← all writing

· 6 min read

How do ad auctions actually set prices?

What determines ad auction pricing in 2026: first- and second-price formats, quality scores, floors, pacing and fees, with a public auction to compare.

Almost every ad you see was priced by an auction that finished before the page loaded — and almost nobody buying those ads can explain how the price came out. Ad auction pricing is arithmetic plus incentives, some of it shown to you and most of it not. Here's the machinery, from the hidden version to the fully visible one. (Disclosure: published by BidSurvivor, which runs the visible one, so weigh the last section accordingly.)

The big platforms: auctions with adjustments

Google, Meta and the programmatic exchanges run auctions where your money bid is only one input. Google multiplies bids by quality scores (expected clickthrough, relevance, landing page); Meta weights by predicted engagement; exchanges add floors, fees and header-bidding layers (the 100-millisecond version). The consequence buyers feel: two advertisers can pay wildly different prices for the same slot, and neither can fully reconstruct why. The price is real; the pricing is opaque.

The formats underneath

Most digital auctions historically ran second-price (winner pays just above the runner-up) and have largely moved to first-price (winner pays their bid) — the strategic difference matters enough that we gave it its own article. Google Ad Manager made the switch for its exchange in 2019, and said why: with several auctions stacked on top of each other, a second-price rule in one layer was no longer telling buyers what they would actually pay. Layered on top of either format: floors (minimum acceptable bids), reserve prices, and budget pacing that quietly spreads your bids across the day.

A worked example: one slot, three prices

Take a search auction with three bidders and a quality multiplier, the simplified form of the rank Google describes in its own help pages. Bidder A bids $4.00 with a quality score of 5. Bidder B bids $3.00 with a score of 8. Bidder C bids $2.00 with a score of 10.

Rank is bid multiplied by quality: A scores 20, B scores 24, C scores 20. B wins with the lowest cash bid but one. What B pays, under a second-price-with-quality rule, is the smallest amount that would still beat the runner-up: the runner-up's rank divided by B's own quality, so 20 ÷ 8 = $2.50, plus a cent. B bid $3.00 and paid $2.50 for a click that A, bidding more, did not get.

Now the part that decides the price and that no invoice shows: B does not know A's bid, A's score, or that a score of 8 rather than 7 was the difference between paying $2.50 and paying $2.86. Change any hidden input and the price moves; nothing the buyer can see explains why. Multiply that across the ~$5.42 average cost per click WordStream reports for 2026 and you have a market where the price is precise and the pricing is a black box.

Fees, floors and the missing middle

In programmatic display the price the advertiser pays and the price the publisher receives are separated by a supply chain that takes a cut at every hop. The best-known audit of that chain, the UK advertisers' body ISBA with PwC in 2020, could trace only about half of advertiser spend through to publishers and could not account for roughly 15% of it at all — the "unknown delta". Whatever the current figure, the structural point holds: the auction sets a price, then several parties between you and the page each apply a margin, and the sum of those margins is part of what you paid for the click.

Floors do something subtler. A publisher's floor of $2 CPM does not simply reject bids below $2; in a first-price world it becomes the price a lone bidder pays, and in a second-price world it becomes the runner-up when nobody else shows up. Floors are set by the seller, adjusted by software, and invisible to you. When your $3 bid clears at $2.99, the floor was $2.99.

The visible version, as a teaching model

Our board strips the machinery to parts you can watch: BidSurvivor sells two twelve-hour slots a day where price formation is the entire visible product. The floor is explicit ($1.00 to take a held slot, $0 for an empty one). The standing bid is public. The "auction" is one rule — beat the standing bid and the slot is yours instantly — with a hard ceiling ($5,000) and no quality score, no pacing, no adjustments. Every clearing price is published on the board tile, the Hall and the brand's page, so you can watch what a slot "is worth" get discovered by bidders rather than computed by a model. It's a minnow next to Google's machinery, and that's precisely what makes it legible.

The demo auction on the site plays the whole pricing sequence in a minute, and it's worth reading as a script. Northwind takes an empty slot for $0. Kestrel takes it at $1.00, the floor. Northwind comes back at $2.50. Tinderbox takes it at $5.00. Kestrel tries $4.00, misses, and is charged nothing — a failed bid releases its authorisation. Whoever holds the slot when the twelve hours end keeps it in the record. Four prices, all visible, and every one of them set by a person deciding the slot was worth more than the last person thought.

Two rules make the visible price honest rather than merely visible. Being outbid is not refunded — Northwind's $2.50 stays spent — which means each bid is a real statement about value, not a free option. And the $5,000 ceiling stops the one failure mode a public open-outcry auction has, the escalation spiral; why a ceiling makes the auction fairer works through that. The complete rules are in how BidSurvivor works.

What this means for a buyer

Three practical rules fall out of understanding the machinery. On adjusted auctions, improve the multipliers before the bid — quality scores discount good ads more than budgets boost bad ones; in the worked example above, B's score of 8 was worth more than A's extra dollar. Distrust any price you can't decompose; if a platform can't tell you why a click cost $7, the answer includes their margin and probably someone else's. And calibrate on transparent markets when you can — a published clearing price is the only kind you can sanity-check someone else's invoice against, and the unit you compare in matters as much as the number.

The less of the pricing you're shown, the more of the price is someone else's decision.

BidSurvivor sells advertising in twelve-hour blocks, at auction. The first brand into an empty slot pays nothing, every account starts with $100 of house credit, and every brand's click-throughs are public before you bid.

See the board

Read next