TL;DR

  • The month's biggest stories in marketing and advertising technology shared one habit: the figure that would settle the matter was sealed, redacted, modelled, or simply never published, from a monopoly ruling kept private for two weeks to a lawsuit with its dollar figures blacked out.
  • Institutions on every side did it. A federal judge withheld her reasoning, a retailer's pricing math sat under redaction, a browser shipped ad measures with no pass mark, and several firms announced acquisitions while declining to state a price.
  • The open questions now carry dates: October 2 for the Google judgment, September 21 for a TikTok hearing, an unknown date for the Amazon complaint to be unsealed, and an appeal window in Frankfurt.

Key Findings

  • A federal court rejected a forced sale of Google's ad exchange and imposed six years of conduct rules instead, but filed its 106-page explanation under seal for 14 days and left the hardest deadlines to later negotiation.
  • Three new measurement tools arrived without the number that would make them usable: an AI search report with impressions but no clicks, a browser ad-load metric with no threshold, and a YouTube audience estimate that only the creator can see.
  • The Federal Trade Commission's case against Amazon accuses it of a hidden surcharge worth a headline 20 billion dollars, yet the rates and dollar estimates that would size the harm are redacted from the public complaint.
  • Two courts pushed the other way, forcing disclosure out of platforms: a Los Angeles judge moved to keep TikTok under a child-privacy order, and a Frankfurt court stripped Meta of a liability shield and ordered it to reveal the reach and revenue of fake ads.
  • Product changes and acquisitions landed with the commercial terms left out, from Google removing free shopping listings across Europe to two ad tech acquisitions announced with no price attached.

Details

The connecting thread across September was not a single company or a single scandal. It was a pattern in how information moved, or failed to move. A marketing professional reading the trade press this month met the same shape again and again: an announcement, a headline number, and then a gap exactly where the most important figure should sit. What follows are five stories that share that shape, told in order of how far each institution went to keep the key fact quiet.

1. A court that explained itself in private

Start with the ruling that set the tone. On September 2, 2026, Judge Leonie M. Brinkema of the United States District Court for the Eastern District of Virginia entered a two-page order in Case 1:23-cv-108, the Department of Justice's advertising technology case against Google. The order did the headline work: it refused the government's request to force Google to sell AdX, its ad exchange, and it rejected two other structural fixes. An ad exchange is the marketplace where the right to show an ad on a web page is auctioned in a fraction of a second while the page loads. Google's exchange is the largest, holding between 63 and 71 percent of open-web display exchange transactions in the data before the court. Brinkema's April 2025 liability ruling had found that AdX charged a durable take rate of 20 percent on each open-web display transaction for over a decade, against rivals who often charged closer to 10 percent. A take rate is the cut the middleman keeps on every sale.

The reasoning behind the refusal ran to 106 pages, and the court filed it under seal for 14 days so the parties could ask for redactions. For two weeks the industry argued about a decision it was not allowed to read. The full opinion emerged around September 16 and 17, and PPC Land published the six-year worldwide conduct decree for Google's ad tech business on September 17, followed by a closer reading of the unsealed opinion ordering AdX and DFP opened to Prebid.

What the opinion says matters less to a general reader than what it does. Rather than change who owns Google's tools, it changes how they must behave for six years, everywhere Google's ad tech operates, backed by a court-appointed monitor and a three-person technical committee that Google will pay for. The core obligation is interoperability with Prebid. Prebid is free, shared auction software that most publishers already run to let many exchanges bid at the same time, a technique called header bidding. Google must connect both AdX and DFP, its ad server, to Prebid. An ad server is the program a website uses to decide, out of all the ads available, which one to actually show. Today Google's advertiser demand flows most easily into Google's own server. The remedy forces that demand to show up for rivals on the same terms. The plaintiffs' own expert calculated that ending one contested practice alone, uniform pricing rules, would cut AdX's 20 percent take rate to roughly 16.6 percent, which puts a measurable number on what publishers can watch over the coming quarters.

