00The week in one paragraph
This was the week the numbers stopped adding up. Last week the industry was busy buying the pipes; this week we found out what’s actually flowing through them — and how little of it anyone can measure. The supply itself is contracting: Ozone’s data shows publisher ad requests down 32–41% year over year in Q2 and programmatic spend down 30.6% across the US and UK, as AI search intercepts the user before they ever reach your page. The money that remains is being spent into instruments buyers openly distrust — the IAB pegs US digital video at $80 billion while 67% of buyers say they lack confidence in open-exchange inventory. The measurement layer is buckling in parallel: Nielsen ONE and VideoAmp both withdrew from MRC accreditation, only 23% of residential IPs reach their intended geography, match rates have fallen below 10% for some clients, and the MMM tools filling the vacuum — Google’s Meridian and Meta’s Robyn — are authored by the very platforms whose inventory they score. Meanwhile the giants kept scaling on their own terms: Netflix’s ad business is on track to roughly double to $3B even as it cuts its public engagement reporting to once a year, and Stagwell built its own AI curation marketplace to decide which publishers get spend. But the most important thread is what publishers did about it. SPUR’s publisher-run content telemetry standard — BBC, FT, Guardian, Sky, The Times, now the AP, 30 members deep — is open for public comment through July 24, the SAIL ledger launched to track AI usage against IAB Tech Lab’s pay-per-query framework, Axios, Forbes, Future, Time and The Washington Post started selling AI visibility as inventory, book publishers sued Google over Gemini, the EU ordered Google to share its Search data and open Android to rival AI, and Reddit said out loud what everyone is thinking — that the AI licensing cheque might be cannibalizing the ad business it was meant to subsidize. One thread ties it together: when the industry’s own numbers stop being credible, the publishers who win are the ones who build and control the meter.
01Publisher ad supply fell up to 40% in Q2 — the traffic collapse has reached the P&L
Ozone data covering ~20 billion impressions across its premium network (The Guardian, News UK, WSJ) shows publisher ad requests fell 32–37% YoY in the US and 39–41% in the UK in Q2 2026. Combined US/UK programmatic spend dropped 30.6% YoY in H1 — US spend down 44%, the UK down only 14.3%, where higher yields cushioned the fall. UK eCPMs rose ~30% YoY in June against ~7% in the US, so scarcity is partly repricing what’s left. Apps were the only growth channel (US spend +23%, eCPMs +42%); outstream video spend fell 76%. Ozone COO Danny Spears attributed it to platforms, particularly Google, “intervening in the user journey and providing content in situ.”
Publisher takeaway
This is the hardest number yet connecting the referral-traffic collapse to actual revenue — it’s no longer a traffic story, it’s a supply story, and supply is the thing you sell. Two responses follow directly from the data. First, the UK/US divergence proves yield management partially offsets volume loss: if your impressions are falling and your eCPMs aren’t rising, you’re absorbing the whole shock instead of half of it. Second, apps grew while outstream collapsed — owned destinations where the platform can’t intercept the journey are now materially outperforming rented ones. Model your business at 60% of last year’s request volume and see what breaks. Push hard on registered, direct, and app-based audience. (Ozone is a single premium, UK-weighted network — directional for the open web, not a market-wide census.)
02Digital video hit $80B — and buyers don’t trust the inventory
The IAB’s 2026 Digital Video Spend & Strategy report puts US digital video ad spend at $80 billion this year, while buyer confidence deteriorates across every buying method: 43% have somewhat-to-no confidence in direct I/O, programmatic guaranteed and self-serve; 55% lack confidence in private marketplaces; 67% lack confidence in open exchange/RTB. IAB VP Chris Bruderle said “buyer trust is being eroded on two fronts” — bad actors injecting invalid inventory, and uncertainty about where legitimate inventory actually originates and runs. Dept’s Lyndsey Garza: “CTV has matured into a premium channel with premium price tags, and when you’re paying TV-sized CPMs, you expect TV-sized transparency.” Agencies including Kepler Group and Crispin are shifting toward direct SSP relationships and curated deals.
