Nielsen changes TV currency today as ad-supported viewing falls to 71.5%

Counting is the quiet part of advertising. Nobody buys a rating, a download or a subscriber; they buy a number that stands in for one, produced by an organisation with its own methods, its own commercial interests and its own release schedule. When that number moves, the market moves with it, whether or not anything in the world has changed.

Four days in late August put an unusual amount of pressure on that arrangement. Nielsen deploys seven separate changes to the currency underpinning United States television trading today. A quarterly report published the week before showed the advertising-funded share of American viewing shrinking. A panel operator now owned by a demand-side platform put a number on how little of that viewing is actually spent looking at advertisements. A podcast company sold exclusive video rights to a streaming service without publishing a measurement rule. A research supplier holding 160 petabytes of purchase data announced an arrangement with OpenAI whose most specific paragraph describes what the deal does not do. And a marketplace covering 3,000 newsletters published rate benchmarks showing that the smallest lists command the highest prices per reader.

The thread running through all of it is the same: who does the counting, what the count leaves out, and what happens to the money in between.

Seven changes land in the currency on the same morning

Nielsen deploys seven methodology changes to its national Big Data plus Panel service today, August 31, 2026. PPC Land reported the schedule when it was set, listing co-viewing capture through wrist-worn wearables, a latency-adjusted DASH universe estimate, a household demographic assignment model, integrated weighting, Hispanic universe estimation, an ACR monitored tuning adjustment and householding for one big data provider. Nielsen's own framing was cautious. No ratings increase is guaranteed.

Seven simultaneous changes to a trading currency is not a routine software update. Each one alters a different layer of the calculation: who is counted as present, how many households the sample represents, which demographics get assigned to which set, how the weights are applied, and how one supplier's duplicate households are collapsed. Any single change would produce a measurable shift in delivered impressions. Applied together on the same date, they make it impossible to attribute a movement in August ratings to any one of them, and PPC Land noted that the combination forces buyers to re-check August delivery rather than assume continuity with July.

The changes arrive against a backdrop of contraction in the thing being measured. Nielsen released its second-quarter Ad-Supported Gauge on August 25, and the topline is that advertising-funded television accounted for 71.5 percent of total United States viewing in the April to June quarter, down 1.3 points from 72.8 percent in the first quarter and down from 74.2 percent in the fourth quarter of 2025.

Inside that shrinking bloc, streaming took 48.2 percent, up 1.6 points quarter on quarter. Broadcast took 26.6 percent, down 1.6 points on the quarter but up 0.6 points on the year. Cable held at exactly 25.2 percent, unchanged from the first quarter.

Two features of that release deserve more attention than the headline share.

The first is the population base. Those figures are calculated on persons aged two and over. Switch to adults 18 and over, the base against which most guarantees are actually written, and streaming falls to 44.4 percent, broadcast rises to 28.6 percent and cable rises to 27.0 percent. The gap between streaming and broadcast narrows from 21.6 points to 15.8 points, and cable lands within 1.6 points of broadcast. Same quarter, same data, two different markets depending on which line of the report is read.

The second is the denominator. A category taking a larger share of a contracting bloc is not the same thing as a category growing. Streaming gained 1.6 points of a pool that lost 1.3 points of the total. Both statements are true and they describe different phenomena, but they are published in the same table and quoted in the same sentences.

The quarter itself was unusually distorted by sport. The FIFA World Cup opened on June 11, three weeks before the quarter closed, with FOX holding English-language rights to 68 of the 104 matches and Telemundo carrying Spanish coverage. Preliminary viewership reached 24.4 million on the FOX Television Network for a single knockout match, and the tournament generated more than 25 billion advertising impressions across FOX, FS1, Telemundo and Universo. The NBA Finals ran five games on ABC domestically, with Prime Video distributing to 17 international markets. In the June monthly Gauge, broadcast reached 19.8 percent, its first June share increase since the Gauge launched in 2021, on a 118 percent month-over-month rise in broadcast sports viewing.

Peacock recorded its first profitable quarter in the same window, with 189 million dollars in adjusted EBITDA, and PPC Land reported that advertising revenue per subscriber rose 43 percent to 4.98 dollars a month while Domestic Media advertising revenue rose 55 percent to 2.16 billion dollars. Comcast attributed the performance to the World Cup, the NBA playoffs and Love Island USA arriving simultaneously, and noted that none of them recur in the same form in the third quarter.