The reason the court gave for refusing a sale can be summarized without the sealed detail. The government never produced a buyer for AdX. Moving a system woven into more than 17 years of shared code and 100 million lines of it looked, on the engineering evidence, slower than simply changing the rules. And the court read the government's core argument as distrust rather than proof, writing that the case for a breakup rested on a lack of trust that Google would obey an order. Digiday's breakdown of who wins and loses, published September 18, quoted Adform chief technology officer Jochen Schlosser describing the outcome as "a six-year game of algorithmic hide-and-seek." Its companion piece, a decoded guide to what changes and what will take years from September 17, set out the timeline: full behavioral change is expected within 15 months.

Now the silences, because they are the point. DV360, Google's tool for large advertisers to buy ads across the web, was left untouched; the court found it played no role in the part of the case where liability was established. Connected television, the term for streaming on a TV set, along with in-app advertising and retail media, all sit outside the order, which reaches only open-web display. The government's demand that Google share bid data covering video and in-app formats was refused for the same reason. The proposal to publish Google's auction code was rejected because the argument that the code was an unauditable black box was never made at the liability stage. The non-discrimination language was called too indefinite and sent back for redrafting. And several mechanical questions were simply deferred: the deadline for AdX to feed bids into rival ad servers is unresolved, with the government asking for six months and Google for twelve; the format of the documentation Google must provide for each ad served is undecided; and no date exists yet for when AdX bids will actually begin arriving in Prebid. The monitor has not been chosen. A joint proposed final judgment is due by October 2, 2026, and only then will the industry learn the dates. On most of the contested mechanics, the judge sided with Google. The backdrop is a shrinking business: Alphabet's Google Network revenue, the segment that pays open-web publishers, declined slightly to 7.30 billion dollars in the second quarter of 2026, from 7.35 billion a year earlier, the only Google advertising line that fell while total ad revenue rose.

2. The measurement that arrives without its most important number

The same instinct, publish the count but withhold the part that converts it into money, ran through four measurement changes this month.

The first was a search report. On August 31, 2026, Google finished rolling out globally a Search Console report for its generative AI features, the AI Overviews and AI Mode answers that sit above the normal search results. The report shows impressions, meaning counts of how often a site's links appeared inside those AI answers. It does not show clicks, meaning counts of people who actually tapped through, and it does not show the search queries that triggered the appearance. PPC Land described the launch as AI search reports that arrive without clicks or queries. For a publisher watching traffic fall while impressions rise, that is the difference between a diagnosis and a shrug: the report says a page was seen inside an AI answer, but not whether anyone arrived, nor what they had asked. In the same window, on September 1, Google quietly unlinked a number it did not disclose of Local Service Ads accounts from the business profiles that feed them, during a migration that deletes historical reporting unless an advertiser exports it first.

The second was a trade-body deck, and it carries a conflict of interest that has to be stated plainly. On September 15, the Video Advertising Bureau, the trade body for television sellers, circulated a 13-page analysis setting Google's own enforcement figures against linear television. The comparison is that Google suspended 38 times more US ad accounts than TV had advertisers: 3.3 million United States advertiser accounts suspended in 2025 against 86,800 brands that ran television commercials, a ratio near 38 to one. Globally the numbers were 8.3 billion ads blocked or removed and 24.9 million accounts suspended. The Video Advertising Bureau exists to sell television against digital, so a comparison flattering to television is self-interested by construction, and its own footnotes undercut it: television time is sold through negotiated deals while Google takes self-serve signups, and one bad operator can open many accounts. MediaPost carried its summary of the argument on September 16. The firmer point survives the bias. Google says its systems stopped more than 99 percent of violating ads before they served, but discloses nothing about how long the remaining share stayed live or who saw them, a remainder that at one percent of 8.3 billion runs to as many as 83 million ads. PPC Land also found that the 16 listed enforcement categories total only about 5.27 billion ads, roughly 63 percent of the 8.3 billion headline, with no account of the rest.

The third was a browser metric. On September 15, Google's Chrome team added four experimental ad-load metrics to the Chrome User Experience Report, a public dataset built from real browsing sessions, in a post by group product manager Alex Cone. The four are ad count, ad density, and two weight measures for the data and the processing time that ads consume. For the first time, anyone can look up how ad-heavy a site feels to real users. The gaps are structural. The metrics carry no threshold, no good or poor band of the kind Chrome sets for its performance scores, so a density figure describes a site without grading it. Only sites that name an authorized seller in an ads.txt file, the text file listing who is allowed to sell a site's inventory, are measured, and the figures are reported at the 75th percentile of page views. The logic that decides which frame counts as an ad sits in documentation Chrome did not publish alongside the launch, and the post is silent on whether the new fields feed Google Search ranking. The party doing the measuring here also sells the ads and runs the dominant ad server, an overlap the documents do not mention. In a disclosure worth repeating, PPC Land noted that it sells display ads itself and is one of the sites the metric measures.

