The Complete Google Ads Quality Score Guide: What It Measures, What Moves It, What Doesn't

The short answer

Quality Score is a keyword-level diagnostic in Google Ads, reported on a 1-10 scale, that summarizes three separate estimates: expected click-through rate, ad relevance and landing page experience. Each of the three is also reported on its own as Below Average, Average or Above Average.

That is what it is. What it is for is the part almost every explanation skips, and it matters more. Quality Score is a diagnostic summary, not a lever. There is no such thing as Quality Score work. There is landing page work, ad group structure work and ad copy work, and the score reports back on how those three are going. Treating the number itself as the target is the error that produces a year of activity with nothing underneath it.

The most useful mental model is a dashboard warning light. It tells you where to look. It does not tell you what is wrong, it does not tell you how much the problem is costing, and turning the light off is not the same as fixing the engine. Every genuinely productive use of this feature starts by ignoring the composite number and reading the three components underneath it.

Where Quality Score actually enters the auction

The mechanism that gives this metric real financial weight is Ad Rank. Every time a search happens, Google computes an Ad Rank value for each eligible advertiser and uses it to decide whether an ad shows at all, in what position, and what the advertiser is charged if the click happens. Ad Rank draws on the bid, on quality signals computed at that moment, on the context of the search — device, location, time, the exact query typed — on the expected impact of assets and formats, and on the rank thresholds that govern eligibility.

Notice what is not in that description: a published formula. Google has never given advertisers one they can compute, and any guide that hands you an equation is inventing the precision. What can be stated honestly is the direction of the relationship, and the direction is enough to act on.

Why relevance shows up as a discount

An advertiser does not pay their bid. They pay something close to the minimum needed to clear the ad ranked immediately below them. That single fact is where quality turns into money. If your quality contribution is stronger, you need less bid to reach the same rank as a competitor — which means you can hold the same position for less, or hold a better position for the same amount. If your quality contribution is weaker, the auction charges you more to sit where you were sitting before.

So relevance is simultaneously a position lever and a price lever, and that is the reason this feature deserves attention even though the reported number itself is close to useless. The thing worth internalizing is that the discount is real and unquantifiable from the interface. Anyone who tells you what a specific score is worth per click has made the figure up.

The one distinction that clears up most confusion

The Quality Score column and the quality signals used at auction time are not the same object. The column is a historical summary attached to a keyword. The auction evaluates the specific query, on the specific device, in the specific context, in real time. They are related, and they are not interchangeable, which is why an account can show a healthy column and still lose auctions it should win.

Component 1 — Expected click-through rate

Expected click-through rate estimates how likely your ad is to be clicked when it shows for a given keyword. It is the component people find most intuitive and the one they most consistently misread, because of one property: it is normalized for position.

Why it is relative rather than absolute

Ads in higher positions get clicked more, regardless of how good they are. If the estimate were absolute, it would mostly be measuring how much everyone was bidding, and it would reward spending rather than relevance — a circular metric that told you nothing. Normalizing for position removes that. The estimate is a judgment about the ad as an ad, held against what ads generally achieve in that slot for that keyword.

This changes what a Below Average label means. It does not mean your click-through rate is low in absolute terms. It means your ad underperforms what an ad in that position, on that keyword, against that competitive set, would normally achieve. A generic keyword with a crowded results page and a specific long-tail keyword with two competitors are being graded against completely different reference points, and comparing their labels to each other is meaningless.

How new keywords with no history are handled

A keyword added this morning has no click history, yet it can carry a rating almost immediately. The estimate in that case is a prior, not a measurement — inferred from related signals such as the performance of similar keywords, the history of the account and the domain, and the ad copy itself. That is a reasonable engineering decision and a terrible thing to make decisions on.

The practical consequence: on a keyword with thin impression volume, the component label is closer to a guess than a finding, and it will move as data accumulates without anybody changing anything. Reacting to it is how managers end up chasing their own noise. Check impression volume before you believe any component label, every single time.