There is a further caveat buried in the Gauge release, and it matters for anyone planning against these figures. The second-quarter Ad-Supported Gauge has not yet migrated to Advertising Research Foundation DASH-based universe estimates. That migration is planned for autumn 2026, and it will produce results that are not comparable with the current production data. A planner reading the second-quarter report and the eventual fourth-quarter report side by side will be comparing two different measurement regimes.

Then there is the lag. The second-quarter Gauge covers April to June and was published on August 25, roughly eight weeks after the quarter closed. The June monthly Gauge appeared on August 18; the May edition appeared on July 28. Upfront negotiations largely conclude inside that interval. By the time the quarterly number is public, the commitments it might have informed have been made.

Nine minutes out of two hundred

Six days before the currency changes deploy, a different measurement company published a figure that sits underneath all of them.

TVision announced Ad Scoreboard 2.0 on August 25, and the arithmetic in the announcement is stark. Of more than 200 minutes of daily video consumption, roughly 24 minutes represent advertising opportunity, and about nine minutes qualify as attentive advertising time. That works out at approximately one attentive advertising minute for every 22 minutes of viewing.

Yan Liu, TVision's chief executive, described the product's purpose bluntly: "TVision Ad Scoreboard 2.0 shows you what most tv measurement doesn't: whether real people actually watched your ads."

The methodology is observational rather than modelled. TVision operates a panel of 5,000 United States households fitted with computer vision sensors, automatic content recognition and digital meters, capturing four variables simultaneously on a per-second basis: whether the screen is showing programming or an advertisement, how it is being delivered, which individuals are in the room, and what their attention level is. Three separate scoreboards break the output into brand performance, pre-purchase network and application planning, and second-by-second creative analysis. The central metric is viewers per viewing household, which counts people in the room rather than devices receiving impressions.

That metric is the collision point. Co-viewing is now being counted three different ways, by three organisations, in the same households, using three incompatible methods. Nielsen adds wrist-worn wearables to the national currency today. Google applied a Total Co-view definition to seven reach metrics on June 2, 2026, expanding reported unique users without changing delivery. TVision uses computer vision. A buyer running linear, YouTube and open connected television in the same campaign now receives three co-viewing figures that cannot be reconciled with one another, and each of the three is treated as authoritative inside its own environment.

The comparative numbers show why the discrepancy is commercially significant. Research from the Video Advertising Bureau's impression gap study put co-viewing at 60 percent on premium connected television platforms against 45 percent on YouTube. That is a 15-point difference in how many people a single impression is credited with reaching, applied across billions of impressions.

TVision's own independence is now a live question rather than a settled one. Viant closed its acquisition of the company on May 5, 2026 for 40 million dollars, 22.5 million in cash and 17.5 million in stock. Larry Madden, Viant's chief financial officer, told the August 10 earnings call that integration of the eyes-on-screen attention data was pacing ahead of initial expectations across the technology stack, and Viant reported second-quarter revenue of 104.3 million dollars, up 34 percent, against a net loss of 1.8 million. A measurement panel owned by a demand-side platform supplies data that the same platform's buyers use to justify placements, which is a structure the industry has historically objected to elsewhere.

The announcement also left the usual gaps. No measurement window was specified. No demographic breakdown of the sample was published. It is not stated whether the nine-minute figure derives from TVision's own panel or from a third party. No Media Rating Council accreditation was mentioned, and no pricing, availability date or client roster was disclosed. Even the product name is rendered inconsistently across the materials.

Attention measurement has been consolidating around a handful of vendors regardless. Lumen Research runs predictive biometric models at impression scale and brought its methodology to Netflix advertising across five European markets in March. Adelaide derives an attention currency from placement characteristics and integrated with Amazon DSP in June and DV360 custom bidding in late July. Teads combined Adelaide, Lumen and TVision into a single layer with CTV Ensemble in June. Competing attention claims have accumulated alongside: InMobi reported 3.5 times more attentive seconds than YouTube equivalents in July, and Kargo claimed 78 percent better viewer retention than industry benchmarks a year earlier.

The picture that emerges from the two releases together is a market spending most of its money against a currency being rewritten today, while a second measurement layer suggests that fewer than five percent of daily viewing minutes are spent attentively watching the advertisements that fund it.