The fourth is the one most likely to be misread, so the correction has to be exact. On September 5, 2026, YouTube launched a co-viewing metric that only creators can see, called views co-viewed. Co-viewing means more than one person watching a single screen, the classic living-room case where three people watch one television that registers as one device. The new metric estimates how many people were on the sofa. It is modelled, not counted: no sensor watches the room, and the estimate is built from three described inputs, demographic patterns, video genre and viewing times, with no error band published and up to 48 hours of processing lag. Crucially, it appears only in the creator's private analytics and covers organic views of the video, which is why a brand handed that number in a sponsorship pitch has no way to check it beyond a screenshot. What this metric is not is the only co-viewing data media buyers have. Google has reported co-viewing on paid ads in Google Ads and DV360 since 2022, and on June 2, 2026 seven Google Ads reach metrics moved to a Total Co-view basis. A buyer purchasing YouTube inventory can pull a co-view figure from a report; PPC Land laid out how the platform now estimates three modelled TV viewers where analytics counted one device. The unauditable one is the organic creator metric alone.

3. The auction with the price blacked out

If the measurement stories show numbers arriving incomplete, the Amazon case shows a number withheld by court redaction.

On August 31, 2026, the Federal Trade Commission and 22 state attorneys general filed a 181-page complaint against Amazon in the United States District Court for the Western District of Washington, Case 2:26-cv-03097, styled United States v. Amazon.com, Inc. PPC Land reported the FTC and 22 states suing Amazon over hidden ad surcharges the same day. This is not an antitrust case. It is a deception case under Section 5(a) of the FTC Act and 22 state consumer-protection laws, and it turns on one sentence Amazon repeated for years: that the winner of an ad auction pays about one cent more than the next-highest bidder, the standard second-price rule. The complaint alleges that stopped being true, and that Amazon added an undisclosed markup its own staff called a "soft reserve" on top of the auction result, a step one internal document described, in the FTC's telling relayed by PPC Land, as "a surcharge hidden in it."

The mechanism matters because a normal reserve is a floor published before an auction runs. The complaint alleges Amazon calculated its reserve after the auction and applied it when it was higher than the second-price result, so that the final charge could exceed the price real competing bids would have set, while never passing the advertiser's own maximum bid. The one figure that survived redaction and travels well is the first-price rate, meaning the share of clicks where the advertiser paid its own full bid rather than anything resembling a runner-up. PPC Land's follow-up, that Amazon charged advertisers their full bid 79 percent of the time on September 1, tracked that rate from 4 percent in late 2020 to 79.1 percent in 2024. The timeline of the alleged practice runs from Sponsored Brands in 2018, to Sponsored Products in mid-2019, to Display Ads by 2023, covering about 1.2 million United States advertisers, more than 500,000 of them small and medium-sized businesses. FTC Chairman Andrew N. Ferguson said the customers were misled into paying significantly higher prices.

Here is the blacked-out center. The complaint's most consequential arithmetic, its average surcharge rates, the size of the guardrails Amazon placed on its reserves, and a January 2025 internal estimate of what optimal bid shading would have saved advertisers, is redacted from the public version. The 20 billion dollar headline arrives with the granular dollar figures that would size it hidden. Amazon rejects the framing and published a same-day rebuttal, arguing that advertisers saved more than 8 billion dollars between 2021 and 2025 because Amazon ranks ads by relevance rather than by bid alone, that average Sponsored Products cost-per-click was flat in inflation-adjusted terms from 2019 to 2024, and that in no scenario does an advertiser pay more than its bid. Those are the company's own estimates, not tested findings, and the case is an allegation, not a ruling. Two sides, two sets of numbers, and the arithmetic that would adjudicate between them sealed by the court.

