Why the Generalized Second-Price Auction Isn't Truthful (and What It Means for Your Google Ads Bidding Strategy)

What the generalized second-price auction actually is

When you run a search ad, you are not paying a posted price. You are entering an auction, and the specific auction almost every keyword marketplace uses is the generalized second-price auction, or GSP. Edelman, Ostrovsky, and Schwarz named and formalized it in their 2007 American Economic Review paper, whose subtitle, "Selling Billions of Dollars Worth of Keywords," is not hyperbole: it is the microstructure through which most of paid search is transacted.

The rule is deceptively simple. Advertisers submit a bid representing what they will pay per click. The slots on the page are ranked, the highest bidder takes the top position, the next bidder the second, and so on. Crucially, a winner does not pay its own bid. It pays roughly the amount needed to hold its position, which is close to the bid of the advertiser ranked just below it. The top advertiser, in effect, pays near the second-highest bid. That single design choice, paying the next bid rather than your own, is where all of the strategic complexity begins.

The Vickrey resemblance, and where it breaks

Paying the second price sounds like a famous and reassuring mechanism: the Vickrey auction, or sealed-bid second-price auction. In a Vickrey auction for a single item, honesty is optimal, not merely permitted. Bidding your true value is a dominant strategy, meaning it is your best move regardless of what anyone else does, because your bid determines whether you win but not what you pay. This property, called incentive compatibility, is why economists prize the design.

GSP borrows the second-price idea but breaks the property that made it truthful. With a single ad slot, GSP and the Vickrey auction coincide, and truth-telling survives. The moment there are two or more slots, they diverge. With multiple positions, what you pay for a higher slot depends on the full ladder of competing bids in a way that can make it profitable to bid less than your true value in order to win a cheaper, lower slot, or to nudge a rival into paying more. The comforting Vickrey intuition quietly stops applying exactly where real search auctions live: many advertisers competing for many ranked positions at once.

No dominant strategy: the result that changes how you bid

This is the load-bearing finding, and it is worth stating precisely. Edelman, Ostrovsky, and Schwarz showed that GSP generally has no equilibrium in dominant strategies, and that truthful bidding is not an equilibrium strategy. Herbert Varian, in his companion 2007 paper "Position Auctions," analyzed the same mechanism independently and characterized its stable outcomes. Both arrived at the same structural conclusion: there is no bid you can name that is best in isolation.

The practical translation is unforgiving. Because your optimal bid depends on the bids around you, and those change as competitors enter, exit, raise budgets, or shift their own strategy, the "right" bid is not a number to be discovered once. It is a position in a strategic equilibrium that moves. Both papers describe the stable configurations of that equilibrium, the set of mutually consistent bids from which no advertiser wants to deviate, rather than a single truthful price. An advertiser who treats a bid as a fixed declaration of value is not playing the game the mechanism actually presents.

The truthful auction the industry chose not to use

A natural question follows: if a truthful design exists, why is the marketplace not built on it? A truthful design does exist. It is the Vickrey-Clarke-Groves mechanism, developed across three foundational papers by William Vickrey (1961), Edward Clarke (1971), and Theodore Groves (1973). VCG generalizes the second-price idea correctly to many items by charging each bidder the cost it imposes on everyone else, the externality of its presence. Under VCG, bidding your true value is once again a dominant strategy.

The advertising industry adopted GSP anyway. The usual account is that GSP is simpler to explain and, under many conditions, produces higher expected revenue for the seller, while VCG pricing is harder for advertisers to reason about. The point for a bidder is not to relitigate that history. It is to notice its consequence: the auction you actually transact in was chosen for reasons other than making honesty optimal for you. The mechanism is not neutral, and it was never designed to be.

What "model your competitors" means for a real bid

Once truthful bidding is off the table, the work of bidding well becomes the work of modeling the field. Your correct bid is a response to a specific competitive structure: who else is in the auction, how much each of them values the same click, how their budgets pace through the day, and how the ranking above and below you is priced. None of that is visible in a single number you can enter and forget.

This is why "just bid higher" and "set a max and let it run" are not neutral defaults. They are strategic choices, usually poor ones, made by advertisers who have not recognized that the auction rewards responsiveness to rivals rather than declarations of intent. The competitors who model the field, revisit their position as it shifts, and treat the bid as a lever rather than a verdict are not being fussy. They are playing the game the mechanism actually is.

The second problem: your true value is not a number you already know

There is a deeper reason the "declare your true willingness to pay" instinct fails. It assumes you know that value. In practice, the true incremental worth of a click is an empirical unknown, and the numbers the platform hands you to estimate it are biased upward by the way advertising is measured.

What the evidence actually settles

It is worth separating what is settled from what is situational. The game-theoretic core is established and peer-reviewed: GSP is not truthful, it has no dominant strategy, and VCG is the incentive-compatible alternative that the industry declined. The measurement findings are established too, though the eBay result comes from one firm and should be read as a mechanism rather than a fixed percentage that applies to your account.

What is genuinely situational is the next step: the specific bid a given competitive field calls for on a given day, and whether a platform's automated bidding is optimizing for your objective or its own. Automated bidding is the platform modeling the auction on your behalf, which can be useful, but its target is not guaranteed to be your incremental profit. Those questions cannot be answered from a paper. They can only be answered from live auction data, watched and modeled over time.

What this means for how a paid budget should be run

Put the pieces together and the operating implication is hard to avoid. The auction has no honest optimal bid, your correct bid depends on competitors who keep moving, and the value you are bidding against is an empirical unknown that platform metrics tend to overstate. None of those facts is a marketing opinion. They are the documented properties of the system your money passes through.

A set-and-forget campaign treats a moving strategic problem as a fixed one. That is not a small inefficiency to be tuned away later; it is a category error about what kind of system paid search is. The conclusion the evidence supports is modest and specific: bidding well is continuous work, it is worth measuring against true incremental return rather than the flattering last-click number, and it rewards attention to the field over confidence in a single declared value.