There is a moment in this corpus worth playing eleven times, which is roughly how many it takes to stop being funny and start being diagnostic.

The co-CEO of a European video ad tech company, a man with twenty years in the business who now runs an agentic AI initiative as the centerpiece of his company's strategy, is describing a meeting with his own key clients. He came in ready. He had the agents. He was, in his own account, expecting standing ovations.

The room rolled its eyes and looked at their phones. More than once, it seems.

So he asked what the problem was, because he has known these people a long time. And they told him. Everybody is talking about AI. We have heard too much about it. There is AI fatigue. Please do not bore us with your agents. They are probably cool. It is an overhyped topic.

Precision about the speaker matters here. This is not a critic. This is not a skeptic with a Substack and a grievance. This is a vendor, describing his own buyers refusing his own product on the grounds that they are sick of the category. He then went on stage at an industry conference and told several hundred people it had happened to him, a level of candor that would improve this business considerably if it spread and shows no sign of spreading.

He said something else that afternoon about the economics of the model business, which turns out to be the most useful sentence anyone in the industry has produced about AI in two years. It comes later.

A distinction is useful before going further, and it belongs to the philosopher Harry Frankfurt, who spent a slim and famous book working out the difference between a liar and a bullshitter.

A liar knows the truth and points away from it. He has to. Lying requires respect for the facts, because nobody successfully steers around something they have not located.

A bullshitter has no relationship with the truth at all. He is not steering around it. He has not located it. He is saying whatever the moment requires, and whether it happens to be accurate is not a variable he is tracking. Frankfurt's argument, worth revisiting every time a conference agenda lands in an inbox, is that the second thing is worse. The liar is at least oriented. The bullshitter has broken the instrument.

The distinction matters because of where the charge in this piece lands.

This industry is not lying about AI. Careful review of these recordings turned up almost no statement the speaker appeared to know was false. What turned up instead, session after session, across nearly every category of company, was an enormous volume of speech that could not be checked, delivered with total confidence, by people showing no apparent awareness that checkability was something anyone might want.

That is the condition. And it matters not because it is irritating, though it is, but because an industry that stops noticing whether its claims are checkable will eventually stop being able to tell. That experiment has been run before. It was called viewability. It was called brand safety. It was called attribution.

Here is what the tools actually do when the demo is over and a working executive is using them on a Tuesday.

A publisher-side executive who works on yield and pricing is describing her team's experience testing AI creative tools. She is careful, and she declines to name the product, which is more grace than the product earned.

Her husband asked the tool to sharpen a photograph of their daughter.

It put an axe into the child.

Her account of what she said next is the most economical description of the current state of generative tooling available anywhere. No. That is not what we meant. But thank you for playing.

She is not anti-AI. She spends the rest of that segment arguing that these tools free people to work on ideas rather than on building the rails underneath the ideas, which is a reasonable position held by a reasonable person. Her operating rule is simply that you have to check your work, because hallucination is real, because as she puts it, God knows whether the answer you are getting is correct, and because she has been burned.

Hold that against the register. In public, the category is a superpower, an unlock, a transformation. In practice, at the level of one competent professional doing one small task, it is a tool that occasionally puts an axe into your daughter and requires supervision every single time.

Both of those things are being said by the same industry in the same year. Only one of them is on a slide.

The visible layer deserves a short section and no more, because it has been covered and it is the least interesting thing happening.

At the bottom of the market, generated imagery has removed the last cost floor under outright fraud. The coat in the ad was never manufactured. Selling a garment used to require a photograph of a real thing, which required a real thing. No longer, and the drop-ship storefront selling a coat that does not exist is now a business model with a supply chain and a chargeback window.

One rung up is slop, which is the correct word and requires no apology. Uncanny hands. Background text that is almost English. A face that changes shape between the second and third second.

None of it is confined to the bottom, and the industry knows it. A senior figure at a large marketing association, who is also an author and an educator, walks through the problem without flinching. A major beverage brand ran a holiday campaign that was entirely AI generated. It drew heavy criticism. In his description it was granular and did not look like professionally filmed video. His summary of the moment is that the whole industry is trying to figure this out in real time while flying, that there are bad characters in the ecosystem, that brands can be damaged, and that a lot of money can go out the window.

That is an accurate description of a mess, delivered by someone with a good view of it. What he does immediately afterward is the subject of Part One behind the paywall, and it is not what you would expect.

Meanwhile reporters in San Francisco this summer found a neighborhood pizza chain whose walls were papered in generated imagery, including a famous soccer player, name misspelled, holding a slice. A business with a lease and a health inspection, hanging a defamation risk on the wall because the alternative cost four hundred dollars and a photographer's afternoon.

Consumers are not fooled by any of it. The finding is stable across the trade research and across the ambient roar of ordinary people complaining about it on Reddit and YouTube to audiences larger than any publication in this business will ever reach.

