Spokes.wiki Search About
Defined Term mechanism updated Wed Aug 12 2026 00:00:00 GMT+0000 (Coordinated Universal Time)

Search ad auction

How a search engine decides which paid results appear, in what order, and at what price. Every figure in this spoke’s paid corner — the CPC in b2b-ppc-metrics, the return in groas — is an output of this mechanism, and until now the spoke held no page on it.

The mechanism in one paragraph

Advertisers bid a maximum price per click on a keyword. The engine ranks the ads and charges the winner based on the bid below it, not its own, plus a minimum increment. Ranking is by bid multiplied by a quality assessment, so a cheaper bid can outrank a dearer one. Nothing is charged unless the ad is clicked.

Three generations

EraMechanismWhat it rewarded
pre-1997manual, cost-per-impression, batch contractsbudget and a sales relationship
1997 (Overture/GoTo)generalized first-price — pay your own bidreacting fastest; bidding robots
2002 (Google AdWords Select)generalized second-price (GSP) — pay the next bid + incrementbidding closer to value, without a cent-by-cent war

The first-price design had no stable resting point: any bid could be undercut by a cent, so the profit went to whoever could re-bid quickest, and the engine’s own revenue could collapse when one bidder automated and the others did not (gsp-auction-paper).

Second-price is not the Vickrey auction

The most-repeated claim about this mechanism is the one the economics disputes. GSP resembles Vickrey-Clarke-Groves, and Google’s early marketing invoked “Nobel Prize-winning economic theory,” but with more than one ad slot they differ: VCG charges each advertiser the externality it imposes on the others, GSP charges the next bid down. Truth-telling is not an equilibrium of GSP — bidding your true value is a mistake in the mechanism as built. The practical residue for an advertiser is that “bid what a click is worth to you” is folk advice, not a theorem.

Quality score, and what it became

In 2002 quality score was the estimated click-through rate, and the effective rank was bid × quality — a rescaling, formally equivalent to a plain auction on adjusted values. Today google-ad-rank names six inputs (bid, ad and landing-page quality, Ad Rank thresholds, auction competitiveness, search context, ad assets) and publishes no weights. Two changes matter:

  • A threshold can exclude an ad from the auction at any bid. Quality stopped being only a multiplier and became a gate.
  • Context is in the rank. Location, device and time of search enter directly, so a keyword has no single clearing price.

Where it connects

The paid and organic halves of this spoke rhyme here. A platform changes a mechanism to remove a gaming incentive; the optimization industry re-forms around whatever the new mechanism rewards; the platform stops publishing the formula. That arc runs 1997 → 2002 → today in the auction, and it is the same one seo-operating-model-shift, publishing-volume-hurts-seo and c-seo-bench describe on the organic and generative sides.

The asymmetry is worth stating plainly. The auction has a peer-reviewed analysis from outside the platform (gsp-auction-paper); generative-engine visibility, so far, mostly does not — which is why this spoke’s paid corner can describe a mechanism while its GEO corner is still arguing about whether its metric measures anything.

gsp-auction-paper · google-ad-rank · b2b-ppc-metrics · groas · seo-operating-model-shift · search-marketing · synthesis