Publisher takeaway
Read the confidence gradient as a price list — trust falls as you move from direct to PMP to open exchange, and so does your yield. That 67% open-exchange figure is the single best argument for investing in direct and curated paths rather than dumping supply into RTB and hoping. The corollary is that provenance is now a product feature: buyers are paying premium CPMs and demanding they be able to see exactly whose inventory they bought and where it ran. Publishers who can prove supply origin — clean schain, minimal reseller hops, verifiable player and placement data — are selling something 67% of the market says it can’t currently get. Audit how many hops sit between you and the buyer, kill the redundant ones, and lead your sales conversations with transparency rather than treating it as a compliance afterthought. (See #3 and #4 — the trust problem has a measurement problem underneath it.)
03Nielsen ONE and VideoAmp withdraw from MRC accreditation
Both Nielsen and VideoAmp withdrew their MRC accreditation applications, per the MRC’s quarterly update. Nielsen cited methodology changes to Nielsen ONE Ads, its product measuring linear and streaming TV ad performance; VideoAmp said it plans to reassess by next year. Both retain Joint Industry Committee certification, and Comscore and iSpot maintain MRC credentials in separate categories. The upshot: the two most prominent cross-platform TV currency contenders now operate without MRC sign-off.
Publisher takeaway
Currency accreditation is the referee in every CPM negotiation you have, and the referee just left the field for the two biggest players. Expect more disputes, not fewer — when buyers and sellers disagree on delivery, the absence of an accredited third party means the argument gets settled by leverage rather than by standard, and the party with less leverage is usually the publisher. Practically: find out which currency your key buyers are transacting on and whether their confidence in it changed this week, get your make-good and discrepancy language reviewed now rather than mid-dispute, and keep at least one independently-verified measurement source in your own stack so you’re never arguing from the buyer’s numbers alone. JIC certification still stands and is worth citing — but it isn’t MRC, and sophisticated buyers know the difference. (Secondary-sourced via MediaPost/MRC quarterly update — confirm against the MRC’s own release.)
04Identity is failing: IP targeting misses 3 times in 4, match rates fall below 10%
Adstra research found only 23% of residential IP addresses reached their intended geographic target. Device-level CTV identifiers proved 24% more consistent than IP over time (71% vs 57% persistence). A separate Truthset study found nearly 40% of open-auction CTV media spend is wasted on inaccurate identity data. Adstra CDO Andy Johnson: “Identity sellers often win tests based upon either match rate or scale. But the matching that is performed is opaque to the customer.” Adstra introduced ID Connection Strength (ICS), scoring identifiers on recency, frequency, signal integrity, congruency and cardinality, widely available in Q3. Separately, a white paper from Idea Peddler with PJX Media, Share Local Media and Locality — “The Attribution Illusion” — reports match rates degraded below double digits, under 10% for at least one client. Idea Peddler CEO Cimin Ahmadi Cohen: “we do very, very little pure programmatic display anymore because it’s just deeply ineffective.”
Publisher takeaway
The addressability premium is being exposed as partly fictional, and that cuts both ways for you. The bad news: if buyers conclude that ID-based targeting doesn’t work, the segment-based CPMs layered on top of your inventory deflate. The good news is bigger — everything Cohen recommends as the replacement (premium CTV, contextual targeting, brand-lift studies, MMMs) is a category where a real publisher with a distinctive audience and a coherent content environment beats an anonymous ID graph. This is the moment to make context and logged-in first-party data your pitch rather than third-party segments you resell at a thin margin. Audit which of your revenue depends on identity resolution you can’t actually verify, and start converting those conversations to contextual and outcome-based ones before your buyers do it for you. (Adstra and Truthset are identity vendors; the white paper is agency-authored — vendor-supplied research, directional not audited.)
05Netflix’s ad business doubles to ~$3B — and its public numbers get thinner
Netflix reported Q2 2026 revenue of $12.56B, up ~13% YoY, just under the ~$12.59B consensus, with net income of $3.4B ($0.80/share). Q3 guidance of $12.86B missed estimates and full-year guidance narrowed to $51.0–51.4B; shares fell as much as 8–9% after hours to an 18-month low, having already dropped 31% in the three months after Q1. The ads business remains on track for ~$3B in 2026 — roughly doubling YoY — with programmatic access expanded to pause ads and live inventory plus AI-powered planning and buying. Members watched 97 billion hours in H1, up 2%. US upfront talks are in “advanced stages,” with interest in the 2027 FIFA Women’s World Cup, an expanded NFL slate, WWE and MLB. Netflix will cut its biannual “What We Watched” engagement report to once a year from 2027. Separately, it’s exploring continuous linear-style channels for subscribers, reportedly licensing inexpensive short-form content from publishers including BuzzFeed and Condé Nast.
