Podcast episode

The Race to the Bottom: Media Quality, Attention, and Why Cheap Reach Is Costing Brands More Than They Think

by Ken Rona ·July 14, 2026 ·Signal & Noise

brand-building contextual-targeting measurement programmatic publisher-economics

TL;DR

Erez Levin (Emmett Advisory, former Google executive) joins Signal & Noise to discuss CIMM's new "Quality Matters" report, arguing that media quality (MQ) has been systematically underpriced in programmatic buying. The episode makes a detailed case for incorporating placement-level signals — time of day, format, device, geography — into buying algorithms alongside (or in place of) weakening identity signals. Solid industry-practitioner content; worthwhile for anyone in programmatic strategy, publisher monetization, or measurement.

What was covered

CIMM "Quality Matters" report: Co-authored by Levin, Gabe DeRose (ex-New York Times) and CIMM Managing Director John Watts, the report attempts to establish consensus definitions and frameworks for media quality (MQ) — distinct from audience quality and creative quality — as an actionable pricing signal.
The "race to the bottom" diagnosis: Two decades of programmatic optimization around identity signals (cookies, device IDs) and short-term attribution (ROAS, CPA) has systematically undervalued placement-level quality, producing MFA (made-for-advertising, low-quality sites built to capture cheap programmatic dollars) proliferation and the "performance doom loop."
Dual time-horizon framework: Levin argues advertising works both short-term (roughly 5% of audience in-market at any given time) and long-term (brand equity / mental availability for the other 95%), and that attribution-centric buying sacrifices the latter — citing Nike, Uber, Airbnb, and Adidas as cautionary case studies.
Attention metrics are necessary but insufficient: Sound-on, in-stream video scored 8.2/10 on a quality index used in the report; muted out-stream scored 1.3/10. Attention alone misses "receptiveness" — situational context like time of day, geography, and device that predicts whether someone is actually susceptible to a message.
CTV as the key battleground: CTV (connected TV) ad delivery shows ~30% of budgets running overnight at flat CPMs, which Levin calls a diagnostic red flag. He argues CTV has the opportunity to avoid display advertising's quality mistakes but is already showing the same symptoms (mislabeling, inventory misclassification, open-auction remnant pollution).
PMPs (private marketplace deals) are not the fix: Levin argues the shift to PMPs — cited as now ~80% of programmatic display spend and ~59% of CTV — gave buyers false security; they still chase the user/cookie across environments without valuing the environment itself.
AI advertising revenue skepticism: Brief segment on AI chatbot ad revenue — Levin argues AI ads will be a meaningful but much smaller business than traditional search advertising, and that current trillion-dollar valuations implying search-scale ad revenue are significantly overhyped.

Notable claims & predictions

Levin: "Stop trusting the algorithm, the platforms' algorithms, to just spend to fill your budget without accounting for some of these quality dimensions." The call to action is manual or custom-algorithm-driven weighting by time of day, geography, and format before ceding control to DSP (demand-side platform) optimization.
Levin: AI will "mostly accelerate the race to the bottom" by flooding the ecosystem with low-quality ad creative ("slop"), but paradoxically will force marketers to pay a premium for genuinely scarce quality inventory because averaging high-quality and AI-generated content becomes untenable.
Levin, on walled gardens: "They can show attribution — doesn't mean it's incremental, doesn't mean it's accurate." Walled garden closed-loop attribution is characterized as systematically misleading for brand-building investment decisions.
Levin, on MQ governance: Media quality may ultimately need a formal governance layer — analogous to JIC (joint industry committee) or MRC (Media Rating Council, the body that accredits measurement vendors) — to standardize definitions and prevent gaming.
Levin, on the open web: Defines "open web" as everything outside walled gardens (platforms where only the corporate owner can buy their own inventory), explicitly including Netflix and CTV — arguing premium CTV is undervalued by buyers because attribution doesn't credit it fairly.
Levin, on AI search advertising: Estimates AI search ad revenue could reach ~13.6% of overall search spending by 2029 — "meaningful but still a fraction of search itself," and explicitly not the 10x search-scale revenue implied by some current AI company valuations.