One panelist stated the systemic version better than any research has. Individual brand discipline does not save anyone, he argued, because being the best driver on the road does not help much when the road is full of drunk drivers. As a consumer, he said, he is not going to notice that one particular brand gets him. He is going to want to turn off the television and the phone.

He is right. It is the most important consumer insight in the entire corpus. The conversation moved on about nine seconds later.

Set the creative argument aside for a moment, because there is one place where AI has produced an enormous, measurable, entirely uncontested effect on this industry.

It is a subtraction.

An adtech consultant and analyst, discussing what is actually happening to publishers, is blunt about it. Any publisher drawing a large share of traffic from search, he says, is really screwed. He cites the figure he has seen for one large digital publishing group: a 90 percent traffic reduction. His assessment of what that means for a business needs no elaboration and he gives it none. It destroys the business.

A platform CEO, speaking separately and with a considerably softer message to deliver, puts the broader industry decline at around 20 percent of search traffic across the publishers he sees, attributes it to AI answer surfaces, and says plainly that it will continue to go down, that he does not know how far, and that the platform driving it is highly motivated to keep moving queries into that surface because it makes more money there.

Two speakers, different incentives, different numbers, same direction.

The consultant, who mentions that he is a former journalist, does not leave it as a business observation. Small and midsize publishers, he says, will have to branch into licensing, events, merchandise, and he does not think that will be enough to keep many of them commercially viable. And then the part that has nothing to do with ad tech at all: no republic functions properly without a good news industry, and the business model that used to finance good journalism is not there anymore.

So here is the scoreboard on this technology's measured impact on media, as described by the industry itself, on the record, in public, this year.

Creative performance: contested, unmeasured, mostly asserted. Publisher traffic: down somewhere between 20 and 90 percent depending on exposure, uncontested by anyone.

The keynotes are about the first one.

The most valuable evidence in any argument like this comes from practitioners who ran the actual test, and there is one in this corpus.

The data lead at a performance agency manages millions in search and millions in social. And because some inventory is only available through the platforms' automated products and some is available everywhere, his team runs campaigns in parallel: manual campaigns, and the platforms' AI-driven campaign types, side by side, at the same time, on the same objectives.

That is a controlled comparison. Very few people in this industry are running one.

His finding, delivered without drama: for many of their more sophisticated backend KPIs, trusting the black box does not work in most cases.

He is not saying the automated products are useless. He is saying that when you optimize toward the metrics that actually matter to a business rather than the ones the platform reports, the box loses. And he describes what the platforms would prefer, which is that you hand over your American Express, set no budget cap, and trust that the right ad will find the right person.

Nobody on that recording disputes him. The conversation moves to whether the loss of control is inevitable.

So why does adoption keep climbing against all of that?

The standard answers are correct as far as they go. It is cheap. No photographer, no retoucher, no copywriter, no studio, no talent, no usage rights. That is not a saving, it is the deletion of a cost center, and a CFO who ignored it would be committing a species of malpractice.

It is a relevance hedge. Large advertisers have concluded that an ad people find mildly annoying is still an ad people saw, and that mild irritation costs less than vanishing. That is a defensible read of a century of frequency research, however unclean it feels.

Neither of those explains the talking.

Cheapness does not require a keynote. Cost reduction has never required a keynote before in the history of this business. Nobody ever booked a ballroom and a hype reel to announce a renegotiated print rate. So why does this particular cost decision need a stage, a framework, a crystal ball question, and four separate uses of the phrase "exciting time" inside a single answer?

The gap is now conspicuous. The tools require supervision every time. The only uncontested number is a 90 percent traffic collapse. The practitioners running controlled comparisons report that the box loses on the metrics that matter. And the buyers, by the vendor's own account, are rolling their eyes.

Every one of those statements came from inside the industry, on the record, into a microphone. And none of them made it into the story the industry tells about itself.

Something is being performed. Eleven months of the performance sits on a drive, timestamped, and it can be sorted.

So it was. Every substantive claim about AI in the corpus, divided into statements that could in principle be checked and statements that could not.

The second pile is enormous.

The first pile has four things in it, and one of them is a number that inverts the entire public story.

Karsten Weide has spent thirty-five years in this industry and arrived, in the fourth decade, at a position that should be impossible to hold. He holds it without strain.

He started as a journalist at Ziff Davis Germany in the PC era, when Ziff Davis was the biggest computer publisher in the world and its name was on every manual you owned. He got out of journalism by accident, pitching a CompuServe forum to accompany a Windows magazine. His boss asked what it would cost, paid for it out of marketing, and then, as Weide puts it, "the damn thing actually made money." That was the end of his print career.

Then Yahoo, which for anyone under thirty-five was Google before Google. Heather Killen, an Australian he calls the dragon lady with what he insists is equal parts fear and respect, asked if he wanted to build Yahoo Germany. Day one, the Ziff Davis liaison walked him into the office space: "There's literally nothing there. It's just big office and there's a landline telephone sitting next to an outlet at the wall. That was it. That's how we started Yahoo Europe."

