Three days of filings, product launches, conference keynotes and quiet policy edits had one thing in common. In each case the question was not what the advertising worked on, but who owns the part of the machine that decides whether it runs at all, and who gets told when the answer changes.

A browser vendor decided a domain would not resolve. A retailer renamed its demand-side platform after the software that now operates it. One mobile ad network went to a San Francisco courtroom to stop a rival's measurement kit from reading its creative. A television measurement body set a deadline for its own dissolution. A league handed thirty-two clubs a generative content pipeline. And the job that is meant to hold all of this together, the chief marketing officer, is now lasting about three years.

None of it was negotiated. That is the through-line.

Game: Definitions settle more measurement fights than maths
Sixteen terms and four groups of four, drawn from four corners of the job. One group is far narrower than its tiles suggest, and it is not the obvious one.

The domain that stopped answering

Apple released iOS 27 on September 14, 2026. Within the week, The Trade Desk could no longer serve advertisements on Safari for users on affected Apple devices, and by late September Apple had begun investigating why.

Two domains are involved. The first is adsrvr.org, which carries The Trade Desk's ad requests and ad delivery and sets its third-party cookies. The second is UIDAPI.com, tied to the company's Unified ID 2.0 programme. Ian Meyer, senior director of engineering at The Trade Desk, raised the matter on GitHub and was explicit about the distinction that matters here: adsrvr.org is, in his words, "The Trade Desk's core ad request and delivery domain, not identity." He asked WebKit engineers for an urgent review, then asked again for an estimate of when findings might arrive.

John Wilander, who manages WebKit privacy and ad tech at Apple, acknowledged the flag and promised an investigation. His last update before Monday, September 29 was narrow, promising only to report back "if and when any changes are available" for testing. Neither Apple nor The Trade Desk commented publicly before AdExchanger published on September 29 at 15:43 Eastern.

The distinction Meyer drew is the whole argument. Safari has blocked third-party cookies outright for years, and the industry built alternative identifiers in response, of which Unified ID 2.0 is among the largest. A browser declining to accept an identity domain is a policy choice the market has absorbed. A browser declining to resolve the domain that physically delivers the creative is something else: the advertisement does not render, the impression does not happen, and no workaround at the identity layer helps.

Four further vendors joined Apple's blocklist in September 2026: ID5, Audigent, LiveRamp and Permutive. Each sits somewhere on the identity and data spectrum, which makes the pattern legible. What does not fit the pattern is that Google's delivery through ad.doubleclick.net kept working across the same period. Whether that reflects deliberate scoping, a difference in how the domains are classified, or an error in the list, nobody outside Cupertino has said.

There is prior art for both readings. WebKit has spent the past several release cycles tightening what scripts and storage can do, introducing new credential APIs in Safari 26 and rebuilding private browsing around link-tracking protection. Safari has also broken things by accident before, and advertisers have found their click identifiers quietly stripped without an announcement. The practical difference between an intentional block and a bug, from the buyer's seat, is close to nil for as long as it lasts.

The Trade Desk has spent 2026 arguing that the open internet needs fewer chokepoints, opening its ecosystem through OpenTTD and building direct supply paths. An operating system release is a chokepoint that no amount of supply-path work reaches. The company cannot appeal it, cannot bid around it, and cannot detect it in advance. It can file a GitHub issue and wait, which is what it did.

Game: Automated media buying still runs on manual vocabulary
Sixteen terms, four groups, and a grid that crosses search, audio and the money side. One of the four groups is far narrower than the tiles make it look.

Amazon renamed the buying desk after the software running it

Amazon held unBoxed in San Francisco on Tuesday, September 29, and used it to rename most of the advertising stack it has assembled since buying the Sizmek ad server in 2019.

The Amazon DSP is now DVA, for display, video and audio. The Amazon Ads console and the DSP are being folded together and relaunched as the Amazon Ads Agent, a conversational interface that Amazon says will span planning, campaign creation, optimisation, measurement and support. Kelly MacLean, vice president of science, engineering and product at Amazon Ads, described the result as "human-led AI supported" and said the company has "made massive progress in simplifying the overall experience."