4. The courts pushing the other way

Not every institution this month chose to say less. Two courts forced platforms to disclose more, which is the same theme viewed from the opposite side: the fight over what stays hidden.

In Los Angeles, Judge George H. Wu issued a 16-page tentative ruling on September 18, 2026 that would refuse the government's own unopposed request to end a 2019 consent order, the binding court-approved promise that settled a children's privacy case against Musical.ly, the app that became TikTok. PPC Land covered the tentative refusal to lift TikTok's child-privacy order. The structure is what makes the silence commercial. The Justice Department announced a 400 million dollar settlement on August 21, 2026, with 300 million payable immediately and the final 100 million due only upon entry of an order vacating the Musical.ly decree, which otherwise runs to 2029. Wu wrote that neither side had shown the order was no longer needed, and that the government's papers did not establish substantial compliance. The settlement terms themselves were never filed with the court; Wu noted the figure reached him through a press release rather than a filing. A hearing is set for September 21 at 8:30 in the morning. The case is 2:19-cv-01439.

In Germany, the Frankfurt Regional Court, sitting as its 6th Civil Chamber in case 2-06 O 234/25, ruled on September 16, 2026 that Meta must stop distributing scam ads that impersonate the financial-education brand Finanzfluss and its co-founder Thomas Kehl, and that it must disclose the reach those fake ads achieved and the revenue they earned, broken down by web address, on pain of a fine up to 250,000 euros per breach. PPC Land reported Meta losing its hosting shield in Frankfurt and set it within a wider weekend of consequential filings. The legal core is that Meta could no longer claim to be a neutral host under European law, because its algorithm and its advertising system decide which ads appear and in what order. That control, the court found, removes the ignorance defense. Finanzfluss reported close to 260 violations in August 2024 alone, and some flagged posts stayed live as long as 62 days on the plaintiffs' account, or up to 20 days on the court's findings. The judgment is not final, and Meta can appeal.

A third filing belongs here as the disclosure that has not yet happened. A proposed antitrust class action against LinkedIn, the Microsoft-owned professional network, was filed on September 14, 2026 in the Northern District of California and made public on September 17. PPC Land described a LinkedIn antitrust class action over Sales Navigator. Sales Navigator is LinkedIn's paid tool for salespeople. The grievance is that on February 1, 2025 LinkedIn moved a feature, the ability to view LinkedIn profiles inside a customer's own business software, from its Advanced tier up to Advanced Plus, which carries a minimum commitment of ten licenses. A three-person team that wanted to keep the feature therefore had to buy ten seats. Core runs 119.99 dollars a month per license and Advanced 159.99; Advanced Plus carries no public price at all and requires a custom quote. The lead plaintiff is a trade association, Fairlinked e.V., and the identity of the litigation funder behind it is not disclosed. LinkedIn has yet to file a response.

5. The changes and the deals announced without the numbers

The last cluster is the quietest, and the most everyday: platform changes that shipped without usage data, and business deals announced without a price.

The largest platform change reached shoppers across Europe. Google removed free product listings and the popular-products carousels from search results across the European Economic Area, the unpaid blocks that let a retailer appear in shopping results without bidding. Search Engine Roundtable reported Google dropping free product listings across the EEA on September 18, and PPC Land documented the two-day disappearance. Independent trackers recorded declines close to total, roughly 90 to 100 percent in Germany, France, Belgium, Sweden and the Netherlands, with the space handed to comparison shopping services. Google's Ginny Marvin confirmed the change followed from the European Union's Digital Markets Act. One tracker, Adriaan Dekker, wrote that "Free listings from Google Shopping look to be disappearing in the EEA." Google did not quantify the traffic effect on merchants or say whether an organic format would return. A second change, reported by Search Engine Roundtable on Google Ads removing manual language targeting and confirmed by Ginny Marvin to MediaPost on September 11, removes the setting that let advertisers pick which language a Search campaign targets, handing the choice to automated matching. Google published a month, September 2026, not a day, and no data on how many advertisers used the setting it retired.

