Premium Inventory: Not What Your DSP Says it Means

In July 2026, TAG, the ANA, and Fiducia published the first statistically rigorous analysis of "AI slop" in programmatic media. This term refers to machine-generated filler content found throughout the open web supply chain. Their research found that this content accounts for between 1.3% and 2.4% of open web programmatic spend, which is roughly comparable to the industry's 1.1% level for "made-for-advertising" (MFA) content.

While that is the main headline, the following details are what truly matter:

AI slop outperformed clean inventory on conventional quality metrics. It carried an invalid traffic rate of 0.05% versus 0.32% for clean supply. It posted higher viewability — 77.2% against 74.9%. And once measurability was factored in, it graded as premium more than 70% of the time.

It is worth reviewing those figures again: this “low-quality” content actually performed better than legitimate inventory on every quality signal reported by typical DSPs.

This issue is not a failure of measurement, but rather a failure of definition. The industry developed a vocabulary for quality based on properties that are incredibly easy to manufacture, and content generators realized this before the buyers did.

What "premium" actually means in your platform

When a DSP surfaces inventory as premium, it is almost always compressing some combination of four measurements: viewability, invalid traffic rate, brand safety classification, and measurability. Sometimes a domain allowlist sits on top. That's it.

Each of these metrics is a property of the ad slot rather than the audience.

For example, a page can be 100% viewable if the ad unit is pinned in a "sticky" container above content that no one reads. It can show near-zero IVT because the traffic is technically human, even if those humans arrived via paid discovery widgets with no intention of engaging. It can pass brand safety checks because AI-generated text about mattress reviews avoids unsafe keywords. Finally, it can be perfectly measurable because a site designed for monetization has every incentive to implement measurement tools flawlessly.

A page optimized specifically to score well on these four metrics will often outperform a genuinely valuable page. This is no longer a hypothetical scenario; it is the core business model of made-for-advertising content. Recent ANA data confirms that AI-generated inventory is executing this strategy even more effectively than MFA ever did.

A page optimized specifically to score well on these four metrics will beat a genuinely valuable page on all four. That is not a hypothetical. That is the business model of made-for-advertising, and the ANA's own data now shows AI-generated inventory doing it more effectively than MFA ever did.

The uncomfortable implication: the harder you optimize toward your DSP's quality metrics, the more efficiently you select for inventory engineered to game them.

The gap this creates is enormous, and it's widening

The ANA's Programmatic Transparency Benchmark tracks how much of a given advertiser's spend converts into qualified impressions — fraud-free, measurable, viewable, MFA-free. In Q1 2026, higher-performing advertisers converted 54.0% of programmatic spend into qualified impressions. The lower-performing cohort converted 32.1%.

That 21.9-point spread is the largest the benchmark has recorded. The market-level index sat at 43.3%.

There are two primary takeaways from this data. First, most advertisers see less than half of their programmatic budget go toward impressions that meet basic quality criteria. Notably, this is occurring after three years of industry-wide efforts to reduce MFA spend, which dropped from 15% to the low single digits. The MFA problem was addressed, but the underlying efficiency problem remains because MFA was merely a symptom of a larger issue.

Second, the gap between the highest and lowest performing buyers is growing. As ANA CEO Bob Liodice noted, programmatic performance is increasingly determined by the ability to actively manage quality, price, and measurement while curating supply at scale. Quality has become the true differentiator. Buyers who treat inventory selection as an active discipline are seeing significantly better results than those who rely on default platform settings.

The original ANA transparency work established the baseline that still governs the economics: of every dollar entering a DSP, roughly 36 cents reached a brand-safe, non-MFA, measurable, viewable impression. About 29 cents went to transaction costs across DSP and SSP fees and data. The remainder went to impressions that failed quality on one dimension or another. Following better practices could raise the effective value from 36 cents to 50 cents on the dollar — a 39% improvement in working media with no increase in budget.

Nothing about that math requires a better algorithm. It requires knowing what you are buying.

The three questions that actually separate premium from packaged

We evaluate supply on three axes that no DSP quality score captures. None of them are exotic. All of them require asking a supply partner something they would rather not be asked.

1. Where did the audience come from?

This is the single most predictive question, and it appears in no quality metric.

An impression served to a reader who typed a publication's name into their browser, or arrived via a newsletter they chose to receive, is worth far more than an impression served to someone who clicked a paid recommendation widget. Although they may share the same viewability scores, IVT rates, and brand safety classifications, they represent entirely different levels of human engagement and intent.