What genuinely moves it

• Language that matches how the searcher phrased the need, in the headline rather than buried in the description — the searcher scans headlines and reads descriptions only after something in a headline earned it.
• Differentiation against the specific ads sharing that results page. If four advertisers say the same three things, the fifth saying them again has no reason to be clicked. This is competitive work, not copywriting work, and it requires actually looking at the page.
• Specificity that a competitor cannot honestly copy — a real service boundary, a real location, a real qualification. Vague superlatives are available to everybody and therefore differentiate nobody.
• Assets and formats that give the ad more surface area on the results page, which changes both how the ad appears and what it can say before the click.

And here is the trade this component quietly invites you to make badly. It is entirely possible to raise expected click-through rate by writing a vaguer, more curiosity-driven ad that attracts people who are not going to buy. The label improves. The account gets worse, because you are now paying for a wider, less qualified set of clicks. If an ad rewrite raises clicks and lowers qualified volume, that is not a win with a caveat — it is a loss with a flattering label.

Component 2 — Ad relevance

Ad relevance assesses whether the language in your ad corresponds to the intent behind the keyword. It is the component that is cheapest to fix, most often misdiagnosed as a copy problem, and least worth obsessing over once it reads Average.

The mechanism is structure, not wording

Ads are written at the ad group level, and every keyword in that ad group is served by the same set of ads. So ad relevance is bounded by a structural fact before a single word is written: how many different intents share the ad group. If one ad group contains a keyword about emergency same-day service, a keyword about annual contracts and a keyword about pricing, no ad can be directly responsive to all three. One of them will always be reading an ad written for somebody else.

This is why Below Average ad relevance is usually a structure statement rather than a copy statement. Rewriting the ad inside a badly themed ad group moves the problem from one keyword to another. Splitting the ad group by intent removes it. Tight thematic grouping is the actual mechanism; ad copy is just what makes the grouping visible.

Where the "one keyword per ad group" advice goes too far

The logical endpoint of tight grouping is one keyword per ad group, and for a while that was standard advice. It does exactly what it promises for this component, and it causes three problems that cost more than the component is worth.

1. It fragments conversion data. Automated bidding models learn from volume, and dividing an account into a large number of tiny ad groups spreads the same conversions across many thin buckets. The structure that is best for a relevance label is frequently the structure that is worst for the bidding system actually spending the money.
2. It makes honest ad testing impossible. Comparing two ad variants requires enough impressions in one place to tell a difference from noise. An ad group receiving a trickle of traffic will never produce that, so every test result becomes a coin flip somebody wrote down as an insight.
3. The isolation is partly notional anyway. Close-variant matching means a keyword catches queries that are not literally the keyword, so the perfect one-to-one correspondence the structure was built to guarantee does not actually hold at the query level.

The workable rule is to group by intent rather than by string. If one honest ad can speak directly to a searcher arriving from any keyword in the group, the group is correctly built and should be left alone. If writing that ad requires hedging so that it fits several different needs, the group holds more than one intent and should be split. That test scales in both directions — it tells you when to split and, just as usefully, when to stop splitting.

Component 3 — Landing page experience

Landing page experience assesses the destination rather than the ad: whether the page delivers what the ad promised, whether it is transparent and easy to move around, and whether it works well on the devices real traffic arrives on. It is the only component that is not primarily about words, and it is the one most likely to sit Below Average for months because it belongs to somebody who does not work on the ads.

Why speed and mobile experience matter here specifically

Every other part of this metric is a judgment about correspondence. This part includes a mechanical reality: a page that takes too long to become usable on a mobile connection is a page some share of your clicks never actually see. You have already paid for those clicks at the moment the auction cleared. A slow page does not reduce your costs in proportion to the visitors it loses — it converts spend into nothing, silently, and the ads interface will show you a click for every one of them.

Mobile matters twice over, because the phone is where the gap between "the page works" and "the page works for a stranger in a hurry on a mediocre connection" is widest. A page checked only on the desk it was built at is a page nobody has actually tested.

