Five items and one costume. This week, every process that calls itself competitive turned out to have one guy in a trench coat writing all the bids. WPP Production put in writing a plan to supply all three of a client's "competing" quotes itself. Amazon is defending itself against an invented-bidder lawsuit by bragging that most of its ads don't go to the highest bid anyway. Prebid, the publisher-first auction code, is now chaired and run by sell-side executives, and nobody noticed because nobody ever notices Prebid. Meta sells the AI that writes the ads and is now being sued over what it wrote. Competition is the costume. Check who's wearing it.

Programming note: no Weekly Reckoning this week. Rosh Hashanah begins Friday at sundown, and the podcast is taking the holiday off along with its host, who will be busy atoning for a year of calling holding companies what they are. Shana tova. The auctions will still be opaque when we get back. They always are.

Leaked internal documents from WPP's production arm, reported by MediaPost, state the objective with the subtlety of a parking ticket: avoid traditional commercial triple bidding and make WPP Production the client's default partner. Then comes the workaround, which is the good part. If a client insists on three bids, the document suggests WPP can produce all three internally, from different markets, locations or director treatments, so the job never leaves the building. Triple bidding exists for price discovery, for creative competition, and because historically the agency running the bid didn't own the production company it was recommending. Three quotes from one P&L is not a triple bid. It's one seller with three letterheads and a straight face.

WPP says the document was selectively quoted and the process is "completely transparent." Read the denial the way you'd read a teenager's alibi. It does not say the document is fake. It does not say the quoted lines mean something else. It says WPP follows client contracts and procurement requirements, which is carrying a lot of weight, because a contract that says "three bids" instead of "three unaffiliated bids" makes the house-brand version technically compliant. The APA called it a threat to the free market in commercial production. AICP called three in-house bids a plan to mislead clients into thinking they're shopping when they're not. And the calendar does the rest: WPP is currently the only bidder left in Coca-Cola's global media review. The holding company that wants to be the default in production is already the default in media, by forfeit. This is principal media with a camera crew and craft services. Clients, pull your production contracts and search for the word "independent." If it isn't there, congratulations, you've been triple-bid by one company.

WARC now forecasts retail media at $200.4 billion globally in 2026, rising to $223.4 billion in 2027. The pitch that built that mountain is closed-loop measurement: the retailer can connect the ad to the sale. The flaw in the pitch is the same one sitting in the FTC's Amazon complaint. The retailer sells the media, owns the data, writes the methodology and grades the result. It's a student who wrote the exam, took the exam, graded the exam and then invoiced you for tutoring. Exposed shoppers buying more than unexposed shoppers is not incrementality. It's loyalty members being easy to find.

Now savor Amazon's own rebuttal to the FTC's invented-bidder allegations, as reported by AdExchanger: roughly 92% of placed Sponsored Products ads are not awarded to the highest bid, because Amazon prioritizes relevance, and advertisers supposedly kept more than $8 billion from 2021 to 2025 as a result. Read that as a buyer, slowly. The company accused of inventing an auction participant is defending itself by saying both the winner and the price are chosen by a system only it can see. That's not a defense. That's a confession with good posture. Sarah Caputo of Fraction Method nailed it to AdExchanger: "auction" has become a marketing word instead of a defined term. Nobody outside Amazon can audit the logs, the budgets keep migrating anyway, and the industry that spent a year dunking on The Trade Desk has suddenly lost its voice. Two hundred billion dollars, graded by the people who sold it.

Let's be honest with each other. Nobody outside the plumbing cares about Prebid. That's not an insult, it's a headcount, and it's exactly the problem. Prebid.org named Garrett McGrath, its former longtime chairman, as only its second president on Thursday. McGrath left in June after five years as board chairman, following his exit as SVP of product management at Magnite; since he no longer worked for a board member company, he had to give up the chair. OpenX CTO Joel Meyer was elected chairman at the end of August. McGrath starts September 15 and co-leads with outgoing president Racic until Racic heads out October 15. The new boss, as they say, is the old boss with a new business card. McGrath's pitch to AdExchanger: "If advertising pays for the internet, then Prebid is critical infrastructure for the internet."