Ask any supply partner for the traffic-source composition of the inventory they're selling you. Direct, search, social, paid acquisition. The ones with genuinely good inventory will tell you. The ones who won't are telling you something anyway.

1. Does the content have a reason to exist besides ad revenue?

The old MFA test was structural — endless slideshows, heavy ad density, thin content. AI-generated inventory largely defeats that test, which is exactly why it grades as premium 70% of the time. The pages look fine. The text is coherent. The ad density is reasonable.

The replacement test is editorial rather than structural: would this page exist if it carried no advertising? A publication with subscribers, a newsroom, a masthead, or a real editorial identity passes. A domain producing hundreds of competent articles a week on whatever topics carry high CPMs does not, regardless of how it renders.

This is why the ANA's guidance points toward scrutinizing smaller and native-format exchanges, which showed higher AI slop rates, and toward tighter management of long-tail exposure. Not because small publishers are bad, but because the long tail is where synthetic supply hides most effectively.

1. How many hops between your DSP and the publisher?

Supply path length is the most easily verified of these three factors, yet it is often the most ignored. Every intermediary takes a fee and creates an opportunity for the impression to be misrepresented. The ANA's breakdown of transaction costs (including roughly 8% for DSP transactions and 13% for SSPs) describes a standard chain. Longer chains increase costs while reducing transparency.

Advertisers should request log-level data to verify these paths. To illustrate how rare this transparency is, the original ANA study found that only 21 out of 67 participating advertisers could actually obtain their own impression-level log data. Only 31% of advertisers had access to the actual record of their purchases.

Ask for log-level data. This is the whole ballgame, and it's worth understanding how rare it is: in the original ANA study, of 67 advertisers who wanted to participate, only 21 could actually obtain their own impression-level log data. Just 31% of advertisers had access to the record of what they bought.

If you cannot get log-level data from a partner, you cannot verify any claim they make about quality. Every assertion in their deck is unfalsifiable by construction.

So are PMPs worth it?

Sometimes. The honest answer is that a PMP is a transaction structure, not a quality guarantee, and the market routinely confuses the two.

A private marketplace deal provides a negotiated relationship with a specific seller, priority access, and a clear point of accountability. However, it does not automatically guarantee better inventory. Many PMPs simply repackage open-exchange supply with a deal ID and a higher price tag. Audits have shown that some "curated premium" packages contain the same long-tail domains available in the open auction for a fraction of the cost.

To determine if a PMP is worth the investment, you must be able to answer the three questions mentioned earlier. If the seller provides a domain list, traffic-source data, and log-level verification, the premium is usually justified by the improvement in working media. If they offer only a promise, you are simply paying for a label.

The same logic applies to the broader curation trend. Curation is genuinely the right structural direction — it moves quality control from post-hoc exclusion to pre-transaction selection, which is where it belongs. But "curated" has already become a pricing adjective faster than it became a practice. The word tells you nothing. The inclusion list tells you everything.

What we'd change on Monday

Move from exclusion lists to inclusion lists. This is a highly effective change, though many teams resist it for fear of losing scale. Blocklists cannot keep up with the speed of automated domain generation. You cannot simply exclude your way out of synthetic supply; however, you can certainly include your way into high-quality supply.

Stop treating viewability as a quality metric. It's a delivery metric. It confirms the ad had the opportunity to be seen. It says nothing about whether the person seeing it was worth reaching, and optimizing toward it actively selects for engineered inventory.

Demand log-level data as a condition of doing business. Not as an audit request. As a contract term.

Measure supply quality by outcome, not by grade. Run your own analysis of which domains actually produce conversions, retained customers, or incrementality — then compare that list to the one your platform grades as premium. In our experience the overlap is partial at best, and the divergence is the most valuable thing you'll learn about your media all year.

The word "premium" survived into 2026 with its meaning quietly hollowed out. It now describes inventory that scores well on four metrics, all of which can be manufactured, and at least one category of manufactured supply now scores better than the real thing.

The advertisers who successfully convert 54% of their spend are not necessarily using better technology than those at 32%. Instead, they are asking the critical questions that their platforms do not ask for them.

Sources

• TAG / ANA / Fiducia, AI Slop Industry Briefing, Q1 2026 (July 28, 2026) — AI slop share of open web programmatic spend; IVT, viewability, and premium-grading comparisons versus clean supply
• ANA Programmatic Transparency Benchmark, Q1 2026 — qualified impression conversion rates by cohort; TrueAdSpend Index
• ANA Programmatic Media Transparency Study (2023) and follow-on reporting — working media economics, transaction cost breakdown, log-level data access rates
• ANA / TAG TrustNet benchmark reporting on MFA spend reduction