Matching in the first screen, not somewhere on the page

The distinction that decides this component in practice is where the correspondence lives. A page that mentions the service the ad promised, three sections down, past a company history and a rotating banner, technically matches the ad. It does not match it in the way that counts, because the visitor makes a stay-or-leave judgment in the first moments and never reaches the section that would have satisfied them.

The promise has to be visible in the first screen: the same service named the same way, the same qualifier if the ad carried one, and an obvious next action. This is not a design preference. It is the difference between a page that continues the conversation the ad started and a page that asks the visitor to start over, and the second one produces exactly the immediate-return behavior that degrades this component and your conversion rate at the same time.

The prerequisite nobody wants to hear

Landing page experience is frequently not an ads problem at all. It is a web development problem, a content ownership problem, or an organizational problem where nobody has the access or the mandate to change the page the ads point at. If that is the situation, no amount of keyword and copy work will move this component, and pretending otherwise wastes months. The honest move is to name it as a dependency with an owner and a date, or to accept the component as it stands and stop reporting on it as though it were being worked.

Reading the diagnostic properly

The three components are reported as Below Average, Average or Above Average, and each label is relative — it positions you against other advertisers competing on the same keyword over a trailing window. Average therefore is not a failing grade. It is the middle of your competitive set, and on a keyword where everybody has done the obvious work, the middle is where competent advertisers sit.

The composite number, meanwhile, is nearly useless on its own. Two keywords can carry the same score for entirely different reasons: one with a strong ad and a slow page, another with a good page and an ad group holding six unrelated intents. The composite reports the same value and prescribes nothing. The three labels underneath tell you which team should be doing which work this week.

There is exactly one defensible use of the composite: as a sorting aid. Sort keywords by cost, look at the ones carrying real spend, and use the composite only to decide the order in which you open them. Every decision after that comes from the components.

Historical column versus auction-time signals

The column in the interface is a summary of accumulated history. The auction, as described earlier, evaluates the query in front of it right now — the exact words typed, the device, the location, the time, the context. The column moves slowly and after the fact. The auction has already happened.

This has a concrete consequence for how you judge your own work. When you rewrite an ad or ship a faster page, the column will not repoint immediately, because it is still averaging in the period before the change. If you wait for the column to confirm the fix, you will conclude the fix did not work, revert it, and be wrong twice.

Judge the change by the thing you changed it to affect. A page speed fix is verified in page speed measurement and in the behavior of arriving traffic, not in a Quality Score column weeks later. An ad group split is verified by whether the new ads speak directly to their keywords and by what the traffic does after clicking. The column is a slow, lossy confirmation of things you should already be able to see more directly.

The diagnostic workflow

This is the sequence I use when opening an account where somebody has flagged Quality Score as a problem. It is deliberately ordered, and the order matters more than any individual step. Before any of it, conversion tracking has to be trustworthy — the conversion tracking audit comes first, because every judgment below about whether a keyword is worth its cost depends on the conversion data being real.

1. Sort by cost, not by score. Open the keyword view, sort descending by spend, and work only on keywords carrying meaningful budget. A keyword with a poor score and negligible spend is not a problem; it is a rounding error with a label attached. Sorting by score is how managers spend a morning improving things that were never costing anything.
2. Add the three component columns and hide the composite. This single interface change alters what the account looks like more than any other step here, because it converts a vague number into three specific accusations you can act on.
3. Check impression volume before believing any label. On thin volume the estimate is inference rather than measurement, and it will drift on its own. If the keyword has not accumulated enough impressions to produce a stable read, the correct action is to leave it alone and come back.
4. Read landing page experience first. It is shared across many keywords and often across whole campaigns, so a single fix propagates further than anything else on this list — and because a weak page is also suppressing conversion rate, the same work pays twice.
5. Read ad relevance second, and treat Below Average as a structure finding. Before rewriting a single headline, list the distinct intents in the ad group. Most of the time the ad is fine and the group is holding jobs that belong to two or three groups.
6. Read expected click-through rate last, because it is the most confounded. Position, competitive set, seasonality, format eligibility and match type all move it for reasons unrelated to how good the ad is. Acting on it before the other two is how good ads get rewritten to solve problems they did not cause.
7. Ask whether the keyword should exist. Below Average on all three components, sustained, on a keyword with real volume, is often not a quality problem — it is the account telling you that this keyword attracts people you do not serve. Removing it is a legitimate outcome of this workflow, not an admission of defeat.
8. Change one thing and write down what and when. Component labels move slowly and for several reasons at once; without a dated record of what changed, you will not be able to attribute any subsequent movement to anything. This is the same discipline the bid governance model applies to bid changes, and it exists for the same reason.

