Hill To Die On

Full analysis

Two things run through this episode, and they are not the AI segment everyone will quote. Adam Heimlich and Gareth Glaser spent most of the hour on how the plumbing cheats you. First, probabilistic identity is being passed off as deterministic in the bid stream, which inflates what you pay. Second, Google has a quiet carve-out in the proposed antitrust remedies that could make Prebid's whole ad-server overhaul pointless.

The decision for an operator is whether to treat either as actionable now. Both are easy to act on cheaply (audit your match rates, submit comment to the remedy process) and hard to undo if you ignore them (you keep overpaying, or the remedy ships with the hole intact). What sets the deadline is the DOJ remedy timeline on the Google side. The identity problem has no deadline, which is exactly why it festers.

The Skeptic

Gareth Glaser calls probabilistic IDs injected into a deterministic field "fraud." Strong word, thin proof. For an outsider: a deterministic ID is a confirmed one-to-one match; a probabilistic one is a statistical guess. Passing the guess as a match is the accusation. But where is the systematic evidence? The academic EID prevalence report ranks how often IDs appear, not how often they are wrong. Prevalence is not fraud. The Trade Desk catching false UID2s via OpenPath is one buyer, one ID type, and they have said almost nothing publicly about scale. On the remedy loophole, Glaser is a vendor who sells into this plumbing. His "hill to die on" is also his business.

The Operator

You cannot verify most of this on your own inventory, and that is the actual problem. A DSP buyer sees an EID in the request and has no resolution mechanism to check whether it is a real match or a guess about a heavy shopper. So the manipulation survives because nobody downstream can price it. The one concrete lever this episode hands a publisher is the COPPA line item story: Glaser's team found publishers who unchecked the Google Ad Manager COPPA box saw 5 to 10% immediate revenue lifts. That is a Tuesday-morning action. Go look at your GAM settings. The identity audit is harder, because asking your SSP to split deterministic from probabilistic match rates is asking it to shrink its own addressable number.

The Market Analyst

The Prebid loophole is the part that moves money. Prebid proposes flipping the hierarchy so Prebid Server runs the final auction and calls Google Ad Manager as just another bidder. Sounds like it breaks the tie between GAM and AdX. Glaser's warning is that it doesn't. When GAM loads directly on the page, it collects click-propensity signals and feeds them only to AdX, so AdX bids higher and wins. Route GAM server-to-server through Prebid and that data pipe breaks, so AdX bids less. Google doesn't have to block the new architecture. It just lets AdX pay less whenever you use it, and publishers switch back on their own because that is where the demand is. The remedy ships, the headline says monopoly broken, and the clearing price tells a different story.

The CFO

Price what each thing is worth. The COPPA audit costs an afternoon and returns a documented 5 to 10% on affected inventory. Do it. The identity cleanup costs you reach: if your segments lean on probabilistic matches, splitting them out shrinks the audience you can bill against, and the whole buy side would have to move at once for it to stop being a competitive disadvantage. Nobody moves first voluntarily. That leaves the remedy comment process, which costs a lawyer's time and pays only if regulators actually force Google to either share those page signals across all supply paths or disclose what AdX uses. Cheap option, long odds, big payoff. Classic thing you fund and forget.

Where they part ways

Two real disagreements sit underneath the friendly agreement on the mic.

Is the identity problem fraud or just incentives? Glaser says fraud, a thing you could litigate. Heimlich frames it as misaligned incentives: DSPs chase the 20% of users who make 80% of purchases, sellers inject those users' IDs, and everyone in the chain benefits except the advertiser paying CPA for conversions that would have happened anyway. If it is incentives, no lawsuit fixes it and only buy-side pressure or disclosure rules do.

Does the Prebid overhaul matter? Both hosts praise it because Prebid ships executable code, not specs, so it has teeth. But Glaser's own loophole argument says the teeth bite air if the click-propensity carve-out stands. You cannot fully believe both that the inversion is a breakthrough and that one data pipe neuters it.

What it hinges on

The remedy question comes down to one fact: will the final DOJ remedy force Google to share or disclose the in-page signals GAM collects for AdX? If yes, the Prebid inversion has real force and publishers get a genuine second auction. If no, AdX keeps paying more on GAM-direct inventory, publishers follow the money back, and the structural tie holds under a new diagram. Everything Glaser is worried about lives in that one clause.

The council leans toward Glaser being right that the loophole matters and wrong that anyone closes it in time. Verify the cheap stuff now: your GAM COPPA setting, and whether your SSPs will even tell you their deterministic-versus-probabilistic split. The answer to that second question tells you more about your supply partners than any prevalence report.

Prediction: The final DOJ ad-tech remedy against Google, expected to be set by Judge Leonie Brinkema in 2026, will not require Google to share Google Ad Manager's in-page click-propensity and engagement signals equally across all demand paths, leaving AdX free to bid higher on GAM-direct inventory than on the same impression routed through Prebid Server.

Confidence: Medium. The carve-out is already in the proposed remedies, and data-sharing mandates are the hardest to draft and enforce.

Why: Glaser's warning rests on a carve-out already present in the proposed remedies: AdX is allowed to bid differently when it receives different information, and GAM loaded directly on a page can collect signals that a server-to-server Prebid call cannot. Remedies that force behavioral separation (don't tie GAM to AdX) are far easier to write and police than remedies that compel a platform to share proprietary signals or standardize what data every exchange receives, which courts are reluctant to micromanage. So the likely outcome is a remedy that breaks the formal tie on paper while leaving the signal asymmetry that makes AdX win in practice. The opposite outcome would require the court to mandate affirmative data parity across supply paths, a far more intrusive step than US antitrust courts usually take.

Revisit by 2027-06-30: We're right if the entered remedy contains no requirement that Google make GAM's in-page engagement and click-propensity signals available to competing exchanges on equal terms, or no requirement to disclose which signals AdX uses. We're wrong if the remedy forces equal signal access across all demand paths or compels Google to publish the AdX signal set.