The default campaign setup is now DVA+, which covers two AI-run campaign types: Performance Plus, tuned to sales and conversions, and Brand Plus, tuned to video branding and prospecting. Both existed before this week, and the four-step DVA+ builder was in US closed beta earlier in the year. What changed is the hierarchy. The manual controls that buyers have used for a decade, covering targeting, deals, supply sources, frequency and bidding, now sit behind a toggle labelled advanced settings. MacLean framed that toggle as offering "that traditional DSP-type experience [buyers] already know and trust."

A third product, Full-Funnel Campaigns, collapses sponsored ads, display, video and streaming television into one campaign. The advertiser supplies products, creative and a budget; Amazon's models handle execution and optimise toward outcomes including new-to-brand customers and long-term sales. Amit Bhattacharyya, vice president of full-funnel products, AI and modeling, put the reasoning plainly: advertisers were already stitching these campaigns together by hand, so "what if we made it really simple, and yet very powerful for advertising?"

Then come the ad formats, which are where the agentic language stops being a rebrand. Branded Conversations is a multi-turn unit inside Amazon's chatbot surface. Amjad Jbara, director of applied science for sponsored products, described it as creating "multi-turn shopping conversations to help shoppers find the right products informed by brand-provided inputs." In the unBoxed demonstration, a shopper searching for a coffee maker met a sponsored prompt reading "Help me decide" - the same phrase Amazon attached to a non-sponsored feature earlier - which opened an experience carrying the brand's visual identity and offering follow-ups such as "maximum automation," "works with capsules" and "fresh ground beans." Sponsored Prompts work more like Sponsored Products, surfacing items in response to a search or a recent purchase.

Advertisers steer the model through Conversational Briefs, structured inputs describing products and the relationships between them. The limits are specific and worth recording: brands cannot cherry-pick reviews, and cannot push particular reviews up the results. A separate cost-per-click product, Request Reviews for Sponsored Ads, targets items with fewer than 1,000 reviews and caps each campaign at 1,000 additional reviews collected, with no purchase incentives allowed.

That review cap sits oddly against Amazon's other position this week. Asked whether Sponsored Ads would show how an individual winning price is determined, the company declined to commit, pointing instead to its disagreement with the Federal Trade Commission and 22 state attorneys general, whose August 2026 complaint alleges secret surcharges in its ad auctions. Amazon has built a conversational layer that tells a shopper why a product suits them, on top of an auction that will not tell an advertiser why a bid cleared. The orchestration layer is getting more articulate in one direction only.

The commercial backdrop is not small. Amazon reported $19.8 billion in advertising services revenue in the second quarter of 2026 and $46.8 billion in third-party seller services, figures that appeared in a separate Amazon announcement on cross-border selling published September 29. The single global DSP account across 34 countries arrived earlier this year; the console merge completes a consolidation that has been running for most of 2026.

Game: The glossary was assembled by accident and never edited
Sixteen terms, four groups of four, and four working vocabularies with nothing in common. One group reads like leftovers until the connection lands.

AppLovin says Unity read its homework

On September 29 at 11am Pacific, a judge in the Superior Court of California in San Francisco heard AppLovin's application for a temporary restraining order against Unity. The case number is CPF26520212, and Digiday reported the filing the same afternoon.

The allegation concerns Unity's Ad Quality software development kit, a creative-review tool that runs inside mobile apps. AppLovin says the kit has "improperly collected data relating to advertisements served by AppLovin," and lists what it means: creative information, user and device data, impression revenue and engagement metrics. It goes further, alleging "on information and belief" that Unity fed this material into competing models and "used AppLovin's own output to train those models."

The relief sought is precise. AppLovin wants an order barring Unity from "collecting, intercepting, extracting, using, or disclosing" what the filing calls Protected Data, with collection switched off within five business days and the Ad Quality SDK modified within 30. Alongside the court application, AppLovin has opened JAMS arbitration with claims for breach of contract, trade-secret misappropriation and unfair competition under California law.

The two sides had already tried to settle this privately. AppLovin sent a cease-and-desist demand in August. Unity offered to wind down Ad Quality's operation with respect to ads mediated through MAX, and said it could remotely disable collection within five business days. AppLovin refused, citing data already collected and collection happening on other platforms.