He's right. Which is why somebody should be watching the infrastructure. Prebid.js is the room where most open-web header auctions happen, and the rules of that room are written by committees drawn from companies that bid in it. Last week ADOTAT traced an SSP keeping roughly $20 of a $30 CPM before the publisher saw a dime. The governance of the auction code sits with a board whose chair seat requires a job at a member company, and the two people now at the top came from OpenX and Magnite. Publishers are being told the referee is on their side while the referee's résumé reads like the other team's roster. Meanwhile Prebid is rolling out a Sales Agent as it starts building for agentic buying, which means the rules for how AI buys open-web inventory are being drafted right now, in a room the trade press covers by reprinting the GlobeNewswire. Publisher-first infrastructure, run by the sell side, watched by nobody. Sleep tight.

Three Facebook and Instagram users filed a class action in San Francisco federal court alleging Meta ads funneled them into WhatsApp groups running crypto schemes, and that Meta's own generative AI tools cranked out hundreds of ad variations optimized to hook vulnerable users. Same week, a Clutch survey of 601 consumers: 86% have run into AI-generated content on social, and 53% say they're less likely to buy from brands they know use it. Asked where they see the slop, 66% named Facebook versus 44% for Instagram. Mark Zuckerberg built a platform so good at AI content that its users now assume the whole thing is AI content. That's a moat, just not the kind you want.

Here's the legal math nobody at Meta wants on a slide. Section 230 protects hosts, not authors. In a similar case last year, a district judge refused Meta's early bid for dismissal, reasoning that if the allegations held, Meta had too big a hand in making the ads to hide behind 230 (the case later died on securities-law grounds, not because Meta was innocent of anything). Every AI creative tool Meta sells advertisers moves it one step from bulletin board toward copywriter. Meta reported a 91% year-over-year drop in Q2 free cash flow while building its agent business, and this week it bought Swedish startup Stilla.ai to automate business messaging. So the pitch to advertisers is that Meta's AI does more of the work, and the pitch to courts is that Meta just hosts it. You can't be the copywriter and the corkboard. Pick one. Consumers already did.

Yes, Mr. Pibb. The soda your uncle drank in 1983. Coca-Cola relaunched the 54-year-old brand in late 2025 with a new recipe and more caffeine, and purchase consideration jumped 11.2 points, with Gen X leading at 15.5. The runner-up, Jimmy Dean, gained 4.7. Mr. Pibb more than doubled the sausage. The backdrop: only 14% of brands grew purchase consideration in 2026, a five-year low, compared with 94% in last year's report.

Two things break here. First, a tracker where 94% of brands grew one year and 14% the next is measuring the national mood at least as much as the brands, and consideration is a survey answer, not a sale. Nobody's buying a boat off a consideration score. Second, the Coca-Cola part, which is delicious. The company has spent the year running a global media review about who operates its data and intelligence systems, the review is now down to one bidder, and its biggest brand win of 2026 came from the product team changing a formula on a drink older than most of its media planners. Morning Consult's read is that squeezed consumers are reaching for familiar, reliable brands. The most effective marketing Coca-Cola did this year fit in a can, and no holding company had to be in the room for it.

The IAB raised its 2026 U.S. ad spend forecast 2.8 points to 12.3%, with social at 16.5%, CTV at 15.6% and commerce media at 13.6%, while linear TV wheezes along at 1.5% and paid search slips to 8.1%; CEO David Cohen's summary was that growth exists but there are no easy wins, which is what trade associations say instead of "good luck." NP Digital scored fintech brands across 68,334 AI-generated answers, with Apple on top and PayPal collecting the most mentions but ranking third because rivals kept getting named first, a lesson in which being talked about is not the same as being picked. Prebid's developer conference lands in Cologne on September 22, for the dozen people reading this who will actually be there. You know who you are. We salute you.

Five stories, one pattern: the appearance of choice is cheaper to manufacture than choice itself. WPP can write three bids. Amazon can crown a winner that isn't the high bidder. Prebid can call itself publisher-first with the sell side at the head of the table. Meta can host the ad and write it too. The costumes keep getting better. The logs are still locked.