What follows explains most of the "inexplicable" Quality Scores: the 6/10 on a keyword that performs beautifully, the 10/10 that suddenly drops to 4, the whole ad group that falls without anybody touching a thing. These are the rules Google documented back in the day and that almost nobody applies any more.
Every search term has its own hidden QS
The visible QS of your keyword corresponds only to the exact match. But every different term that triggers that keyword carries its own internal Quality Score. An example from the classic system:
| Search term | Keyword | Visible QS | Real internal QS |
|---|---|---|---|
| hotels london | hotels london | 7/10 | 7.542 |
| hotels in london | hotels london | 7/10 | 9.950 |
| hotels in london | hotels in london | 10/10 | 9.950 |
The reading: the same traffic («hotels in london») scores 9.950 whichever way you look at it; but as long as it comes in "through the bag" of another keyword, you see a 7/10 while the system applies the term's QS, not the bag's. From this comes one of the most profitable tactics of the classic QS, and one more relevant than ever now that match types have been inflated: pull the good terms out of the bag and give them their own exact keyword, their own ad and their own history.
One QS per device, per network, per area and per time of day
- Devices: desktop, mobile and tablet calculate separate Quality Scores. The number you see when advertising on all of them at once is a blend that does not exist in any specific auction.
- Networks: Google search, search partners and Display each have their own QS, and performance on one does not contaminate the others. The visible QS is only the Google search one. Careful: the partners' CTR is usually worse and dilutes the CTR you see in the column — an "average" 3.4% could be a real 5% on Google. Segment by network before judging a keyword.
- Geography: CTR is also evaluated by location. An area with a bad CTR penalises every keyword targeting that area, including your best ones. Watching geographical performance (and excluding bad areas or splitting them into their own campaigns) is pure QS optimisation, even if nobody calls it that.
- Time of day: the hidden QS can vary by hour. A bad night-time CTR is not "free".
What does NOT affect it (and you have been told it does)
| Factor | The reality |
|---|---|
| Conversions (number or rate) | They do not enter the calculation. Google recommends them as a metric for improving the landing page, nothing more |
| Pausing or deleting a keyword | It changes nothing in itself: the history belongs to the system, not to the row. Pausing the bad stuff prevents future harm; it does not erase the past |
| The match type you choose | The keyword's QS does not depend on exact/phrase/broad; what changes is which terms come in (and those do bring their own QS) |
| Performance on search partners | It does not affect your QS on Google search |
| Old versions of the landing page | Only the latest one counts: rebuilding a bad landing page cleans you up; the system does not hold a grudge |
| Editing ads | An edited ad keeps its history; it is not reset. Do not be afraid of testing (the thing that does go into learning is Smart Bidding) |
Vertical and horizontal relevance: the one you control and the one you do not
- Vertical: keyword ↔ ad ↔ landing page inside your ad group. This one you control: thematic ad groups, related keywords, a specific ad. Every "intruder" keyword breaks the chain a little; broad without control breaks it on its own by bringing in terms that have nothing to do with you.
- Horizontal: what your keyword means to the rest of the world. If the term has several meanings (a brand that is also a common word, «pool rehabilitation»...), some of the searches will never be for you, the term's overall CTR suffers and your QS with it. This one you do not control: all you can do is recognise it and not fight it — split it out, fence it in with negatives, and do not beat yourself up over a 6/10 that is not your fault. A real case: a brand keyword with a 10/10 dropped to 4/10 when the term became popular with another meaning, and it took years to settle at 6/10 without anything in the account having got worse.
The landing page: it can only hurt you
Landing page experience works differently from the other two factors: there is no landing page "prize" that raises your QS; there is only pass, penalised or suspended. It is assessed when the page goes live, it is applied at ad group level (combining the ad group's landing pages with the rest of the domain's in the account) and its list of penalties is specific: impossible promises, pop-ups, slow loading from the robot's country (the bots come in from several countries: measure your international speed, not just your local one), sneaky redirects, duplicate content, a lack of transparency (no privacy policy, no "about us")... and one silent classic: blocking AdsBot-Google in robots.txt, which sinks the QS of the whole account without ever owning up to it. Serious sanctions escalate: ads disapproved, account suspension, domain suspension.
💡 Ninja trick: almost all of this small print is about coverage: how much of your real traffic comes in through keywords with a good QS and how much comes in "through the bag". The QS Analyzer calculates that coverage term by term, proposes pulling the good ones out of the bag (exact keyword + ad group + ad), spots the areas and devices that dilute the CTR and reviews the landing pages against the penalty list — robots.txt included.
What you should remember
- Every term has its own hidden QS: pull the good ones out of the bag.
- Separate QS figures by device, network, area and time of day; the visible one is a blend.
- They do not affect it: conversions, pausing, the match type, the partners, old landing pages, editing ads.
- Vertical relevance you control; horizontal you do not: split it out and do not beat yourself up.
- The landing page can only penalise: international speed, transparency, robots.txt.