Unity's public answer is that AppLovin runs an equivalent tool. Its Ad Review product, formerly SafeDK, is by Unity's account "far more intrusive" than the one under challenge. Unity will oppose the restraining order.

Both descriptions can be true at once, which is the uncomfortable part. Mediation SDKs sit between an app and every network bidding into it, and their legitimate job - checking that a creative is not malicious, does not autoplay audio, does not break the app - requires seeing the creative and the outcome. The industry built these tools on the assumption that the companies running them would not also be competing for the same impressions. That assumption has not held for some time. AppLovin faced its own data collection controversy in a previous cycle, and the measurement layer underneath both firms has spent 2026 trying to insulate itself, with Google, Meta, Moloco and Unity all investing in AppsFlyer explicitly to keep the mobile measurement partner neutral.

Neutrality bought with a cheque from four competitors is a fragile construction. This filing tests what happens when one of them decides the arrangement no longer holds.

Game: The glossary was assembled by accident and never edited
Sixteen terms, four groups of four, and four working vocabularies with nothing in common. One group reads like leftovers until the connection lands.

The currency nobody agrees to

Television measurement reached another impasse this week, and the body convened to fix it gave itself a countdown.

The Coalition for Innovative Media Measurement is forming a Strong Foundations Implementation Task Force, announced by managing director Jon Watts on September 23 and reported by Digiday on September 30. Its brief is to strengthen accreditation and produce what Watts called "a provider-neutral currency requirements framework." The notable clause is the sunset: if progress proves insufficient, the task force disbands within 12 to 18 months.

The conditions that produced it are a matter of record. VideoAmp and Nielsen both withdrew from the Media Rating Council accreditation process over the summer. Nielsen has now declined certification from the US Joint Industry Committee twice. Comscore, iSpot.tv and VideoAmp had their J.I.C. certifications renewed last week. So the market has a currency most of it still transacts on, which has stepped outside both accreditation regimes, and three alternatives that have stayed inside them.

Buyers described the practical cost. Mariel Estrada, head of video currency at Omnicom Media Group, pointed to the evaluation burden Nielsen's methodology changes impose, noting the two years spent comparing one dataset against another. Dan Riess, executive vice president and chief operating officer of US advertising at TelevisaUnivision, said Nielsen "can be challenging when they change methodologies." Nielsen altered seven US television currency metrics with effect from August 31, after ending its legacy ratings service earlier in the year, so the complaint has dates attached to it.

Alternatives are in live use rather than in pilot. Versant-owned CNBC transacts on Cogent, per Tom Winiarski, its executive vice president of ad sales strategy. Tim Hill, chief investment officer at Initiative, said the agency uses VideoAmp for audience buying at substantial volume. Disney's Jamie Power still treated Nielsen as the baseline in a September 22 session. Brittany Slattery, chief marketing officer of OpenAP and a founding member of the US J.I.C., named the blocker: a genuine multi-currency environment requires comparability that does not yet exist. Helen Katz, executive vice president of research at Publicis Media, framed the next phase as less about the complexity of any single system. Tooling has gone some way toward papering over the gap, with network groups now running two currencies inside datafuelX without rebuilding their workflows, but running two numbers is not the same as reconciling them.

The disagreement is not confined to linear television. MNTN and AppsFlyer announced a two-way integration on September 28 via Business Wire, and the numbers it produces will not match each other by design. MNTN sends AppsFlyer campaign IDs, creative IDs, impression IDs and IP addresses; AppsFlyer returns installs, in-app events, event values, revenue, device IDs, timestamps and geography. The integration went live on June 30, 2026, with setup documentation updated on August 5 and a technical paper published on August 24.

The mismatch is structural. MNTN applies a 30-day default conversion window with household-level, device-agnostic attribution. AppsFlyer applies a maximum 72-hour probabilistic view-through window that requires a same-IP-network match, on top of a standard 24-hour view-through lookback. Ten times the credit period on one side, a stricter matching rule on the other. The same campaign will show two conversion counts, permanently, and neither platform is wrong within its own definitions.

