own sitelinks and callouts.
- Analyse its CTR by geographical area (the user location segment) and exclude the areas that are below average.
- Do the same with the demographics (age, gender) if the campaign shows them: exclude the segments with the worst CTR.
- Add all the audiences in observation mode and, after a few weeks, exclude the ones with the worst CTR (or lower their adjustment if bidding is manual).
- Review its landing page against the penalty list (lesson 3) and its international speed.
- Wait. Weeks to have data, months to see the visible QS rise. Note it in the change log with the date.
It works. And here is the method's small print: it is only worth it on keywords with a fair amount of traffic, it will take you weeks to have data and months to see the QS rise. Now multiply that procedure by the dozens or hundreds of keywords in an account, add the continuous watching of terms, areas and components… and you will understand why, in the age of automation, almost nobody does it any more. It is not that the method has stopped working. It is that doing it by hand stopped being viable.
Prioritising: which keywords get the recipe
Not all of them. In order of impact:
- Keywords with high spend and a QS ≤ 4: the quick money (lesson 4).
- Keywords with a QS of 5-6 and a lot of spend with one clear "below average" component: a single specific action lifts them.
- Good terms sitting "in the bag" of another keyword and converting: pulling them out is the coverage tactic (lesson 3).
- The rest: leave them alone. A QS of 6 on a keyword spending €20 a month is not worth an hour of work.
The bottleneck
Quality Score was never a knowledge problem. It is a persistence problem: watching hundreds of keywords, three components, per-term coverage, areas and landing pages every day is machine work that was asked of people for years. People, understandably, stopped doing it. The next lesson is the plan for doing it with a machine.
💡 Ninja trick: every step of the recipe has its own tab in the QS Analyzer: critical QS (step 1 of the prioritisation), keyword and negative suggestions (pulling out of the bag), charted components (what to fix), AI landing page analysis (step 6), and the TODO tab that sorts everything by impact in euros. And its sibling, the RSA Optimizer, does step 2 — ads written for each ad group — with the relevance the QS Analyzer itself measures.
What you should remember
- Six golden rules; the goal of all of them is CTR (properly understood: CTR × conversion).
- The recipe: exact in its own ad group → its own ad and assets → areas → demographics → audiences → landing page → wait.
- It is only worth it on keywords with traffic; prioritise by spend × low QS.
- The method works; what does not work is doing it by hand at scale.