The deals told the same story from the commercial side. On September 18, AdExchanger reported Infillion acquiring Foursquare, folding a location-data company covering more than 100 million points of interest and 250 million United States devices into an ad tech group that has bought MediaMath and Catalina before it. The price? "Terms of the deal were not disclosed." Three days earlier, PPC Land reported Mediaocean's venture arm offering AI startups access to a 200 billion dollar spend system on September 15, an investment program backed by CVC Capital Partners, TA Associates and Eterna Growth Partners with no fund size attached to the money on offer. And Digiday examined the complex anatomy of agency AI bills on September 17, in which all three agency executives interviewed declined to give specific AI spend figures. The one hard number in the piece, that about 31 percent of companies spend more than 10,000 dollars a month on AI, came from Ramp, a corporate spend-management vendor with a commercial interest in the size of that bill. That conflict is worth flagging as squarely as the trade-body one earlier: the only figure on offer came from a supplier.

Recommendations

For readers who track this industry, the practical response is to treat each of these as an open file with a date attached, and to watch which way the disclosure moves.

  • Watch October 2, 2026 first. The joint proposed final judgment in the Google ad tech case is due that day, and any provision still disputed must be filed in both sides' versions. The benchmark that changes the story is the appearance of firm deadlines for AdX bids in Prebid and for AdX feeding rival ad servers. If those dates arrive, the remedy has teeth; if they slip to the monitor, expect years, not months.
  • Treat the Amazon complaint as incomplete until it is unsealed. The 79 percent first-price rate is the number to hold; the surcharge rates and the January 2025 savings estimate are the ones to wait for. Until they are public, the 20 billion dollar figure and Amazon's 8 billion dollar rebuttal are two claims that cannot be reconciled from outside.
  • Use each new metric for what it can support and no more. Impressions in the AI search report describe visibility, not traffic. A Chrome ad-load figure describes a site, not a grade, until thresholds exist. The YouTube views co-viewed number belongs in a creator's own reading of TV reach, not in a brand's audited buy, where paid co-view data already exists.
  • On deals with no price, note the pattern rather than any single transaction, and press for the terms that were withheld before drawing conclusions about value.

What to watch next, in plain dates:

  • September 21, 2026: the TikTok hearing in Los Angeles, which tests whether the child-privacy order stays and whether the final 100 million dollars ever moves.
  • October 2, 2026: the joint proposed final judgment in the Google ad tech case, the day the deferred deadlines become real or stay open.
  • Date not yet set: the unsealing of the Amazon complaint's redacted arithmetic, the surcharge rates and dollar estimates that would size the alleged harm.
  • Appeal window open now: the Frankfurt disclosure order against Meta, which is not final and can be challenged at the Frankfurt Higher Regional Court.

Caveats

Several of these stories rest on documents that are, by design, one-sided or incomplete. The Amazon complaint is an allegation written by the agency bringing the case; no liability has been found, Amazon rejects it, and its internal quotations were selected by the FTC. The Frankfurt judgment and the TikTok ruling are both non-final: the German decision can be appealed, and Wu's ruling is tentative and could change at the September 21 hearing. The Video Advertising Bureau analysis is commissioned by a trade body that sells against digital advertising, and the one hard figure in the Digiday agency piece comes from a vendor, Ramp; both should be read with those interests in view. The Brinkema opinion's most specific commercial figures may still carry redactions even in the unsealed version. Where numbers here are described as modelled, estimated or claimed, that is because the source presented them that way rather than as audited fact, and several, including Wurl's streaming-reach share and Amazon's savings estimate, arrived without a published methodology. Finally, the reporting window closes on September 20, 2026; the dates above fall after it.

Also noted

  • September 15: Riddle reported that 78 percent of interactive-content shares escape web analytics, across 3.8 billion views with a 112-second average engagement, a case that owned formats may offset lost search traffic, though the vendor sells the formats in question. PPC Land
  • September 15: Wurl said a single connection reaches more than 140 billion monthly connected-TV impressions across more than 70 percent of the ad-supported streaming ecosystem, a share it published without a supporting methodology. PPC Land
  • September 15: Clear Channel Outdoor won back the advertising concession at Pensacola International Airport, a role it last held until 2012, without disclosing the contract length or the size of its promised investment. PPC Land
  • September 17: Google Discover on desktop, tested and briefly rolled out over the past two years, has quietly failed to appear on Google.com for several months, with no announcement either way. Search Engine Roundtable