Alex Yip, director of product strategy at AppsFlyer, described the problem the integration addresses: "TV has historically been difficult for mobile marketers to connect to real business outcomes." The screen is large and loud, Yip added, and the campaign runs in hope that it surfaces somewhere in the app metrics afterwards. Moloco's install data suggests roughly two-thirds of app installs from CTV advertisements land within six hours, which arguably favours the tighter window. But MNTN's incrementality report, Site Impact report, Attribution Paths report and visit metrics do not run for app-only campaigns, so the control-group holdout that could adjudicate between the two windows is unavailable precisely where the dispute arises. IAB research from July 2026 found 43% of CTV buyers questioning where their advertisements actually ran.

MNTN reported 4,225 active Performance TV customers on a trailing twelve-month basis as of June 30, 2026, up 40% year on year, on revenue of $82.5 million in the second quarter, growth of 21%. Average revenue per customer fell about 14% to roughly $74,200, and around 95% of customers starting campaigns in 2025 were new to television. The integration is US-only.

A third measurement item landed in the same window. Cint opened self-service outcomes measurement inside its Lucid Measurement Study Creator, and Magnite became the first platform to offer it to clients running ClearLine or streaming campaigns. The product pairs brand lift with sales lift in one live dashboard across four purchase measures: sales lift, conversion rate lift, average order value lift and purchase frequency lift. Benchmarks launch for retail, telecom, technology and quick-service restaurants. Transaction data comes from Affinity Solutions, covering more than 150 million cards and $4 trillion in spend, and linear television measurement is in beta. Laura Manning, senior vice president of measurement at Cint, said the point was making the connection "easier to access and faster to act on."

What the announcement omitted is the familiar list: pricing, minimum thresholds, geographic availability, benchmark base sizes and control-group methodology. Magnite's streaming contribution reached $97.1 million in the second quarter of 2026, up 36% and 51% of total contribution excluding traffic acquisition costs, so the commercial stake in the numbers is real. Cint itself was taken private on August 24, 2026 at SEK 5.60 per share, valuing it at SEK 1,989 million.

Three measurement stories, three different answers to the same question, and no shared arbiter for any of them. CIMM's task force has been given a year and a half to find one.

Game: The client brief lands on four desks that never talk
Sixteen terms, four groups of four, and a grid where the obvious reading is usually the wrong one. The hardest group is four entirely ordinary words.

The grocery list becomes the interface

Kroger, Giant Eagle and DoorDash have each built assistants into their shopping apps, and what they have learned runs against how the technology is usually pitched.

Yael Cosset, Kroger's chief digital officer, described the object as narrowing a gap rather than replacing a workflow: "The shopping assistant is helping close the gap that exists between inspiration and basket shopping." Kroger's assistant handles a last-minute grocery run or a basket with several dietary restrictions, and accepts uploaded handwritten lists and recipes. Cosset's other observation was about behaviour: "Customers don't wake up and say, 'Oh, I'm going to shop online today.'" On pace, he was cautious - "speed with precision and direction is good."

Giant Eagle put its AI recipe generator directly into the main grocery app rather than standing it up separately, and consolidated ecommerce with marketing and merchandising last year. Shweta Prabhu, vice president of digital, ecommerce, martech, loyalty, data analytics and AI, reported that omnichannel shoppers carry 30% higher lifetime value than single-channel buyers. Heather Feather, vice president of digital, ecommerce and customer analytics, offered the line that explains why the app matters more than the assistant does: "No store receives as much weekly traffic as our website and mobile app do."

DoorDash found the same thing from the other direction. Co-founder Andy Fang said users ignore the standard chat window and reach instead for practical functions, of which "reorder my usuals" is the clearest example. The company is building tools for modifying orders in real time and finding substitutes for out-of-stock items. DoorDash leads US grocery orders and has been steadily opening its advertising inventory to third-party tools.

Kroger is also acquiring Giant Eagle, with closure expected in 2027, so two of these three assistants will eventually answer to the same commerce media operation.

Set against Amazon's Branded Conversations, the contrast is instructive. Amazon has built a conversational surface and sold sponsorship inside it. The grocers report that shoppers largely decline the conversation and want the list to work better. Both are building the same buyer agent plumbing, and both will end up selling closed-loop measurement against it, but only one has evidence about what shoppers actually do with a chat box. The retail media networks that read the DoorDash finding carefully may spend less on the chat window and more on the reorder button.

