In the arrangement described by the court, Google's ad exchange could see the highest bid from competing exchanges and then adjust its own offer. A publisher might receive more than that outside bid without receiving as much as a genuinely even-handed contest could have produced. The advertisement eventually displayed on the page did not reveal the difference.^[1]

This was Last Look, examined in the district court's 17 April 2025 opinion in the Google ad tech case. It is a historical arrangement in a specific record, not a description of every auction running today. Its importance is easier to understand once we separate the jobs hidden behind the expression online advertising.^[1]

A publisher has advertising space. Advertisers want opportunities to reach readers. Exchanges organize bids, while a publisher ad server helps compare and deliver advertising opportunities on the website's side. Google's DFP was a publisher ad server; AdX was an exchange. Advertisers' buying tools occupied another position. One corporate name covered several different jobs.^[1]

The connections among those jobs matter as much as the names. Who can deliver a bid? Who learns what another source offered? Who controls the final comparison? Thinking only of a large company facing smaller companies can miss the particular design choices through which size becomes consequential.

Useful work, disputed conditions

Advertising can support the work of publishing. Producing material, maintaining a site and making it available all require resources; an advertiser needs a way to find an audience. Bringing those parties together is a service, not merely a convenient explanation invented after litigation began. Its usefulness should remain visible when its conditions are challenged.

Integration has an intelligible appeal too. A publisher does not have to build every connection itself. Advertisers can find opportunities through familiar tools, and some of the work of comparison and delivery is handled within one system. Ease, reliability and revenue can be genuine reasons for choosing it, rather than evidence that the chooser has been deceived.

Google made that argument against the Justice Department's proposed breakup in September 2025. It said separation would make monetizing content harder and reaching customers more expensive, particularly for small businesses. Its own proposal emphasized interoperability, allowing third-party tools real-time access to advertiser bids. These were the company's arguments and proposed solution, not an independent assessment of their eventual effects.^[2]

The argument merits a serious hearing. But useful integration and the necessity of a particular auction advantage are different propositions. Bringing additional buyers is one contribution. Giving an affiliated exchange information its rivals lack is a further decision. The contribution cannot, by itself, justify every condition attached to obtaining it.

The district court did not treat all advertising as one market. It found unlawful monopolization of open-web display publisher ad servers and ad exchanges, and unlawful tying of DFP and AdX. The advertiser-ad-network claim failed because the plaintiffs had not established the relevant market they proposed. The finding was bounded, not a verdict against all Google advertising.^[1]

That distinction affects how we can tell the story. These were liability findings after trial, not merely allegations surviving an early motion. Yet they do not establish that every acquisition, every integrated function or every advertising product was unlawful. Nor is this the separate Google Search case, whose results cannot be imported to complete an argument here.

Getting another bid into the room

Publishers and other industry participants developed header bidding to obtain real-time offers from multiple exchanges outside Google's existing architecture. Prebid was an open-source collaborative effort in that development. The winning outside bid could then enter DFP as a floor against which AdX competed. An alternative comparison had become technically possible.^[1]

That was more than adding another supplier's name to a list. It moved part of the comparison to a place publishers could inspect. Instead of relying entirely on a closed system's account of the result, a publisher could arrange for competing sources to meet and examine what happened. Choice acquired a means of being exercised.

But the winning outside offer also revealed a threshold. Under Last Look, AdX could respond after competing exchanges had already submitted their bids.^[1] The significant detail is the order of operations. We need not invent an executive's private motive or a particular publisher's losses to identify something concrete requiring an explanation.

There is no contradiction in a publisher benefiting from AdX demand while still receiving less than a more even contest might yield. An additional buyer can be helpful. That does not establish that the buyer needs an opportunity to inspect the best competing offer before responding. The useful service and the informational advantage have to be evaluated separately.

The question also extends beyond one sale. If competing exchanges have difficulty winning transactions, their ability to develop scale, data and services may suffer. A present auction design can affect who will be capable of offering a credible alternative later. Establishing that effect requires market evidence, not simply an observation that one transaction was better than no transaction.