Publisher takeaway
Three separate signals, all worth acting on. A $3B ad business doubling in a year is premium video budget leaving the open web permanently — and pause ads and live inventory going programmatic means Netflix is now competing for the exact automated dollars your video inventory chases. Cutting engagement reporting to annual while charging TV-sized CPMs is the same transparency contradiction buyers flagged in #2, and it’s an opening: sell against it with granular, verifiable delivery data. But the licensing note is the real opportunity — a major streamer shopping for cheap short-form publisher content is a distribution and revenue channel that didn’t exist for most publishers a year ago. If you have a video library sitting idle, its streaming licensing value just went up. (Softening subscriber-growth and guidance figures are as reported; the licensing talks are WSJ-sourced and not confirmed by the parties.)
06Publishers build the meter: SPUR’s telemetry standard and the SAIL ledger
Two publisher-side infrastructure moves landed in the same week. SPUR (Standards for Publisher Usage Rights) is a publisher-led coalition building a telemetry standard to make AI content usage measurable and licensable — founding members the BBC, Financial Times, The Guardian, Sky, The Times of London and MediaHaus, with the Associated Press now joined, spanning 30 publisher members and six affiliates. The standard tracks five events: content retrieved, grounded, cited, displayed and engaged. The spec was announced June 12 and public comment is open through July 24, 2026. MediaHaus CEO Gert Ysebaert called AP’s participation a “milestone.” Separately, Next Net and Sundial Media & Technology Group (Essence, Refinery29, Afropunk) launched SAIL — the Standardized Agentic Intelligence Ledger — tracking how AI systems use publisher content and blend it with other sources, built to be compatible with IAB Tech Lab’s CoMP AI framework and its pay-per-query model. Next Net has been “vectorizing” Sundial’s catalog since January on NVIDIA infrastructure; CEO Franklin Rios takes a negotiated percentage of revenue returned to publishers. Sundial CEO Kirk McDonald frames it as protecting publisher authority over editorial judgment.
Publisher takeaway
You cannot license what you cannot count, and this is the week the counting infrastructure became real. Every AI licensing negotiation to date has been a lump-sum guess because no publisher could measure usage — a telemetry standard converts that into metered, auditable, per-use commerce, which is how every other functioning media market works. The July 24 comment window is the single most actionable item in this issue, and it costs you an afternoon: the five tracked events (retrieved, grounded, cited, displayed, engaged) will determine what you can invoice for, and “grounded” versus “cited” is the difference between getting paid when a model uses you invisibly and only getting paid when it links you. Read the spec, submit comments before Friday, and evaluate SAIL-style ledgers as the accounting layer. The open obstacle is real — none of this bills anyone until OpenAI and Google implement it — but a standard with the BBC, FT, Guardian and AP behind it is the strongest demand-side leverage publishers have assembled yet. (See #8: litigation and metering are two routes to the same destination.)
07AI visibility becomes publishers' newest sellable currency
Axios, Forbes, Future, Time and The Washington Post are packaging AI-visibility metrics into products sold to brands that want discoverability inside LLMs and answer engines — building listicles and sponsored placements engineered to surface in AI answers. Forbes’ Nina Gould: “Publishers aren’t just selling impressions anymore. They’re selling visibility within the AI knowledge ecosystem.” But there is no standardized measurement; Time COO Mark Howard noted analytics firms “all have different methodologies. Their numbers are all different.” Cited context: 56.4% of news publishers block at least one AI crawler, 187% growth in AI-driven traffic during 2025, and Time ranking in the 98th percentile for AI bot activity among 7,000 publishers in TollBit’s network. Insiders warn some GEO vendors are closer to “snakeoil salesmen” than credible partners.