Game: The org chart changed, and the vocabulary stayed put
Sixteen terms, four groups of four, drawn from four corners of digital marketing. One group leans on a measurement discipline most people only half-recall.

Thirty-six months, and the title is disappearing too

While the platforms rearrange the plumbing, the executive expected to make sense of it is departing faster than at any point in sixteen years.

Median tenure for a sitting chief marketing officer now stands at 36 months, a 35% decline since 2010, on a Findem and CMO Huddles survey of 13,000 US marketing professionals at companies with more than 100 employees. The generational split inside that figure is sharper than the median suggests: CMOs who started in 2010 typically stayed four years, while those appointed from 2022 onward are lasting 2.6. Software marketing chiefs average 27 months.

Comparisons put the number in context. Chief information officers and supply chain chiefs both average 5.2 years. Chief executives average 7.4. Only chief financial officers, at 2.1 years, turn over faster, and the nearest cultural analogue is an English Premier League manager at fifteen months. Client relationships with their media agencies, at 3.7 years on the 4As and ANA tenure report, now outlast the marketers who signed them.

The title itself is thinning out. Forrester's analysis puts the share of Fortune 500 companies using the CMO designation at 36% in 2026, down from 55% a year earlier. Forty-six percent of CMOs report to somebody other than the chief executive. Jay Pattisall, the Forrester analyst, called the reporting structure a clear misalignment, and the budget data supports the reading: Gartner's 2026 CMO Spend Survey puts marketing at 7.8% of company sales, against 11.2% in 2018. Only 15.3% of CMOs are allocating budget to AI, and 30% consider themselves ready to scale it.

Mandates are part of the problem. IDC's 2025 work found 41% of chief executives wanting a new customer acquisition strategy while 30% of their marketing chiefs prioritised something else. Drew Neisser, chief executive of CMO Huddles, framed the pattern as companies handing marketing leaders the job of fixing growth without the levers to do it. Only a third of CMOs make it past five years in a role.

Adweek reached a similar diagnosis from a different direction. Jessica Apotheker, chief marketing officer at Boston Consulting Group, told an Adweek podcast published September 29 that "strategy only matters if it survives execution," and split the function into orchestration, art and science - connecting marketing to business planning, storytelling and creative, and analytics and CRM. Her argument on AI is that the efficiency framing is what gets marketing chiefs into trouble with finance, and that the conversation has to move from cost reduction to revenue. The relationships she treats as decisive are with the CFO, over a measurable growth equation, and with the CIO, over a technology-heavy agenda.

Agency leaders at Brandweek Atlanta were asked what they would change over the coming year, and their answers clustered on people rather than technology. Sharon Napier, founder of Partners + Napier, wants young women brought into the AI transition: "They are our future, but they can help us transform AI much faster than probably anyone in this room." Andrew O'Dell, co-founder and chief executive of Pereira O'Dell, prioritised upskilling, arguing the examination has to cover "our businesses and our people." Frances Webster, chief executive of Walrus, described a moment of major disruption carrying opportunity for anyone flexible enough. O'Dell also founded Silverside AI, which produced Coca-Cola's AI-generated holiday campaign and Svedka's Super Bowl spot, so the upskilling point comes from somebody who has already shipped the output.

Research from earlier in the year runs parallel. Acquia found only 41% of CMOs confident their team can govern AI, and a separate strand of the same work tied governance discipline to measurable search gains. A three-year median tenure against a technology transition measured in half-decades is an awkward arithmetic, and it is not obvious that either number moves first.

Game: The buy side and the sell side rarely agree on terms
Sixteen terms, four groups of four. This one leans on knowing which supplier does which job, and on reading four very ordinary words correctly.

A league, thirty-two clubs and a generative content pipeline

The National Hockey League and Adobe announced a partnership on September 28 at 8:00 a.m. Eastern, and the detail that matters for advertising is that the tooling reaches all 32 member clubs rather than the league office alone.

The stack named is Adobe CX Enterprise, with GenStudio for Performance Marketing, Experience Platform, Journey Optimizer, Express and Firefly underneath it. GenStudio reached general availability on October 14, 2024 and gained custom models and ad platform integrations on October 28, 2025, with routes into Amazon Ads, Google Marketing Platform, Innovid, LinkedIn and TikTok. Firefly's training corpus is Adobe Stock, openly licensed material, public domain works and NVIDIA libraries following the March 2026 partnership.