A podcast crosses into streaming without a measurement rule

If television measurement is contested, streamed podcast video does not yet have a contest, because it does not yet have a rulebook.

Netflix and iHeartMedia announced on August 25 that two Martha Stewart programmes move to Netflix as video exclusives. PPC Land reported that "Martha Cooks" launched the same day and "The Martha Stewart Podcast" followed on August 26 with Alexis Ohanian as the opening guest. Audio versions continue on iHeartRadio and open podcast applications. The video exclusivity covers all new episodes plus selected library episodes.

Martha Stewart framed the move as a format change rather than a distribution change: "I'm delighted to bring my podcast to Netflix, and with that, introduce the medium of video to the series." Will Pearson, president of iHeartPodcasts, emphasised reach.

What was not disclosed is longer than what was. No CPM. No named sales owner. No advertising format specification. No measurement methodology. No financial terms. No detail on programmatic availability.

Those omissions are not cosmetic, because episodes distributed inside Netflix fall outside the IAB Tech Lab's version 2.3 measurement framework, which covers open RSS feed delivery. Whether a Netflix view registers in podcast metrics, in streaming metrics, or in both, is unresolved. The Tech Lab has been working the other direction, pulling video podcasts under the same download counting rules that govern audio, but a proprietary streaming environment is not a feed and does not report like one.

The budget question follows directly. United States podcast advertising revenue reached 2.9 billion dollars in 2025, up 17.6 percent. Digital video budgets run to roughly 78 billion. A show that moves from one to the other has not changed its content, its host or its audience, but it has changed which planner signs for it and which benchmark that planner is measured against.

Duplication compounds the problem. The same show now reaches audiences through two channels that are separately priced, separately sold and separately measured. Bob Pittman, iHeartMedia's chairman and chief executive, put audience overlap at probably less than five percent, without describing how a buyer would verify it.

Then there is the permanence issue. Baked-in advertisements, recorded into the episode file itself rather than inserted dynamically, travel with library content wherever it goes. A sponsor who bought a host read for an open-feed audience in 2024 now receives impressions inside Netflix that were never sold, never priced and never measured for that environment.

Supply conditions do not favour sellers either. Research from MillionPodcasts found that 77 percent of 34,208 video podcasts carry no advertising at all, a large body of unsold inventory competing for the same attention.

iHeartMedia is running the strategy in parallel across several platforms. Second-quarter revenue reached 977.2 million dollars, with podcast revenue up 20.7 percent to 162.1 million. The company recorded 150.3 million United States streams and downloads in June across 945 active shows reaching 29.5 million unique monthly listeners. Six iHeartMedia shows arrived on Disney+ and Hulu from August 14. Netflix, for its part, reported 250 million monthly active viewers on its advertising plan at its 2026 upfront and is tracking toward roughly 3 billion dollars in advertising revenue this year.

A publisher taking the opposite approach appeared in Digiday on August 31. Sara Guaglione reported that The Telegraph is expanding its podcast operation into video, events and paid memberships rather than licensing exclusivity to a platform. Its podcast audience grew 161 percent year on year. Three daily shows average 2.4 million weekly views on YouTube. "Ukraine: The Latest" has passed 75 million views globally since launch and grew 115 percent on YouTube in the past three months. "Iran: The Latest" records more than 350,000 weekly plays and more than 60,000 daily downloads. Homepage video views tripled between the first and second quarters. The audio team has doubled to more than 40 staff, five regular video podcasts are running, and four more are planned by year end, including true crime and sport.

Matthew Bayley, executive editor of audio and visuals, described the underlying shift: "Video is now becoming the primary method by which news and journalism is consumed."

Revenue is explicitly not the current metric. The Telegraph, acquired by Axel Springer in June 2026, is measuring listeners, viewers, watch time and engagement, with separate podcast subscriptions and YouTube membership programmes planned but not scheduled. The Economist launched a comparable offering, Economist Play, earlier in the summer.

Two models, then. One sells exclusivity into a measured-by-somebody-else environment and takes the cheque. The other keeps the audience relationship and defers the revenue question. Both are bets on where podcast video ends up being counted.

A data supplier draws a line around what the model may see

NIQ announced an arrangement with OpenAI on August 25, and the most precise language in it concerns exclusion.

