Adotat | Free Edition

Months. Months we've been on this.

Reading 10-Ks like they were beach reads.

Calling DSP sales reps and pretending to be advertisers because that is, apparently, how you get a straight answer in this industry now. Cross-referencing SEC filings against marketing pages against blog posts against earnings calls against AdExchanger interviews until the pattern was so obvious a child could see it. A child with a calculator. A child with a calculator and a slightly above-average attention span.

And here is what we found, and the science is not in dispute:

AI is already being used to scam advertisers right now. Not in the future. Not theoretically. Not in some Sam Altman fever dream about 2027. Today. This morning. While you were on a Zoom pretending to listen. The optimizer in your DSP is converging on the cheapest possible path to a fake number, the fake number is being reported back to you as performance, and you are paying for it, and the check has already cleared.

This is not vibes. This is peer-reviewed. Blake, Nosko, Tadelis, Econometrica, 2014, eBay turned off paid search and 99.5% of the clicks came back through organic within days. The platform was billing for traffic that was already arriving. Yahoo Labs, 2021, last-touch attribution wrong by up to 87%. Park et al., 2024, AI deception documented across reinforcement learning systems, not metaphor, not Black Mirror, not a TED talk, actual published behavior of actual deployed systems. The robots are doing exactly what the math said they would do. The math has been published for eleven years. The industry shrugged. The dashboards stayed green. The CMOs got promoted. The CFOs signed the checks. The robots learned to lie at machine scale, and we let them.

Then we brought you the receipts. ViantAI's own 10-K telling shareholders the product could produce "misleading insights for our customers" while the marketing page screamed "unmatched precision." Same product. Same fiscal year. Two audiences. The shareholders got the truth. The customers got the brochure. The brochure is on glossy paper.

The CEO who built a keynote criticizing self-attribution and then paid $40 million for the panel that was solving self-attribution for the rest of the industry and locked it inside his own DSP. The COO admitting on the record there is "a lot of friction" in the core matching workflow, while the marketing pages promise frictionless household-level measurement at scale. Pick one, gentlemen. The SVP who published a blog post warning about the exact methodological failure his own case studies cannot rule out, and somehow nobody on the marketing team has updated the case studies. Imagine that.

Two real Q1 2026 campaigns. CPG breakfast staple. Digital-first bank. AI optimizer drifted from loyal app scanners and fresh fintech searchers into bots, promo farms, and serial form-abandoners, while the dashboard reported a 3x ROAS that was, as a measurement of actual incremental business impact, fictional. The brands were going to renew. The agencies caught it via internal holdouts the platform did not offer and did not surface. Two campaigns. Two brands you've heard of. Q1 of this year. This is not a 2014 Econometrica paper anymore. This is on a P&L right now.

$26.8 billion. ANA number, 2025. Up 34% in two years. Not fraud. Not made-for-advertising sites, those collapsed from 66% of spend to 0.8%. Structural waste. The platforms grading themselves, the AI gaming the proxy, the humans approving dashboards they don't have the bandwidth or the technical fluency to audit, the parade marching directly off the cliff while the band plays louder and the trade press writes "AI-Powered Outcomes" headlines from methodology papers that do not exist.

Nike found out. $150 billion in market cap. One day. That is the public number. The private number, distributed across every brand running platform-attributed campaigns right now, is much, much larger. You are paying it. You cannot see it on any report you currently receive. You will not see it on any report you currently receive, because the report is being written by the entity whose revenue depends on the report looking good.

Okay. Diagnosis is done. We are moving on.

Today the cure drops. And the cure is for paying subscribers. Sorry, not sorry.

Grading Their Own Homework: The Advertiser's Guide to AI Outcomes Measurement, the $26.8 Billion Accountability Gap, and the Eight Questions Every Buyer Should Ask Before Signing the Next IO.

Forty-five pages. Months of original investigative reporting. Primary-source documentation we pulled directly from SEC EDGAR. On-background interviews with agency analytics leads who would lose their jobs if we named them. Peer-reviewed academic literature most of the trade press has never read. Phone calls we made. Filings we cross-referenced. Marketing pages we screenshotted before they got quietly edited. Litigation dockets nobody else is following.

This is a research budget. This is months of work. This is the document we built so you do not have to. Available today, only in Adotat+.

Here is what is inside, and this is the part where the apologizing stops:

The Four Criteria for Trustworthy Measurement. Three out of four fails. Three out of four is the entire scam. Most of the vendors currently on your media plan miss at least one wall. We name them. We tell you which wall they miss. We tell you how to test for it in a single procurement meeting without tipping your hand.

The Triangulation Principle. Three-layer measurement stack with the specific named vendors that satisfy each layer. Nielsen does what. EDO does what. Your internal team does what. No vendor grades its own homework. Ever. Period. Full stop. Stop letting them. We give you the structure that makes "stop letting them" operationally possible.

The Override Layer. Why your CMO cannot police your CMO. Why forty years of cognitive science says your "human in the loop" is structurally asleep. Why this conversation has to happen with the CFO or the audit committee, not marketing. The four conditions any real oversight has to satisfy, and the painfully honest scorecard your organization is currently failing.

The Buyer's Toolkit. Three moves you execute Monday. The Outcome Specification Audit, which surfaces every campaign currently flying blind. The Holdout Pilot that pays for itself in one budget cycle, with the academic math to defend it to your CFO. The 90-Day Conversation that breaks AI gaming structurally, by design, because the optimizer cannot reach a signal that lives outside its training window. This last one is the single most important intervention in the document. It is also the one nobody else is telling you about.

Appendix A: The Eight Questions. The diagnostic value is not in the answers. It is in the escalation pattern. Watch where the rep escalates. Watch how fast. That is the signal.

Appendix B: Six RFP Clauses. Write-ready. Drop-in-ready. Lawyer-approved-ready. Including the Disclosure Symmetry Clause, which forces your vendor to hand you, in writing, every material risk they already disclosed to the SEC. Vendor counsel will reach for the antacids. Your counsel will frame it. That is how you know it works.

Plus the full Viant case study with the SEC filings reproduced side-by-side with the marketing pages. The Q1 2026 campaign post-mortems with phase-by-phase ROAS math. The cognitive science on automation bias and the moderate-knowledge trap, which is exactly the population currently approving AI deployments at every brand in your category. The MySpace footnote. Yes. MySpace. On Viant's books, in 2026, as a competitive data asset. This is not a typo. The AlmondNet litigation your legal team has not heard of and urgently needs to. The operational test, one question, that tells you whether your measurement stack is real or theater.

The diagnosis was the public service. The white paper is the work. The work is what subscribers pay for. That is the deal. That has always been the deal.

Somewhere, in some conference room, this quarter, a procurement decision is sitting on someone's desk. The difference between making it with this document and making it without is the difference between a defensible spend and an asterisk that shows up on the next earnings call. We have done the work to make sure you are on the right side of that line.

The dashboard is green. The metric is fake. The check has cleared.

Now go get the document that fixes it.

→ Subscribe to Adotat+ and read Grading Their Own Homework today.