David Lehanski, the league's executive vice president of innovation and growth strategy, and Lara Balazs, Adobe's chief marketing officer and executive vice president of global marketing, both spoke to the announcement. The scale cited is 200 million-plus supporters across 147 countries, with media distribution reaching more than 250 countries and territories, and rosters drawing players from over 20 countries. That last figure is the commercial argument for personalisation at club level: a Columbus supporter in Helsinki and one in Ohio want different things from the same highlight.

What the announcement withheld is as informative as what it disclosed. No financial terms, no contract length, no exclusivity position, no launch dates for the fan-facing tools, and no data ownership framework. The consumer features are described with "could" rather than "will."

Timing places the deal inside a busy quarter for Adobe. The company reported record revenue of $6.76 billion for the third quarter of fiscal 2026, ended August 28, up 13% year on year, with total annualised recurring revenue of $27.50 billion and AI-first ARR growing more than 150%. Shantanu Narayen announced in March he would step down after 18 years once a successor is named, and on September 24 Photoshop and Firefly arrived inside Google Gemini on all plans. Adobe became presenting sponsor of MLB Opening Day for 2026, 2027 and 2028 in March.

One compliance date hangs over any generative sports content aimed at European audiences. Article 50 of the EU AI Act became binding on August 2, 2026, carrying a ceiling of 3% of global turnover for unmarked synthetic output. A league distributing AI-assisted fan content into 147 countries has a labelling question to answer that the announcement did not address.

The demand-side case is easier to evidence. Live sports advertising spend on Magnite rose 56% between January and July 2026, with more than 5,800 new live sports advertisers joining. A Gracenote survey published September 1 found 26% of sports fans often unable to work out which channel carries a specific game, rising to 35.6% among viewers aged 18 to 34.

That discovery problem was on display the same week. Fubo signed a multi-year deal to carry regional sports networksfor the Carolina Hurricanes, Columbus Blue Jackets, Minnesota Wild and St. Louis Blues, announced September 28 and launching before the 2026-27 regular season opened on September 29. All four teams lost their television home when FanDuel Sports Network shut after the 2025-26 season, and the league now produces the telecasts itself and signs distributors one by one. Todd Mathers, Fubo's executive vice president of content strategy, said subscribers can continue to enjoy their local teams despite the rights churn. Channel names, game counts, pricing and plan placement were all left unannounced. Fubo carries roughly 5.75 million North American subscribers and is about 70% owned by Disneyfollowing the October 2025 merger. Fans in those four markets can now reach the games through Fubo, Prime Video at $19.99 a month or $99.99 a year for some teams, Spectrum cable, or selected over-the-air broadcasts. Four routes to one local game is not a discovery improvement.

Disney supplied the week's other streaming number. Toy Story 5 recorded 20 million views in its first five days on Disney+, September 23 to 27, the service's most-watched premiere week for a film and a top-ten theatrical debut on the platform. The definition matters: Disney counts a view as total stream time divided by runtime, so at roughly 102 minutes the figure implies about 34 million hours, not 20 million people. Inside Out 2 managed 30.5 million views over the same window in September 2024. The film took $1.15 billion at the global box office and reached the platform 96 days after its June 19 theatrical release, with franchise lift of 168 million hours since the teaser landed on November 11, 2025. For advertisers the disclosure is incomplete in a specific way: Disney reported 122 million ad-supported Disney+ and Hulu subscribers in early 2026, but does not run advertising on children's profiles or preschool content, and published no split by tier, profile type or market. The ad-supported share of that 20 million is unknown.

Game: Half of this industry exists to audit the other half
Sixteen terms, four groups of four, and a Friday grid pitched at the hardest band of the week. One group is a great deal less obvious than it looks.

Five thousand games, and the parents playing them at 3am

Overwolf launched an audience mapping tool on September 17 and PPC Land examined it on September 28. The pitch is that a media brief can be matched against players across more than 5,000 titles rather than a handful of flagship games.