The improvement playbook

Concrete actions per weak component, in the order that returns the most for the least. Nothing here is exotic, and that is the point — the accounts with strong component labels are not doing clever things, they are doing ordinary things consistently.

If landing page experience is weak

1. Measure the page loading on a phone on a normal connection, not on the machine it was built on. Fix whatever dominates that measurement first — usually oversized images, render-blocking scripts or fonts arriving late.
2. Put the promised service, in the ad's own words, in the first screen, along with the single next action you want taken. Remove anything above it that is about the company rather than the visitor.
3. Make the page match the ad's qualifiers. If the ad said same-day, the page has to say same-day. A page that quietly drops a qualifier the ad carried is the most expensive kind of mismatch, because the click was bought on a promise the page does not keep.
4. Give every campaign a destination that corresponds to its theme. Pointing four differently themed campaigns at one homepage guarantees this component sits weak across all of them, and no ad copy work will change it.
5. Check the page on the browsers and screen sizes your traffic actually uses, then check that forms submit and that phone numbers dial on a phone. Broken conversion paths degrade this component and your revenue at the same time, and they are invisible from the desk.

If ad relevance is weak

1. Export the ad group's keywords and sort them by intent rather than by volume. The split lines will be obvious once they are in front of you, and they are almost never where the original build put them.
2. Split until one honest ad can serve every keyword in the group — and then stop. Splitting past that point buys a label and costs you the data volume that automated bidding and ad testing both depend on.
3. Write the headline for the intent, not the string. Stuffing the exact keyword into a headline is a decade-old habit that produces awkward ads; a headline that plainly answers what the searcher wanted is more relevant in the sense that matters.
4. Read the search terms report before rewriting anything. Ad relevance is judged against the keyword, but the searchers arriving are on queries the keyword matched, and those often reveal that the intent you are serving is not the intent you thought you bought.

If expected click-through rate is weak

1. Search the keyword yourself and read every competing ad. This is the step most people skip and the one that supplies the answer, because the component is a comparison and you cannot reason about a comparison you have not looked at.
2. Find the claim that is true for you and unavailable to them, and put it in a headline. Coverage area, response time you actually hold to, a service boundary, an accreditation, a guarantee you honor.
3. Use qualifiers deliberately, understanding the trade. Adding price framing or a scope limit will reduce clicks and improve their quality. That is often the right call and it will make this component look worse. Decide which one you are managing before you make the change.
4. Make full use of the assets and formats available to the ad, so it occupies more of the results page and answers more before the click.
5. Give a change enough time to be measured against something other than noise, and hold the rest of the account still while you do it — the same governance logic that applies to bids applies to copy.

When to ignore Quality Score entirely

The honest scope of this metric is narrower than its prominence in the interface suggests. There are several situations where the correct action is to read the label, understand why it says what it says, and then do nothing.

• Low-volume keywords. Below the impression volume needed for a stable estimate, the labels are inference and they will move without you. Acting on them adds work and removes nothing.
• Newly launched keywords and new accounts. Early labels lean on priors rather than your own history. Give them the traffic before you give them a verdict.
• High-intent, bottom-of-funnel keywords in competitive sets. A narrow, purchase-ready keyword in a crowded auction can carry a weak expected click-through rate label permanently and still be the most profitable line in the account. A lower score on a keyword that produces revenue is a trade worth making every time.
• Keywords where the results page is dominated by aggregators and comparison sites. You are being graded inside a competitive set you cannot resemble, and the label reflects the company you are keeping rather than the quality of your ad.
• Cases where the only real fix is a page the business genuinely should not build. Sometimes the correct answer is that this keyword does not deserve a dedicated page, and the component stays where it is.

