The highest bid loses more often than people expect. That’s the part most explanations leave out, and it’s the part that changes how you run an account.

Someone types a query. Before the page has finished drawing, Google has worked out which ads are eligible, scored them, decided how many to show and in what order, and calculated what each advertiser will pay if the ad is clicked. That happens on every search, which means there is no fixed price for a keyword and never was.

We wrote a while back about how ad delivery works across platforms generally. This is the search auction specifically, because unlike most of the machinery in digital advertising, it has rules you can do something about.

What eligible means

Scoring only happens to ads that are allowed in the room.

Your keyword has to match the query under whatever match type you chose. The campaign has to be running, in the right location, at the right hour, on the right device, with budget left. The ad has to pass policy. A campaign that spends its daily budget by lunchtime simply isn’t in the afternoon’s auctions, which is a mundane reason for a keyword you thought you owned to be invisible when your customers search.

Ad Rank, in plain English

Ad Rank decides the order, and whether ads appear at all.

Your bid is one input. Alongside it sit the quality signals: how likely your ad is to be clicked, how well the ad matches the query, and what happens on the page you send people to. Then the context of that specific search, which includes device, location, time, and the exact wording of the query. Then the expected effect of your sitelinks and other assets. Finally the thresholds, which are the minimum standards a position demands before an ad can occupy it.

Two consequences are worth carrying around. A well-matched ad on a relevant page can outrank a bigger bid. And if your Ad Rank falls below the threshold, you don’t appear at all, no matter what you’re prepared to pay.

You don’t pay your bid

Your bid is the ceiling, not the price. What you pay is roughly the least it takes to clear the Ad Rank of the advertiser sitting below you, given your own quality.

Which produces the only genuine discount available in a Google Ads account: better quality means clearing the same bar for less money. Two advertisers can hold the same position on the same query and pay materially different amounts, and the difference isn’t negotiated with anybody, it’s earned by relevance.

Quality Score is a diagnostic, not a target

The 1 to 10 number in your keyword columns isn’t what the auction uses. It’s a historical summary, aggregated across past auctions, of the same three components: expected click-through rate, ad relevance and landing page experience.

Use it the way a mechanic uses a warning light. Look at which of the three reads “below average” and go and fix that. Chasing 10s across the account is a hobby. Nobody has ever taken a Quality Score to a board meeting and had it mean anything.

What the auction is really measuring

Whether your ad and your page answer the question that was asked.

The chain runs query, keyword, ad, landing page. Break it anywhere and quality slips, cost rises and position falls. Most accounts break it in the same place: broad keywords pointing at ads that describe the company, pointing at a homepage that makes the visitor start their search again.

Tightening that chain is dull work with better returns than almost anything else in the account. If somebody searched for a specific product, the page they land on should be about that product, and the ad should have said so.

The bidding moved under everyone’s feet

The mechanics above are older than most accounts. What changed is who sets the bid.

Smart bidding sets a bid for each individual auction using signals you can’t see and can’t override: device, location, time of day, browser, remarketing lists, the exact query wording, and a pile of other things Google doesn’t publish. You’re no longer choosing a price. You’re choosing an objective and supplying the data it learns from.

So the highest-value work moved. It’s no longer bid management, it’s what you feed the model. Conversion actions that represent real business outcomes rather than every form submission including the spam. Values on those conversions, so the system optimises toward money instead of counts. Enough volume for it to learn from. Change the attribution setting and you change what the bidder sees, which is why attribution models stopped being a reporting question and became a buying one.

Feed it badly and it will efficiently buy you the wrong thing, at scale, without complaint.

Reading the auction from inside your account

Two columns tell you most of what you need. Search impression share lost to rank means you weren’t competitive enough on quality or bid. Lost to budget means you could have won and chose not to fund it. They call for opposite responses, and plenty of accounts have been bid up when the real problem was the budget cap, or funded harder when the ads simply weren’t good enough.

Auction insights shows who you’re actually up against, which is often not the competitor set in your marketing plan. And the search terms report shows what you really matched, which on a broad match campaign is regularly a surprise worth an hour of anyone’s Monday.

Where accounts leak

Broad match with a thin negative list. One ad group holding forty loosely related keywords and a single ad, so nothing can be relevant to much. Everything landing on the homepage. Conversion tracking that counts newsletter signups and quote requests as the same event. Bidding strategies swapped every fortnight, so no model ever finishes learning.

Southern Phone had a version of several of these when we took over. The accounts were underperforming, attribution was unreliable enough that nobody could see where users were arriving from, and campaigns had been set up and then left alone while costs climbed. We rebuilt the strategy, the buying and the creative testing, and fixed the measurement underneath it. Account cost per acquisition came down by $22, and the business chose to reinvest 68% more spend the following year, which is the response that tells you the numbers were believed. The Southern Phone case study has the rest.

What to do on Monday

Pull the search terms report for the last 30 days and add negatives for everything that isn’t your business.

Take your five highest-spending keywords, read the ad each one serves, then open the landing page and ask whether a stranger would find their answer above the fold.

Check which conversion actions are set as primary, and whether you’d be happy paying for every one of them.

Then leave the bidding alone for a fortnight and let it learn. If that sequence sounds like more discipline than time allows, it’s the everyday work of our paid search team, and it’s where most of the recoverable money in an account is sitting.

How the Google Ads auction actually decides who wins