What is Quality Score?

Quality Score is a diagnostic Google Ads reports per keyword on a one-to-ten scale. Microsoft Advertising reports its own equivalent on the same principle. It summarises three estimates: how likely the ad is to be clicked when shown for that keyword, how closely the ad matches the intent behind it, and how useful the destination page is to somebody who clicked.

The critical framing, and the one most often missed: the reported score is not the thing the auction uses. The auction computes its own estimates live, conditioned on the specific query, the device, the location, the time and everything else about the moment. The score is a periodically refreshed, rounded-off summary of the same underlying signals, produced so that an advertiser has something to act on.

That distinction explains several things that otherwise look like inconsistencies. A keyword can hold a score of 7 and perform differently across two queries it matches. A score can sit still while costs move. And a change to an ad can lower costs before the number updates.

Each is reported as below average, average or above average against other advertisers competing on the same keyword, which makes it a relative measure rather than an absolute one.

Expected click-through rate. How likely this ad is to be clicked if shown for this keyword, historically and predictively. Google does not publish how the three components are weighted against each other, so any claim about which matters most is an inference; what is stated is that the auction ranks on expected value, and a platform paid on clicks earns more from an ad it expects to be clicked. It is also the component most responsive to the ad text itself.

Ad relevance. How closely the ad matches the intent behind the keyword. This is a question about proximity: does the ad say what the query asked about, in the words the query used. It responds to structure more than to writing — a keyword sitting in a group of thirty loosely related terms served by one generic ad will report poor relevance however well that ad is written.

Landing page experience. How useful and relevant the destination is to somebody arriving from that query. Speed, whether the page answers the question the ad raised, whether the content is substantive, and whether navigating it is straightforward.

Auctions rank on expected value rather than on money. A platform paid on clicks earns the bid multiplied by the probability of a click, so an ad more likely to be clicked is worth more at any bid.

The consequence is the reason the score is worth attention: improving the things it measures wins more auctions and lowers the clearing price of the auctions already being won. A bid increase does only the first, and pays for it.

That is the entire economic case, and it is a strong one. It also explains why the score is worth improving even on keywords that already perform — a working keyword with poor quality signals is a keyword the account is overpaying for.

There is no control that sets the score. It is computed from behaviour, which means every intervention is indirect and lagged.

That produces two common errors.

The first is treating the number as a target. Accounts that chase the score for its own sake end up optimising a diagnostic, and the work that moves it is worth doing only because of what it does to costs — which is measurable directly.

The second is impatience. The components are built from accumulated history, so a change made on Monday does not show by Friday. Judging a copy change too early and reverting it is the most common way an account stays uncompetitive: a sequence of good changes, each abandoned before it registered.

Read the components, not the score. The number is three signals compressed into one, and the three have completely different remedies. An account looking at a 4 knows nothing; an account looking at a 4 with below-average relevance knows to open the ad group structure.

Read it beside cost and value. The score says nothing about conversion rate, revenue, or whether the traffic is worth having. A low-scoring keyword with a good cost per acquisition is an opportunity to pay less for something that is already working, not a candidate for pausing.

Weight by impressions. An account-wide average across keywords with no volume is meaningless. The keywords carrying the spend are the ones whose quality signals are costing money.

Expect it to differ by match type. A broad-matched keyword serving a wide range of queries will report weaker relevance than an exact-matched one, not because the ad is worse but because it is being asked to be relevant to more things.

Structure first: fewer, tighter groups, so each ad serves queries it can speak directly to.

Then the ad text: repeat the language of the queries, answer the intent in the headline rather than the description, and test against click-through rate specifically.

Then the page: speed, and whether it answers the question the ad raised. This is the component most often owned by somebody outside the advertising team and least often checked.

Then subtraction: negatives on queries the ad was never written for, which improves relevance by removing the traffic that was dragging it down rather than by writing anything new.

For all the ways it is misread, the diagnostic earns its place in two specific jobs.

The first is triage across a large account. Sorting the keywords that carry the spend by their weakest component produces a work queue in a couple of minutes: these groups have a relevance problem, those landing pages need attention, this set of ads is not earning clicks. No other single report organises the work that way.

The second is settling an argument about where the money is going. An account paying more than a competitor for the same position has a reason, and the components name it. That is a more productive conversation than one conducted entirely in bids.

It does not measure whether the traffic converts, what it is worth, or whether the keyword should exist. Those are business questions and the score is indifferent to all of them.

It does not measure the account. The three components are relative to other advertisers competing on the same keyword, so a score is a statement about a competitive set rather than about absolute quality, and a set that gets stronger will lower a score that did not change.

And it does not measure what the auction did. The live estimates move with the query, the device, the location and the moment; the reported number is a periodic, rounded summary. Treating a whole number that updates occasionally as the live input to a real-time auction is the root of most of the confusion about why costs and scores do not track each other.

• Work the component that is reported as below average rather than the score. The three parts have different remedies and the number on its own says nothing about which one to open.
• Tighten the grouping so the ad can repeat the language of the queries it serves. Relevance responds to proximity between query, ad and page more than to any other single change.
• Test headlines against expected click-through rate specifically, because it is the component the ad text moves most directly.
• Fix the landing page for speed, for relevance to the query, and for whether it answers the question the ad raised. It is the component most often left to somebody else and least often checked.
• Add negatives and remove queries the ad was never written for. A keyword forced to serve traffic it does not match will report poor relevance no matter how the ad is written.