PPC Land reported that the company's proprietary data remains governed and available only through authorised NIQ products, client agreements, permissions and applicable security controls. The collaboration neither exposes that data to the public through ChatGPT nor changes existing access requirements.

The asset being fenced is substantial. NIQ operates a 160-petabyte data engine spanning roughly 260 million product items and 10.5 billion product attributes, ingesting 4.3 trillion consumer transaction records weekly, a figure up 23 percent year on year. Operations cover more than 90 countries, approximately 82 percent of the world's population and 7.4 trillion dollars in consumer spending.

Three products are covered. Optiq Chat is an insights agent answering business questions grounded in retail measurement and consumer panel data. Optiq Bridge is described as a governed route for organisations to integrate NIQ data, models and generative AI into their own platforms. Optiq Mobile is already live on the Apple App Store in the United States. Expanded versions of Chat and Bridge are due in early September 2026.

Troy Treangen, chief product and AI officer, said the arrangement aims to help clients get more value from NIQ intelligence by moving it into more applications, systems and workflows. Irina Stoian joined as chief AI commercial officer from Palantir in July 2026, and the ConnectAI Charter Program embeds NIQ engineers and data scientists inside client organisations, a forward-deployed model recognisable from her previous employer.

The omissions are as instructive as the disclosures. No financial terms were published, covering equity, revenue share, contract length or client commitments. No launch customers were named, though the Charter Program has five signed participants of which only Purina is public. No pricing model was specified for machine access to retail measurement data, and Jim Peck, executive chairman and chief executive, has indicated experimentation with consumption-based and bundled models without settling on one.

One technical detail was conspicuously absent. Treangen confirmed on the August 10 earnings call that Optiq Bridge is the company's product for Model Context Protocol consumption. The August 25 announcement used vendor-neutral language instead, referring to applications, systems and workflows.

The infrastructure underneath has been assembled over two years. The Discover platform launched in July 2024, consolidating consumer panel and retail measurement data. A global data clean room on Snowflake followed in October 2025. GeoPurchase purchase-based audiences operate in 11 countries, and programmatic distribution runs through Adsquare. Commercially the direction has been consistent: AI-native revenue grew 34 percent in the second quarter, with 51 percent of the top 100 clients using AI-native solutions and adoption up 64 percent year on year.

Four questions remain open in the arrangement as published: what an agent is permitted to retrieve, whether retrieved data may be cached in a client environment, how audit trails work when the requester is software rather than a person, and what happens if licensed measurement surfaces in a downstream artefact.

The contrast with what people volunteer is sharp. A DuckDuckGo-commissioned survey of 1,944 United States adults, fielded June 17 to 27, 2026 through the PureSpectrum panel with census-balanced quotas and a margin of error of 2.2 points, found that 56 percent of self-described AI enthusiasts disclosed something to a chatbot that they withheld from close friends, parents, colleagues, doctors or therapists. Across all AI users the figure was 32 percent. Among parents using AI it reached 43 percent, against 25 percent of non-parents.

Awareness lagged behind behaviour. Fifty-three percent were unaware or unsure that conversations train AI systems by default. Twenty-five percent knew that chatbot conversations can be subpoenaed. Forty-three percent knew none of six basic facts about how chats are stored, reviewed and disclosed. After being informed, 58 percent were uncomfortable with conversations being used for training, 30 percent very uncomfortable. Fourteen percent said they mostly or completely trust large AI companies with their data; 39 percent reported no trust at all, and 48 percent reported no trust in the United States government on data protection.

The survey's provenance is worth stating. DuckDuckGo commissioned it and sells a competing product. Results are unweighted, and the report itself flags inconsistencies in how two of its disclosure figures were defined between summary and body.

Set the two announcements beside each other and the asymmetry is the point. A commercial data holder with 4.3 trillion weekly transaction records negotiates contractual limits on what a model may see, publishes the boundary, and keeps the corpus behind authorised products. Individual users supply a comparable volume of far more sensitive material with no contract, no permission layer and, for roughly half of them, no accurate understanding of where it goes. Research documented in May found conversation text reaching advertising networks across 20 chatbot services, and a federal judge in November 2025 required production of 20 million de-identified ChatGPT logs in litigation.