Publisher takeaway
This is the most encouraging story of the week — a genuinely new inventory type, being sold now by named publishers rather than piloted. It also reframes the crawler question: pure blocking forgoes a revenue line, and the publishers monetizing AI visibility are the ones AI systems can still read. But sell it carefully. Without standards, “AI visibility” is a metric you and your buyer are defining together, and an unstandardized metric sold to brands is exactly the setup that produced viewability fraud a decade ago — the difference between building a category and burning it is whether you publish your methodology. Define your measurement transparently, price it as an experiment rather than a guaranteed outcome, and vet GEO vendors hard. The publishers who establish credible AI-visibility measurement first will set the terms for everyone. (Note the tension with #6: metering AI usage and selling AI visibility are complementary, but both need the standards neither yet has.)
08Book publishers sue Google over Gemini training
Hachette Book Group, Cengage Learning, Elsevier and novelist Scott Turow filed a class action against Google in the U.S. District Court for the Southern District of New York, alleging Gemini was trained on books from Google Books in violation of an agreement permitting only snippet display. The complaint cites an internal Google document indicating the company faced “$10Bs-$100Bs” in potential fines — offered as evidence Google understood the legal risk. The filing follows a similar May action against Meta by nearly the same plaintiff group. Google has notably refused to strike licensing deals with digital publishers where competitors have; that refusal has pushed some publishers, including USA Today, to consider delisting from Google Search within six to 12 months. Anthropic previously settled a comparable authors’ class action for $1.5 billion.
Publisher takeaway
The significance is the pattern, not the filing. The alleged violation is scope creep — a permission granted for one use (snippets) extended unilaterally to another (training) — which is precisely the risk in every crawler permission and licensing agreement on your books today. Go read your own agreements against that lens this week and ask what a counterparty could do under them that you never intended to allow. The $1.5B Anthropic settlement is now the market’s reference price for training on unlicensed content at scale, which strengthens every publisher’s negotiating position whether or not you ever litigate. And the USA Today detail is the most striking line in the story: delisting from Google Search is being seriously modelled by major publishers, which tells you how far the traffic value of search has fallen relative to the cost of feeding it. (Allegations are unproven; Adweek’s date is derived from a relative timestamp, ±1 day.)
09The EU orders Google to share Search data and open Android to rival AI
The European Commission issued two binding specification decisions under the Digital Markets Act. On data: Google must share anonymized Search data with third-party search engines and AI chatbots that have search features, using a multi-layered anonymization method aligned to draft joint DMA/GDPR guidelines from the Commission and the EDPB, under a “fair” pricing formula and a transparent access process — sharing data equivalent to what it uses to optimize its own services, while retaining a right to assess security risk per requester. On Android: Google must let third-party AI assistants be voice-activated (comparable to “Hey Google”) and access Android functions currently reserved for Gemini — reported as 11 features opened to competitors, including delegating app actions. Timeline: anonymized dataset by November 2026, pricing proposal by January, data sharing from January 2027, Android changes by July 2027. Google’s Kent Walker said the decision could undermine “vital privacy and security guardrails for millions of Europeans.”
Publisher takeaway
This is the first serious attempt to pry open the dataset that makes Google’s search monopoly self-reinforcing, and it extends EU enforcement into the AI assistant layer — the surface that’s currently eating your referral traffic. The publisher-relevant question is whether cheaper, better-fed challengers actually emerge: more viable search and AI entrants means more discovery surfaces competing for your content, which is the only structural force that improves publisher terms. Don’t restructure anything around a 2027 timeline, but do two things now. Track which challengers take up the data — they are your future distribution partners and they will be courting content while Google isn’t. And note the precedent: a regulator has now defined “fair access” to a gatekeeper’s data as an enforceable obligation, which is the same argument publishers are making about AI training data in #6 and #8. (Decision terms are per the Commission’s own page; the 11-feature figure is via Reuters.)
10Reddit asks out loud whether its AI data deals are eating its ad business
Reddit EVP of ads monetization Roelof van Zwol said at Cannes Lions that the company is “intensely debating” whether licensing user content to OpenAI and Google for AI training undercuts its own ad differentiation. Reddit sells Reddit Max (automated buying) and Community Intelligence (advertiser insights) on the premise of exclusive “contextual intent” understanding drawn from that same corpus. Van Zwol argued licensed raw content is distinct from the engagement signals powering ad targeting, but conceded “we’re still working, we’re a very young platform.” CEO Steve Huffman has separately raised concerns about external AI systems eroding Reddit’s organic traffic. Analyst Claire Holubowskyj noted Reddit’s interest-based targeting is less exposed to AI disintermediation than identity-based platforms.