The underlying platform, Gamer Grid, draws on more than 3 billion gameplay signals and 113 million monthly consumers across eight markets: the United States, Canada, the United Kingdom, Germany, France, Italy, Spain and the Netherlands. Nathan Lindberg, vice president of global brand partnerships, and Shahar Sorek, chief marketing officer, fronted the launch.

The worked example is US parents who play, and it is the sort of segment that would not survive a panel-based methodology. Overwolf identifies 3.75 million of them. Of those, 1.9 million, or 51%, play after 6pm; roughly 940,000, about a quarter, play between midnight and 6am; and around 625,000 play 35 or more hours a week. Narrowed to parents of children aged nought to three, 53% are in market for a mortgage, 26% for a vehicle, with minivan purchase intent running 1.9 times the baseline. Sixty-seven percent identify as coffee lovers, with a Starbucks spending index of 1.5, and 61% report child-influenced shopping.

The commercial context is that Overwolf paid $300 million to game creators in 2025 and targets $1 billion in cumulative creator payouts by 2030, having paid $201 million in 2023 and consolidated gaming ad tech through the NitroPay acquisition. Renewing brands for 2026 include McDonald's, The J.M. Smucker Co., Universal Pictures, Paramount, Logitech, Samsung and Amazon.

Undisclosed: pricing, activation routes, the identity of third-party data providers, match rates and any independent measurement partnership. The IAB has published standard metrics for gaming advertising measurement, which makes the absence of a named verification partner conspicuous rather than unusual. A 1.9x minivan index is a claim about inference quality, and inference quality is exactly what the AppLovin filing three paragraphs of this newsletter earlier is fighting over.

Game: Most of the craft is knowing which word applies where
Sixteen terms, four groups of four. Four working vocabularies that rarely meet, and one group that looks settled until the fourth tile refuses.

Google's week of small failures and quiet policy edits

Nothing Google shipped this week carried an announcement, which is itself the pattern worth recording. Search Engine Roundtable logged four separate items in two days, none of them launches.

Google Analytics went down on the evening of September 29. The company's status page acknowledged it was investigating at 8:57pm Eastern and, two minutes later, said service had been restored for some users with a full resolution expected in the near future and no timeframe attached. About an hour after the first report, Google declared the problem resolved and apologised. Barry Schwartz, who filed the report, noted he saw no access problems himself and few complaints, which suggests a partial outage rather than a global one. An analytics platform that goes dark for an hour on a Tuesday evening in the final week of a quarter is a reporting gap somebody will find later.

Google Business Profiles gained a policy the same morning. The guidelines now prohibit misuse of support channels, naming "false or misleading reports" and "reports made in bad faith" as prohibited conduct, and warning that repeated unfounded submissions can result in all further complaints being rejected without assessment. The stated reasoning is that Google provides the channels in good faith for legitimate business and technical issues. The practical effect lands on the local SEO practice of reporting competitor listings, and it arrives after a run of tightening in the same product, including stricter verification on business links and a ban on countries and states in service areas. A complaints channel that can be switched off per complainant is a meaningful change in how local disputes get resolved, published as a guideline edit.

Two interface items rounded out the set. Google is testing two rows of action buttons in local business panels on mobile, spotted by Khushal Bherwani, after previous experiments with a "more +" expander and horizontally scrollable carousels. Separately, a bug is placing map pins for service area businesses in the middle of the ocean, reported by Dipa Gandhi, affecting profiles that do not publicly display a physical address. The service area itself renders correctly; only the pin is adrift. Service area businesses are the category least able to absorb a mapping error, since the map is the only spatial signal a customer gets.

On the regulatory front, Google filed appeals at the EU General Court against the Digital Markets Act decisions requiring it to share search history data with competitors. Oliver Bethell, senior director of competition, said the company is "appealing decisions that will force us to share people's private search history." Nick Fox, senior vice president of knowledge and information, framed search queries as the most private questions a person asks anyone and raised cybersecurity exposure once data leaves Google's own data centres. The compliance deadlines are January 2027 and July 2027, the original portability mandate dates from June 2026, and non-compliance carries up to 10% of annual global revenue, rising to 20% for repeat violations, with periodic penalty payments of up to 5% of average daily sales available alongside.