The smallest lists carry the highest prices

Newsletter advertising prices the market in the opposite direction to television, and the Paved benchmarks published this quarter show how steeply.

PPC Land reported that human resources newsletters command 18.47 dollars per thousand subscribers while lifestyle lists fetch 1.22 dollars, a spread of roughly 15 times. The methodology draws on twelve months of live marketplace data across more than 3,000 publishers and 253 million subscribers, with categories included only where at least twenty campaigns completed.

Crucially, Paved reports price per thousand subscribers rather than a conventional CPM based on delivered impressions, which is a different unit and worth holding in mind when comparing against display or video rates.

The full ranking runs from human resources at 18.47 dollars, marketing at 13.88 and startups at 13.28, through education at 7.21, finance at 7.11, self-improvement at 5.73, health and wellness at 5.60 and business at 5.34, down to technology at 4.76, family at 4.62, pets at 4.17, travel at 3.94, news at 2.47, entertainment at 2.28 and lifestyle at 1.22.

Set the prices against the list sizes and the relationship inverts cleanly. Human resources, the most expensive category, has an average list of 24,439 subscribers. Lifestyle, the cheapest, averages 766,110. Marketing sits at 57,780, startups at 80,836, technology at 166,912, family at 243,868, news at 555,418 and entertainment at 560,778. The correlation is close to monotonic across the table: the larger the audience, the less each thousand readers is worth.

This is scarcity pricing applied to identity. A newsletter with 24,000 human resources subscribers is not selling reach; it is selling a list on which nearly everyone holds a budget for the products being advertised. A lifestyle list with three quarters of a million subscribers is selling reach and little else, and reach is the commodity in most abundant supply anywhere in digital media.

The engagement figures in the same dataset are less coherent, and buyers reading the report should notice. News lists reported a 31.23 percent open rate. Health and wellness reported 15.22 percent and technology 10.73 percent. But marketing showed a 2.49 percent open rate against a 1.24 percent click rate, education showed a 2.57 percent open rate against a 5.21 percent click rate, which is arithmetically impossible under a conventional definition, and pets reported a 0.10 percent open rate with clicks rounding to nothing. Those figures are drawn from marketplace reporting rather than a controlled panel, and the inconsistencies point to publishers measuring the same events by different rules, which is the same structural problem visible in television, only with less money attached and no accreditation body arguing about it.

Open-rate measurement has been unreliable since Apple introduced Mail Privacy Protection in 2021, pre-fetching tracking pixels regardless of whether a message is read. A category reporting 31 percent and a category reporting 0.1 percent are not necessarily describing different reader behaviour; they may be describing different proportions of Apple Mail users and different pixel implementations.

The category has been attracting infrastructure investment through the same period. Google Ad Manager added support for advertisements in email newsletters, and beehiiv doubled its advertising sales team as monetisation picked up. The Paved report circulated in July and was re-promoted on August 27.

A tabletop game shop bills 4,300 euros to recycle 130 euros of cardboard

The last story concerns a cost that never appears in a media plan and determines what can be advertised anyway.

Kevin Toet runs The Role Play Goblin, a tabletop game shop in Den Haag. On August 13, 2026 he filed evidence with a European Commission consultation drawn from his own transaction records, and PPC Land reported on August 30 that annual packaging compliance costs across nine European Union markets run between 2,400 and 4,300 euros against roughly 130 euros of actual recycling fees, a multiple of 18 to 33 times the underlying environmental obligation.

The volumes are small enough to check by hand. Over eight and a half months the shop shipped 98 parcels to nine costed countries, generating 10,861.54 euros in cross-border revenue. The compliance bill would consume between a fifth and two fifths of that.

Individual markets are worse than the average. Belgium took 30 parcels; registration and representation there costs 350 to 1,050 euros a year against 2 euros of eco-fees, a multiple between 175 and 525. Spain took 14 parcels at roughly 302 euros against 3 euros of fees. Austria took exactly one parcel, worth 82.32 euros in revenue, which triggers an authorised representative fee of 300 to 500 euros plus a licensing minimum of 90 to 150 euros.

The structural problem is that authorised representative fees make up 70 to 90 percent of the burden and are entirely insensitive to volume. A shop shipping one parcel pays what a shop shipping ten thousand pays.