Fact check

Claim (hosts, paraphrasing report): "80% of programmatic display spend now transacts through PMPs and programmatic guaranteed deals." Assessment: Unverified / context-dependent. This figure circulates in industry discussions and is plausible as a directional trend, but the precise percentage varies significantly by measurement methodology, market, and how "PMP" is defined. The hosts themselves note some of the CTV classification may be misleading. Treat as a rough directional data point, not a precise figure.

Claim (Brett House): "Reuters reported $13 billion in 2025 revenue for OpenAI subscriptions, expected to reach $280 billion by 2030." Assessment: Partially accurate but requires context. OpenAI's reported 2025 revenue projections (~$12–13 billion) have been widely cited, but the $280 billion figure by 2030 reflects OpenAI's own internal targets leaked to media — a highly optimistic internal projection, not an independent forecast. Presenting it as a straightforward Reuters-reported projection elides that it is a company aspiration, not an analyst consensus.

Claim (Levin): "Sound-on in-stream video scored 8.2/10; muted out-stream scored 1.3/10" on the report's quality index. Assessment: Unverified externally. These scores come from the CIMM report itself, which Levin co-authored. The methodology behind the 1–10 index is not detailed in the podcast. Listeners should consult the full report before relying on specific numeric scores in buying decisions.

Incentive flag — Levin generally: Levin is founder of Emmett Advisory, which consults advertisers, publishers, and technology companies on media quality strategy. The entire episode's argument — that media quality signals should be priced and prioritized — directly expands the market for the consulting and advisory services he provides. This doesn't make the underlying argument wrong (much of it aligns with established effectiveness research), but listeners should note that "media quality frameworks need governance and external expertise" is also a commercially convenient conclusion for an MQ-focused advisor.

No other claims clear the bar for false/misleading.

Why this matters for ad-tech operators

**DSP and SS

Full analysis

The pitch on this episode is that "media quality" — the environment an ad runs in, not just who sees it — has been systematically underpriced for two decades, and that the fix is to weight placement signals (time of day, format, device, geography) in buying algorithms instead of chasing an ever-weaker cookie. A former Google exec turned advisor is behind a CIMM report making the case.

What's actually being decided for the ecosystem: whether buyers start paying a premium for context and environment — and whether sellers of premium inventory can finally get credit for it. This is a Type 2 (easily reversible) question at the level of any single campaign, but a Type 1 (hard to reverse) question at the level of industry standards and governance. No hard forcing function; the CIMM report and the slow decay of identity signals are the pressure.

The Market Analyst

This is a repackaging of effectiveness research that Byron Sharp and Les Binet have preached for a decade, now aimed at programmatic plumbing. In plain terms: the industry keeps rediscovering that cheap reach isn't cheap. What's new is the attempt to turn "quality" into a priceable signal a machine can bid on. Follow the money: the verification vendors (DoubleVerify, IAS) get cast as blunt instruments, which threatens their premium narrative. Custom-bidding players like Chalice and DV360's own tooling are the beneficiaries. And the AI-search-ads skepticism — 13.6% of search spend by 2029, not 10x — is the sober take that undercuts the trillion-dollar valuations. That number, if it circulates, is a useful anchor against the hype.

For the non-specialist: the claim is that ads work partly by building a brand over years, and today's buying systems only reward the sale you can see this week — so they overpay for junk.

The Skeptic

The load-bearing assumption is that "media quality" can be defined precisely enough to price without being gamed the moment it's defined. That's a heroic bet. The instant an 8.2/10 score exists, every publisher optimizes to the score, not the underlying value — exactly what happened to viewability and fraud metrics. And note the incentive: the advisor concluding that "MQ needs governance and outside expertise" happens to sell MQ advisory. The scores (8.2 vs 1.3) come from his own unaudited report. This isn't wrong, but it's the same medicine the industry has swallowed three times — new metric, new committee, new gaming, repeat.

For the non-specialist: every time the ad world invents a quality score, sellers learn to hit the score instead of actually being good.