Publisher takeaway
This is the most useful public admission of the year, because Reddit is the strongest possible test case — if a platform with a genuinely unique, defensible corpus can’t be sure that licensing it is net-positive, no one selling a smaller archive should assume it is. The distinction van Zwol draws is the one to internalize: licensing your content is different from licensing the signals derived from your audience’s behaviour, and the second is usually the more valuable and more defensible asset. Structure your AI deals accordingly — license the archive if the price is right, but keep the behavioural and engagement layer exclusive, because that’s what you actually sell to advertisers. And run the arithmetic honestly: a licensing cheque that arrives once against ad differentiation that compounds annually is not obviously a good trade, and “we’re intensely debating it” is a more credible position than most publishers’ public confidence.
11Stagwell builds its own AI curation marketplace
Stagwell is launching Stagwell Curate, an AI-powered platform aggregating CTV, online video, display and audio inventory from publishers and ad tech partners, bringing curation in-house. It uses AI agents built with Claude, integrates The Trade Desk’s OpenSincera for domain health assessment, and scores inventory on ad quality, supply-chain quality and technical performance. Matt Adams, global CEO of Stagwell Media Platform: “This is about making sure that we have our own controls over client supply that are bespoke to them.” Stated goals: reduce reliance on third-party paths like OpenPath, lower tech fees, cut auction duplication by concentrating spend on preferred SSPs, and increase client transparency. Gartner’s Andrew Frank noted Omnicom, Dentsu and WPP are building comparable in-house tooling.
Publisher takeaway
A holdco-owned curation layer is a gatekeeper with a scoring rubric, and the rubric is published: ad quality, supply-chain quality, technical performance. That is unusually actionable — you know what you’re being graded on before the grading starts. Concentrating spend on preferred SSPs also means the number of paths into a Stagwell client’s budget is about to shrink, so being on the wrong SSP shortlist is now an existential distribution question rather than a yield-optimization one. Three moves: get your OpenSincera domain-health signals clean and complete, because a third party’s data is now describing you to buyers whether or not you participate; ask your SSPs directly whether they’re on the preferred lists at Stagwell, Omnicom, Dentsu and WPP; and invest in the direct agency relationship, because when curation moves in-house the people picking the list are agency staff, not an algorithm you can arbitrage. (Ties to #2 — curation is the industry’s answer to open-exchange distrust.)
12Google’s Meridian and Meta’s Robyn — the walled gardens are writing the measurement layer
With tracking degraded, marketing mix modeling has become the fallback measurement method — and the two most-used open-source MMM tools are authored by the platforms whose media they score. Google’s Meridian (2024) and Meta’s Robyn (2020) are both free and open-source. Google senior director of data science Harikesh Nair argues open-sourcing Meridian is “crucial for building trust,” but Mike Ryan of Smarter Ecommerce says it maps “the entire Googleverse of media and data in loving, intricate detail.” Measured CEO Nick Stoltz notes Google pushes Meridian partly to drive Google Analytics upgrades, and Google has tied sales-force KPIs to Meridian adoption. Mutinex CEO Henry Innis: “there’s a huge amount of power in setting where everybody starts in a solution.” Three measurement vendors and agency executives told AdExchanger that Meta has dismantled its Robyn engineering team. Amazon Ads’ Jamie Fellows confirmed Amazon’s MMM aim is making “Amazon signal and media available.”
Publisher takeaway
This is the quiet one, and it may matter more than anything above. Measurement doesn’t just describe budget allocation — it determines it, and a model that represents Google’s channels in fine detail while treating open-web inventory as an undifferentiated line item will systematically under-credit you, with no bad faith required anywhere in the chain. Tying a sales force’s KPIs to adoption of your own measurement tool tells you how strategic the platforms consider this. What to do: ask your major advertisers and agencies which MMM they run and how open-web display is represented in it; supply clean, granular, well-labelled data so your inventory can be modelled as a distinct channel rather than lumped into “other digital”; and support independent measurement bodies and vendors, because the alternative to a neutral referee isn’t no referee — it’s the other team’s referee. Whoever writes the measurement layer decides which media looks like it works. (Meta’s Robyn team dismantling is sourced to three unnamed vendors/executives — treat as reported, not confirmed.)