Regulation (EU) 2025/40 applied in full across all 27 member states from August 12, 2026, requiring cross-border sellers to register with national packaging schemes in each destination country and appoint authorised representatives. PPC Land had reported earlier that the law obliges sellers to register in up to 27 countries and that sellers were already withdrawing from markets rather than comply.

Exemptions exist but are inconsistent. Ireland requires no registration below 10 tonnes of packaging or one million euros of Irish turnover; the shop sits roughly 666 times under that threshold and pays nothing. Finland permits direct registration without a mandatory representative. The Netherlands sets a domestic threshold of 50,000 kilogrammes annually, meaning a Dutch seller shipping the identical product to a Dutch customer has no obligation at all. Crossing a border triggers the full regime.

The legislative history explains why the relief has not arrived. The Commission proposed suspending the representative requirement until January 1, 2035 in December 2025. Parliament draft reports in May 2026 proposed limiting exemptions to firms under 49 employees and 10 million euros of turnover. On June 24, 2026 the Council discontinued negotiations on the suspension proposals, citing member state reservations.

Toet's four requests are procedural rather than ideological: a consistent de minimis threshold across all member states of the kind Ireland and Finland already operate, recognition of home-state registration for small cross-border volumes, removal or capping of the mandatory representative requirement for micro-enterprises, and a single European registration point in place of 27 national portals.

The advertising consequence is direct, and it is the reason this belongs in a marketing newsletter rather than a compliance one. Shipping eligibility and advertising availability are the same variable inside a product feed. A seller who disables a destination country to avoid registration does not pause a campaign; the products simply stop being eligible there, and the effect surfaces in Google Merchant Center as disapprovals and coverage loss rather than as a deliberate budget decision.

There is a recent precedent for how fast this moves. The European Union removed its 150-euro duty exemption for low-value imports on July 1, 2026, replacing it with a flat 3-euro charge. PPC Land reported that Temu cut its Google Shopping spend within a week and Shein approached full exit by early August, and that Pinterest's European advertising growth slowed to 12 percent as regulatory pressure landed on Asian sellers. The mechanics here run the other way. Customs charges scale with volume and can be absorbed by large operators; per-market registration is fixed and prices out the smallest sellers first.

Toet is careful about the limits of his own evidence, which is what makes it useful. The cost figures are researched estimates of published rates rather than invoices, since he has not registered anywhere. Shipped product weight of 374 kilogrammes differs from the packaging weight actually assessed. Of 866 products, 328 lack recorded weights and were estimated. The window is eight and a half months rather than a full year, and it covers nine of the 21 destination countries reached. He also sells EPR Tracker software at 100 euros a year, a commercial interest the filing does not hide.

What the filing amounts to is an audit trail rather than a lobbying position, and it points at a cost curve running away from the smallest advertisers in the European market. Those advertisers are the base of the self-serve layer that every major platform has spent three years building products for.

Also noted

• August 30: Version 4.0.0 of the Google Ads API Developer Assistant, released August 25 by Bob Hancock of the Google Ads API team, is explicitly incompatible with all prior installations, converting the tool from a local workspace project into a globally installed plugin bundle with ten slash commands, cached API version locking and strict linting of generated Python. PPC Land
• August 30: WhatsApp replaced the six-digit numeric PIN used for two-step verification with a full alphanumeric password supporting special characters, and now permits multiple passkeys per account, with more than one billion accounts having set up a passkey; no rollout dates or migration details were published. PPC Land
• August 30: Escrow.com's second IPv4 Investment Index, published August 27 and covering the quarter ended June 30, recorded 107 transactions worth 9.37 million dollars, with average deal value down 6.6 percent year on year, large-block pricing at 24.67 dollars per address and a blended rate of 21.35 dollars. PPC Land
• August 31: Target Beauty Studio, launching September 10 across more than 600 stores with 1,600 products from 90 brands, includes two Black-owned brands, Briogeo and Glamazon Beauty, or roughly 2 percent of the curated selection, prompting consultants to contrast the retailer with Ulta Beauty and Sephora on the Fifteen Percent Pledge. Digiday
• August 30: Beth Galetti, Amazon's senior vice president of people experience and technology, published an account of writing more than 100,000 lines of code with AI tools after a 25-year gap, alongside a 2.5 billion dollar Future Ready 2030 skills initiative, ten months after she communicated roughly 30,000 corporate job cuts. PPC Land