What this metric should never be used for is performance reporting to a client or a stakeholder. It is an internal diagnostic. Putting it on a monthly report creates pressure to move a number that does not pay anybody, which leads directly to the failure in the next section.

The metric that lies

Average Quality Score across an account is one of the most seductive vanity metrics in paid search, because it moves, it is easy to chart, and it feels like a measure of craft. It is also trivially gameable in a way that destroys accounts, and the mechanism is worth spelling out because it happens in slow motion and nobody notices until the quarter closes.

It goes like this. Someone puts average Quality Score on the reporting deck. The obvious way to raise an average is to remove the low values, so the low-scoring keywords get paused. The average rises, the chart looks like progress, and the report writes itself. Meanwhile the keywords that were paused were disproportionately the specific, high-intent, competitive-set-disadvantaged ones — which is to say, the ones producing revenue. Volume drops, cost per acquisition rises, and the two charts sit on the same page pointing in opposite directions without anybody connecting them.

The second number that catches it: cost per acquisition and conversion volume for the same segment, before and after every pause. If a pause raised the average score and raised cost per acquisition, the pause was a mistake regardless of what the score chart says. Quality Score and profitability are different axes, and the account only pays one of them.

The related failure is quieter. An account can hold a healthy composite everywhere while spending most of its budget on broad, cheap, high-scoring traffic that never buys. High scores on the wrong keywords is a very comfortable place to lose money.

What automation changes, and what it does not

Smart Bidding removes the bid from the list of things a person tunes. That does not make relevance work less important — it makes it more important, because it is one of the few inputs still genuinely under the advertiser's control.

The mechanism has not changed. Quality signals are priced into the auction whether a human or a system set the bid. If a landing page is slow and an ad group is incoherent, the automated bid buys less than the same money would have bought against a coherent account. Automation optimizes within the conditions you give it; it does not repair the conditions.

What does change is where the advertiser's attention should go. Under manual bidding, a lot of time went into bid adjustment. Under automation, that time is better spent on the inputs the system cannot supply for itself: which conversion action is being fed to it and whether it corresponds to real value, how the account is structured so conversion data is concentrated rather than scattered, what the creative says, and what the page does after the click. That pattern is not confined to bidding: across most of the marketing stack the useful AI turns out to be the platform models already running underneath the tools, with the human contribution moving to the inputs — an argument set out at length in the assessment of where AI earns its place across SEO, content, ads and social.

There is also a structural tension worth naming. The ad group structure that produces the best ad relevance labels — many small, tightly themed groups — is often the structure that most starves an automated bidding system of the data it needs. When those two pull against each other, the bidding system wins, because it is spending the money. Accept an Average ad relevance label in exchange for a bidding model that has enough volume to work.

None of this removes the need for a written rule about when changes get made and on what evidence. The bid governance framework covers that discipline for bids and budgets, and the same logic should govern copy and structure changes — without a threshold agreed in advance, every component label becomes an invitation to react to noise.

The myths, which live in a separate post

There are four persistent misconceptions about this metric — about bids, about cost, about broad match and about when to pause — and rather than restate them here, they are handled properly in the companion post on Quality Score myths. Read that one if what you are looking for is what is untrue; this one is what is true and what to do about it.

Where to start

Open the account, sort by cost, add the three component columns, hide the composite, and look at the top twenty keywords by spend. In most accounts the pattern is visible within a few minutes and it is the same pattern: landing page experience weak across the board because nobody owns the site, ad relevance weak in two or three ad groups that were built by pasting a keyword list, and expected click-through rate broadly fine. That is a work order, and it is a far more useful output than a number between one and ten.

If the account has never had a structured review, the component columns are the wrong place to start — a full account audit will surface larger problems than these three labels can, and quality diagnostics are best read after the structural issues have been found. The framework this sits inside is documented on the Authority page, and I run this work through Arcetis, the growth systems practice I founded. The principle underneath all of it is the one this post opened with: the score is a readout, and the only thing worth optimizing is what it is reading.

Related reading