The liability opinion itself separated the benefits of introducing real-time advertiser demand from the challenged preferences within the arrangement.^[1] That is a more workable inquiry than deciding whether integration is good or bad in the abstract. It directs attention to what a system could retain while changing a particular condition.

An apparent option to switch off a feature needs the same scrutiny. If doing so also gives up access to important buyers, a publisher is choosing between packages, not deciding about one isolated setting. Both the value retained and the value lost belong in the account. A choice on a menu is not necessarily an economically usable route.

An open connection is a task, not a slogan

After liability came the harder question of remedy. In its 2 September 2026 opinion, the court rejected structural remedies including an AdX sale and chose modified behavioral measures. It judged those measures sufficient to restore competition. That was a judicial assessment of a remedial design, not a measurement of a market years after implementation.^[3]

Google had argued that interoperability could address the problem without disrupting useful integration. The plaintiffs argued that conduct rules would not adequately prevent evasion and sought structural change as well. Their disagreement mattered even though many proposed behavioral provisions overlapped. Selecting a remedy does not make its risks cease to exist.^[2][3]

The opinion required connections between AdX, DFP and Prebid, access to AdX real-time bids through rival publisher ad servers, and ways for publishers to obtain and export their own data. The Justice Department's 16 September explanation likewise emphasized interoperability, data portability and nondiscriminatory bidding. These were specified obligations and objectives, not evidence that every integration already worked.^[3][4]

Translated from the technical language, the questions are recognizable. After replacing a tool, can the publisher still receive important bids? Can it take its records and configuration with it? Can another provider understand and use what arrives? Permission to choose a rival is of limited assistance if the business cannot make the journey.

Building that route requires work. A connection may exist but be too slow, unreliable, costly or incomplete to support comparable service. Checking that an interface has been announced is therefore not equivalent to checking that another provider can compete through it. The quality of the connection belongs inside the remedy, not outside as an incidental engineering detail.

The same is true of information. Exporting a file is only the beginning if the recipient cannot interpret it or integrate it into an operating system. Documentation, technical support and usable formats are part of making portability real. Otherwise the publisher may formally possess its records while remaining dependent on the original provider to make sense of them.

This is where oversight has to look beyond promises. What can an outside tool actually do? What happens when it fails to obtain information? Can a publisher produce enough of a record to show the discrepancy and receive an answer? Identically worded commitments can conceal markedly different conditions in everyday use.

The remedies opinion provided for a monitor and technical committee and selected a six-year judgment term; the department's account also described that oversight.^[3][4] A monitor is not an infallible substitute operator. It needs relevant competence and a route from identifying a problem to requiring a response. Its own reach, evidence and performance should remain examinable too.

The dates must stay distinct from the outcome. The opinion set a remedial path, with further work on the judgment's text and implementation. Publication did not mean the required interfaces were built that day. An obligation to provide access is not a measured increase in publisher revenue. This reading stops at the public opinions and September explanation, rather than issuing a premature implementation report.

The disagreement over structural and behavioral remedies remains worth understanding on its own terms. A forced separation can create transition difficulties; an ongoing conduct regime requires capable, durable supervision. Neither concern is answered by announcing that competition is desirable. A remedy must explain how its particular design addresses the established harm and what will make its commitments effective.

The wider lesson is not restricted to advertising. Many services that appear replaceable depend on connections people rarely see: whether records can travel, whether trading partners can remain reachable, and whether a new tool can obtain the information needed to function. Looking for those connections often gets closer to the point where a nominal option becomes a practical dependency.

This does not require every publisher to become an auction engineer. A useful arrangement should allow people to rely on professional work without making their inability to rebuild the machinery a reason to accept every condition it imposes. Verifiable connections and external review are ways of constructing choice, not decorations added after choice already exists.

The reader of a webpage still sees an advertisement. Behind it lie decisions about whose offers arrived, who could inspect them and who performed the comparison. The case invites us to look at those decisions without declaring all integration suspect: not only whether another route has a name, but whether anyone has done the work needed to connect it.