1: \begin{proof}[Sketch] (A rigorous generalization of this proof can be found in Theorem~\ref{thm:vm-sp-char}.) Note that the GSP price $p_i$ is the minimum bid $i$ could have submitted while maintaining the same rank in the auction.

2: 

3: First, assume that bidders are simple value maximizers. Fix a bidder $i$ and bids $b_{-i}$. By the taxation principle, we can think about bidder $i$ choosing among the slots at prices $p_{i,j}$. Let $V_{i,j}\subseteq\Re$ denote the set of value reports for which $i$ wins slot $j$ under a particular auction.

4: 

5: For a pricing to be truthful for a value maximizer, it should be that $i$ wins a particular slot $j$ if and only if (a) it is willing to pay $p_{i,j}$ for slot $j$ and (b) is not willing to pay $p_{i,j-1}$ for the slot immediately above it. This implies that (a) $p_j\leq\inf V_{i,j}$ and (b) $p_{j-1}\geq\sup V_{i,j-1}$ and thus

6: \[\sup V_{i,j+1}\leq p_{i,j}\leq\inf V_{i,j}\]

7: that is, $p_{i,j}$ is the threshold value (bid) at which $i$ moves from slot $j+1$ to slot $j$. This is precisely the GSP price.

8: 

9: Lemma~\ref{lem:svm-roi} extends this to value maximizers with an ROI constraint.

10: \end{proof}

11: