An ad request is a by-product. It exists only because a browser loaded a page, and a browser loads a page only because something sent a reader to it. Break that first link and the last one disappears quietly, well before it shows up in a revenue line. That is what has happened across the open web through 2026: search and social have stopped forwarding readers at the old rate, so there are fewer pages loaded, fewer slots on those pages, and fewer impressions to sell. A decade of work optimising the auction does nothing about it, because the shortage sits upstream of the auction entirely.
How far has publisher ad supply actually fallen?
The clearest public measurement comes from Ozone, whose benchmarking data was shared exclusively with Digiday and reported by Jessica Davies on 15 July 2026. Across roughly 20 billion impressions, publisher ad request volumes between April and June 2026 were down by roughly 32% to 37% year on year in the United States and 39% to 41% year on year in the United Kingdom.
Two things make that dataset worth taking seriously. It measures the software Ozone runs for its publisher members rather than its own ad sales, so it counts ad opportunities rather than sold inventory. And the cohort is premium: Ozone's publisher network includes the Guardian, News UK and Dow Jones' Wall Street Journal. This is not the long tail thinning out. It is the top of the market losing about a third of its inventory in twelve months.
The revenue consequence followed. Across the first half of 2026, combined programmatic spend in Ozone's US and UK cohort was down 30.6% year on year. The regional split is the interesting part: spend fell 14.3% in the UK, where yields rose sharply, and 44% in the US, where they did not.
Format-level detail from the same June data shows where what remains is going. Apps were the only channel growing year on year, with spend up about 23% and eCPMs up roughly 42% in the US. Web inventory rebounded 11% month on month but stayed around 38.5% below the same period a year earlier. Out-stream video spend fell roughly 76% year on year.
Why does falling referral traffic destroy ad supply so efficiently?
Because the relationship is mechanical, not behavioural. Every ad call requires a page load, and most publisher page loads are, or were, the end of a referral chain that started somewhere else. When an AI answer resolves the question in the interface, the page is never loaded, so the auction never runs. There is no partial credit.
Danny Spears, chief operating officer at Ozone, put the cause plainly in the Digiday report: "Platforms, particularly Google, are intervening in the user journey and providing content in situ rather than redirecting to the underlying website as they used to with classic search."
It is not only AI search. Matt Barash, chief commercial officer at Nova Studio, framed the wider shift in the same reporting: "The open web is becoming smaller not because demand for content has fallen, but because discovery has moved." His summary of the mechanism is the cleanest one available: "Search created pageviews. Social creates engagement. As consumers spend less time navigating the open web, premium display inventory becomes scarcer and more valuable."
Reach plc gives the audited version of this arithmetic. In half-year results for the six months to 30 June 2026, the UK's largest commercial news publisher reported Google referral volumes down 55% year on year, on-platform page views down 40%, and revenue per thousand page views up 49%. Digital revenue fell 11.4% to 54.2m pounds. Group revenue fell 9.0% to 232.9m pounds and the interim dividend was halved to 1.44p. Chief executive Piers North told investors the company must work to the assumption that referrals are unlikely to recover.
Multiply a 40% volume loss by a 49% price rise and you land near an 11% decline, which sits close to the 11.4% digital revenue fall Reach reported. The correspondence is looser than it looks, because Reach says only about a third of its digital revenues are exposed to on-platform activity and that third is programmatic. But the shape of the trade is exactly right. A 49% price increase recovered most of a 40% volume loss. Most is not all, and the price increase is the larger of the two numbers.
If supply is scarce, why has price not made publishers whole?
Prices have moved, and in the direction textbook scarcity predicts. In June 2026, average eCPMs were about 30% higher year on year in the UK and 7% higher in the US. Demand has not walked away either: writing in What's New in Publishing on 28 July 2026, adtech and data monetisation consultant Mimmo Palmieri noted that bid density in the Ozone June data held steady at 5.4 bids per ad request. Buyers are still bidding. They are chasing fewer impressions and paying more for the ones left.
Liza Simonova, senior customer success and go-to-market manager at Ozone, gave the honest two-part reading: "The fact that eCPM is increasing is actually a healthy sign of the market because it means that we [publishers] have scarce resource and people are willing to pay more for it." And then the second half: "But because we have less ingredients to monetize, ultimately the spend is down."
That is the whole problem in two sentences. Yield is a multiplier on a shrinking base. A 30% price rise against a 40% volume fall is a net loss, and the US case shows what happens when the price rise is only 7%.
Gabe Dorosz, advertising initiative lead at INMA, explained why the repricing is slower in the US: "The U.S. open web is bigger and messier, with a long tail of commodity supply still diluting yields, so the repricing is naturally showing up more slowly but my prediction is it will accelerate." Worth noting for anyone benchmarking a US portfolio against UK numbers and concluding they are underperforming. They may simply be earlier in the same curve.
It is also worth remembering that this is not the first year of price rises. Across the first half of 2025, average eCPMs were already up around 42% in the US and 36% in the UK while supply was starting to fall. Both markets were effectively yield-led before this year's deeper supply shock arrived. The easy headroom has already been used.
Where is the yield ceiling?
Nobody knows, and the people closest to the data say so.
Luke Stillman, managing director at Madison and Wall, pointed to television as the precedent in the Digiday report: "People were highlighting the risk of a yield-led market in TV for more than a decade, and yet marketers who prioritize TV are willing to pay multiples of what they used to for the same inventory, despite smaller audiences." His conclusion is that the ceiling is hard to call: "It's very difficult to say when 'enough is enough' in terms of pricing that actually spurs channel shifts." In his view, most migration from the open web into walled gardens will happen for other reasons anyway, specifically data availability, measurement credibility and convenience.
That should temper both the optimistic and the pessimistic reading. Publishers should not assume the price rises stop next quarter, because TV suggests a premium channel can reprice for years. But they should also not assume price is what keeps buyers, because on Stillman's argument budget leaves for reasons that have nothing to do with CPMs.
There is a structural argument on the other side that gets less attention. Palmieri points out that open programmatic was built on the assumption of abundant supply: real-time bidding and the systems around it were designed for large pools of impressions, and buy-side bidding algorithms are tuned to operate at scale. Scarcity changes that maths. Smaller sample sizes make optimisation harder and more expensive, which gives the buy side its own reason to be selective rather than to chase reach. That points towards curation and direct deals rather than towards an ever-rising open-auction clearing price. Citing eMarketer, Palmieri notes that direct and programmatic guaranteed deals already account for more than three quarters of US programmatic spend, and that share is likely to grow as sellers with strong direct operations conclude the open auction no longer makes economic sense for their best inventory.
Why chasing volume back makes the problem worse
The instinct in a falling-supply market is to manufacture supply: add ad units, buy traffic, lean harder on low-value inventory. Palmieri's argument is that this treats a structural shift as a seasonal dip, and that it is self-defeating on its own terms. Buying traffic is getting more expensive because there are fewer users to buy, and it dilutes precisely the quality signals that now command the premium.
Some publishers are already doing the opposite. Dorosz noted in the Digiday reporting that part of the supply shrinkage is deliberate: "Some publishers are also deliberately shaping supply, cutting ad load and low-value bid requests to protect or increase attention and price." In a market where price is rising and volume is falling regardless, cutting the worst 20% of your requests costs less than it used to and improves the average of what is left.
Dorosz was blunt about what is ending: "What you're seeing is publishers realizing that buying into the 15-year programmatic strategy of 'infinite supply' has effectively just steadily driven CPMs down, and AI-driven traffic declines have now killed the idea that there actually is infinite supply and this is without doubt a race to the bottom."
The uncomfortable implication is cultural rather than technical. Yield teams have spent fifteen years optimising for pageviews, impressions and session counts. A scarcity market asks for engagement, retention and lifetime value instead. That is a harder change to make than absorbing the revenue hit, because it invalidates the metrics people are measured on.
What actually replaces the lost inventory?
Nothing replaces it one for one. Anyone promising a like-for-like substitute for a third of your ad requests is selling something. What is available is a set of partial replacements, and the sensible position is to run several.
Higher yield on protected supply. This is already happening and it is the largest single offset in the data. It means moving the best inventory, logged-in users, premium placements and high-value formats, out of the open auction and into direct and guaranteed deals.
Evidence that your impressions are different. Dorosz frames the opportunity as sitting with publishers who can surface meaningful signals, "through first-party, contextual, attention or other data", to show their impressions meet the quality thresholds the buy side is now actively seeking. In a scarce market the premium goes to provable difference, not to scale.
Environments the platforms do not intermediate. Apps were the only growing channel in the June data. Spears observed that "Publishers with subscription businesses and apps look a lot more resilient in the face of supply decline." Newsletters and events belong in the same category. They are smaller than the display business they partly replace, and they are owned.
Licensing and content deals, with clear eyes about the distribution. Only the largest publishers have been offered meaningful terms, and the sums are small relative to advertising. Reddit is the instructive case: PPC Land reported that its data licensing revenue reached 36 million dollars in a quarter against 549 million dollars in advertising, which is licensing running at roughly 6.5% of the advertising line at a company with unusually attractive data.
Monetising the demand that replaced the pageview. This is the piece that is missing from most versions of this analysis, and it is worth being precise about it.
Where does AI agent traffic sit in these numbers?
It does not sit in them at all, and that is the point.
The Ozone figures count ad requests. An AI agent retrieving a page does not generate one. It issues an HTTP request, takes the HTML, extracts what it needs and leaves. No JavaScript executes, so no auction runs, no impression is recorded and nothing is counted in any of the numbers above. The reader consumption did not stop. It moved to a channel that is structurally invisible to the measurement systems reporting the decline.
So the supply crunch is understated as a description of what is happening to publisher content. Ad supply fell by up to 40% while content retrieval by machines rose. The gap between those two lines is the commercial question of the next two years.
This is the layer blankspace operates in. Detecting Live Search Agent and LLM traffic at the CDN edge, before the request reaches the origin, makes those retrievals countable, and injecting contextual brand facts into the response makes them saleable. The honest framing is that this is a different transaction from a display impression rather than a replacement for one. It does not restore lost ad requests, it is not currently at the scale of the loss, and the industry does not yet have an agreed way to count it, which is what the IAB's forthcoming attribution framework is trying to address. What it does is put a price on consumption that presently earns nothing at all.
For a yield team, the practical consequence is narrower than the strategic one. Any forecast built only on ad requests is now measuring a declining share of the audience. The first job is to instrument the traffic that does not produce an ad call, so that the size of it is a known number rather than an assumption.
Frequently asked questions
How much did publisher ad supply fall in 2026?
By up to 40% year on year in the second quarter, according to Ozone benchmarking data reported by Digiday on 15 July 2026. Ad request volumes were down roughly 32% to 37% year on year in the US and 39% to 41% in the UK between April and June, measured across around 20 billion impressions from a premium cohort including the Guardian, News UK and the Wall Street Journal. Combined programmatic spend across that cohort fell 30.6% year on year across the first half.
Why are eCPMs rising while ad supply falls?
Because demand has not fallen at the same rate as supply, so buyers are competing for a smaller pool. Bid density in Ozone's June data held steady at 5.4 bids per ad request, which indicates buyers are still present and simply have less to bid on. Average eCPMs in June were about 30% higher year on year in the UK and 7% higher in the US. Scarcity pricing is functioning as economics would predict; it is the volume underneath it that has changed.
Can higher CPMs fully replace lost ad volume?
Not so far, and not in any published dataset. Yield is a multiplier on a shrinking base, so a 30% price rise against a 40% volume fall still leaves a net loss. Reach plc's half-year results are the clearest worked example: on-platform page views down 40%, revenue per thousand page views up 49%, digital revenue still down 11.4%. Price recovered most of the loss and not all of it, in the market where prices moved the furthest.
Should publishers cut ad load in a shrinking market?
Several already are. Gabe Dorosz of INMA noted that part of the supply decline is deliberate, with publishers cutting ad load and low-value bid requests to protect or increase attention and price. The logic is that in a market rewarding scarcity, removing your weakest requests costs little revenue and improves the quality signal on what remains. The counter-argument is that it accelerates the volume decline you are trying to survive, so it is a decision to make with data on which requests actually clear.
Does AI agent traffic show up in ad supply figures?
No. Ad supply data counts ad requests, and an AI agent retrieval does not generate one, because no JavaScript executes and no auction runs. That means the reported supply decline understates the change in how publisher content is being consumed: some of the audience did not disappear, it moved to a channel that produces no measurable ad opportunity. Publishers who want the full picture need to instrument agent traffic separately, at the server or CDN edge, rather than inferring it from analytics that cannot see it.