Brands buying a moment rather than an audience

Two campaigns this week bought a window of time rather than a demographic.

Hoka is running a five-week programme around the New York City Marathon, which falls on November 2. The activity starts with a block party on October 3 and covers paid social on TikTok and Meta, six creator partnerships, digital out-of-home across the city and a sponsored Strava challenge tracking running progress at borough level. Allie Tsavdarides, vice president of North America marketing, is running it with Jellyfish, where Michael Walsh Kirwan is vice president and creative director. The logic is that the race drew more than 59,000 finishers last year, and the training period before it is longer, cheaper and less contested than race day itself, when Nike, New Balance, Brooks, Adidas and On are all present. Sky Canaves of eMarketer, an ultra-runner, and Dan Salkey of the indie agency Small World both weighed in. Hoka posted $704 million in first-quarter 2026 global revenue, up 8%, with parent Deckers Outdoor Group guiding to low double-digit growth for the rest of the year; selling and administrative expenses rose 13% to $420 million over the same quarter, which is the spending this campaign sits inside.

TikTok took the other route, buying its way into a live event rather than the period around one. Its LIVE team ran Music on Stage through August, inviting musicians to submit original videos under a hashtag to compete for a festival slot, and placed the winners on stage at All Things Go from September 25 to 27. The US regional winner was Las Hermanas Machado, a Cuban sister trio based in Miami, appearing on a bill that included Zara Larsson, Hayley Williams and Mitski. Shen Gao, TikTok's head of LIVE operations for the Americas, described creators sharing a stage with established names. All Things Go has programmed around women and LGBTQ+ performers since 2014, against roughly 21% to 22% female representation on major festival lineups. The New York date at Forest Hills Stadium was cancelled for severe weather and will not be rescheduled, which is the risk in buying a fixed date rather than a season.

Both are discovery plays without a discovery surface attached, which distinguishes them from most of what appears above. Hoka is betting that attention accumulates over five weeks of training; TikTok is betting that a platform contest converts into physical presence. Neither depends on a browser resolving a domain, a currency being accredited, or an attribution window holding still.

Also noted

  • September 28: Cloudflare published BEACON, an open dataset of Core Web Vitals from the 10,000 largest sites on its network, updated daily through BigQuery, showing Safari on iPhone at least 10% worse than Chrome-based browsers on largest contentful paint or interaction to next paint in 46 countries and 50% worse in Cambodia. PPC Land
  • September 29: Meta created a business unit called Meta Enterprise Platform and hired Chirantan "CJ" Desai, MongoDB's chief executive of ten months, as chief enterprise platform officer reporting to Mark Zuckerberg, with no pricing, dates or customer names disclosed against $59.36 billion of second-quarter advertising revenue. PPC Land
  • September 29: Advocate General Manuel Campos Sanchez-Bordona's opinion in C-458/25 holds that Article 83(7) of the GDPR bars national rules shielding privately constituted, publicly subsidised schools from fines, over a 1,000 euro penalty stemming from a July 2019 parental complaint about a pupil wellbeing survey. PPC Land
  • September 28: Lob's fourth annual consumer report, surveying 2,000 US adults in July 2026, headlined that 62% say AI has made them warier of digital advertising, a figure drawn from what appears to be a half-sample question the release does not flag, alongside a 34% trust figure whose chart shows a base of 690 respondents. PPC Land
  • September 29: Alphabet's drone delivery arm Wing appointed Minjae Ormes as head of marketing, moving from vice president of global brand and consumer marketing at LinkedIn, with earlier stints as founding CMO of Visible and seven years at YouTube, as Wing expands partnerships with Walmart, Lowe's, DoorDash and PepsiCo. Adweek

By the numbers

  • $59,400 - Australian penalties paid by two retailers whose countdown timers outlived their own deadlines. PPC Land
  • 60% - compliance cost reduction self-reported by early users of Amazon's consolidated product testing. PPC Land
  • 24% - extra unique reach measured when vertical video buying was pooled into one DV360 workstream. PPC Land
  • 2,185 euros - what a Dutch platform recovered after proving its regulator acted unlawfully in 2019. PPC Land
  • February 2027 - when unrendered display advertisements stop counting toward invoices in two Google platforms. PPC Land