The Operator

Try to execute this Tuesday. A trader wants to weight bids by time of day and format before letting the DSP optimize. Where does the reliable, cross-publisher placement data come from? The episode itself flags that CTV inventory is mislabeled — 30% running overnight at flat CPMs, blank Thursday Night Football slots. You can't weight on signals you can't trust. The one place this gets real is where a seller actually holds proprietary placement data others can't replicate — a "moneyball" set of signals about how ads perform on a given page or player. That's a genuine edge, and it's exactly what a video-infrastructure company with player-on-page data would try to package. But at 90 days, the break point is attribution: buyers won't pay a context premium if their measurement still only credits the last click.

For the non-specialist: you can't charge more for a better ad slot until buyers have proof it worked, and today's measurement mostly doesn't provide that.

The CFO

The math the episode is really attacking is the agency incentive to hit the lowest CPM. That structure is sticky because low CPMs are easy to defend in a QBR and brand equity is not. For a premium publisher, the payoff of an MQ framework is obvious — it's a reason to raise prices. For an agency, it's a reason to spend more per impression, which nobody's client asked for. Someone has to eat the cost of proving context is worth it, and pilots always look good; scale is where the premium evaporates. The honest read: this pays back for sellers of genuinely scarce quality inventory and for measurement vendors selling incrementality. For everyone buying cheap reach at volume, it's a cost.

Where the council splits

Is quality priceable, or just gameable? The Operator sees a real edge for whoever holds proprietary placement data; the Skeptic says any published score becomes a target and decays. Both can be true — the edge lasts exactly until the signal is standardized.
Who pays the premium? The Market Analyst and CFO disagree on whether this is a buy-side or sell-side story. It's a sell-side story dressed as a buy-side reform. Premium publishers and CTV owners want context priced; agencies optimizing to low CPMs have no reason to volunteer.
Does the identity collapse force this? The real tailwind isn't the report — it's that cookies and device IDs keep weakening, so something has to fill the bidding signal vacuum. Context is the obvious candidate whether or not "MQ" ever gets a governance body.

What it hinges on

Three beliefs. First, that measurement moves from last-click attribution toward incrementality — without that, no context premium sticks. Second, that placement data becomes trustworthy enough to bid on, which the CTV mislabeling problem directly undercuts. Third, that identity signals decay fast enough to force buyers off the cookie and onto context.

The council leans skeptical that this becomes a priced, standardized signal soon — but agrees the underlying direction (context and environment mattering more as identity fades) is real and already showing up in custom-bidding adoption. The move to de-risk: any seller pursuing this should lead with proprietary, hard-to-replicate placement data tied to outcome proof, not join a race to publish another quality index.

Prediction: No CIMM/MRC-style governance body will launch an accredited, industry-adopted "media quality" scoring standard by the 2027 IAB Annual Leadership Meeting (Jan 2027); the conversation will still be at framework-and-whitepaper stage.

Confidence: Medium — standards bodies move slowly and vendor incentives conflict.

Why: The episode reveals this is currently a report and a set of proposed definitions, not a ratified standard — and it explicitly floats that governance "may ultimately" be needed, which is early-stage language, not a launch. Industry measurement standards (viewability, cross-media measurement via the JIC) have historically taken years and stalled on exactly the disagreement flagged here: whose definition wins and how to stop gaming. The competing incentives — verification vendors defending their turf, custom-bidding vendors wanting proprietary edges, publishers wanting to self-score — make fast consensus the less likely path. The opposite outcome (a real accredited standard in under 18 months) would require the fragmented sell side to agree to be measured by a common yardstick, which almost never happens quickly.

Revisit by 2027-01-31: We're right if there's still no accredited, cross-industry MQ scoring standard in market and the topic remains white papers and panels. We're wrong if CIMM, MRC, or a JIC launches an adopted MQ accreditation with named participating buyers and sellers.

Meanwhile the durable, un-hyped shift — context signals gaining weight in bidding as identity fades — will keep happening quietly through custom bidding, with or without a governance layer. That's the part operators can act on now; the standard is the part worth not waiting for.

Listen to the episode →