Sixteen years at IDC followed, which is where he learned the thing that separates him from everyone else who talks about this business in public. "Research sometimes tells you stuff that you don't want to hear." He describes a content delivery network that arrived with a gigabyte of session data from a global sporting event, certain HD drove better engagement and ad effectiveness than SD. "We spent weeks on these data, massaged them in any which way we could imagine, burned the midnight oil, and no matter what we do, we couldn't find an effect." He reported that, because that was the finding.

Now he is independent. Nobody pays him to be excited.

So: is AI in advertising real, or the next metaverse? "No, it's absolutely the real deal." A revolution within a couple of years. He means it.

Rapid fire, same conversation. Name the trend everyone is excited about that you think is overhyped. One word, no hedge.

"AI."

Both, simultaneously, without noticing they might need reconciling, because to him they do not. The technology is real. The talk is not. "Everything's AI now. The toaster is AI now." A month before, these same companies were machine learning. "That's probably the most disappointing part about the AI revolution, that there's a lot of talk about AI that's not really AI." On the industry-first fully automated agentic campaign everybody applauded: "I think they were still pressing the button, telling the agent to do it."

Then the one he said on camera, knowing it was rolling.

"The industry is lying. Everybody knows the industry is lying. And if we all just stopped, we'd have more fun."

He does not mean a few bad actors. He means the Google verdict: "They lied, they cheated, they broke the law." He means the model companies: "They basically steal content." And the part nobody funded by this ecosystem will say: "Not only do they steal content, they also destroy legitimate businesses that are providing livelihoods to people." The New York Times can fight back. "Dogs.com, like a mom and pop website, they can't."

He means the conference circuit, and he means you. "Everybody is so excited to go to every conference, but when you talk to them, they hate it." LinkedIn used to be about connecting for business. Now: "It's about people hyping their business."

And then the sentence that makes him worth quoting instead of insufferable: "Everybody is in on it. Everybody knows it's a lie and everybody plays the game. Including me sometimes. It's not like I'm better than anybody else, but it makes you wonder why are we even doing this?"

Play him the standard sentences and he sorts them cold. Our platform drives incremental lift: "True." We take brand safety seriously: "Half true," because companies want to protect brands and the tech will not let them. "They only do an 80% job. Not because they are unwilling, it's because the tech is not." Programmatic beats direct: half true. AI will transform advertising: "Absolutely."

We put the consumer first.

"No. It's untrue."

The biggest lie the industry tells itself every day? "We are really important." And on the people who claim otherwise: "The people who say that we're saving democracy. I'm not sure that's true."

He is not gentler about the work. On Super Bowl creative: "Man, they were lame. I mean, it was terrible." On the industry's founding romance, that great advertising is effective advertising: "The assumption that creativity equals effectiveness or efficiency is not necessarily true." He spent years in ad testing on attention tracking. "It's not always the prettiest ad. It's the ad that gets the most attention."

And what is actually coming is not comforting. "A lot of the menial jobs, a lot of the junior jobs, they'll just go away." Why? "If you have a machine that can do a media plan that's faster, better than what a person can do, then why wouldn't you do that?"

Asked what would have to be true for him to look back in ten years and say the industry finally stopped lying, he does not perform hope. "That's very hypothetical, because I don't think it's going to happen. Maybe there will be a renaissance of business ethics. I sure hope that's going to happen, but I'm doubtful."

Given five desert island slots he took his wife, his three kids, and Brad McQuaid, the late developer of EverQuest, a game he has played for twenty-five years and still plays. No colleagues. No analysts. Thirty-five years in media and the one non-family human he wants stranded with is a man who built dragons.

He is careful to say he likes these people. Bright, open-minded, worth talking to. Advertising is a worthy cause.

"However, we're not saving the whales."

Below: the figure that appears in all three contracts, the clause to search your own MSA for, how two former agency finance leads say it gets booked, and the four questions to ask before you sign.

If you are renewing in the next two quarters, find the clause first.

How we reported this. Three sources. A corpus of conference sessions, keynotes, panels and podcasts recorded between September 2025 and August 2026, across US, European and Asia-Pacific events, all public or open-registration with every speaker on the record. A recorded on-camera interview with independent analyst Karsten Weide for The ADOTAT Show. And published material, principally the IAB and Sonata Insights study "The AI Ad Gap Widens" and its 2024 comparison wave. Every quotation was checked against the surrounding audio for context rather than clipped from the qualification that followed it.

Limitations: the corpus is not a random sample, skews English-language, and over-represents vendors relative to brand-side marketers, who speak less often and more carefully. The IAB advertiser sample is 104 executives against 75 in 2024, so movement between waves is directional rather than statistically robust. The workflow range, the 15 percent figure, the query multiplier, the sampling increase and the model-training costs are practitioners' estimates offered in conversation, not audited measurements. Characterizations of third parties are speakers' opinions, not ADOTAT findings. ADOTAT has no financial relationship with anyone named here, and no subject received advance review.