For twenty-five years the open web ran on a simple loop: a page loads, the pageview fires an ad request, the auction clears, the publisher gets paid. New Ozone data shows that loop breaking. Publisher ad supply fell by up to 40% in Q2 2026 as AI search cut the referral traffic that creates pageviews — the clearest measure yet of how publisher ad revenue in AI search erodes when discovery moves into the answer.

  • Ozone benchmarking shows publisher ad request volumes down 32–41% year over year across the U.S. and U.K. in Q2 2026.
  • Prices are masking the damage: June eCPMs rose about 30% in the U.K. and 7% in the U.S., so spend fell less than volume.
  • Rising yield on shrinking supply is a delay, not a recovery — the audience now sits inside AI answers publishers do not monetize.
  • Brands should budget for visibility inside AI answers; publishers should price the future on citation, not on a traffic rebound.
  • The missing piece is an ad layer that pays the publishers whose content AI engines cite.

What an ad request is, and why it is vanishing

An ad request is the signal a publisher's page sends to the ad market every time it loads, offering that impression for sale. No pageview, no ad request, no revenue. The double-digit decline Ozone measured is not a pricing event; it is the disappearance of the pageviews themselves, as search and social increasingly keep users from ever reaching the page.

The pageview economy versus the answer economy

In the pageview economy, attention was monetized on the publisher's page, one ad request at a time. In the answer economy, attention is captured inside an AI-generated answer that reads the publisher's content but never loads their page. Demand for the content has not fallen. The surface where it is consumed — and monetized — has moved.

Google still earns from the ads around its AI answer. The publisher who supplied the underlying facts earns nothing. The value did not vanish; it changed hands.

What Ozone reported, and what the wider data confirms

Ozone's data, shared exclusively with Digiday on July 15, 2026 and drawn from roughly 20 billion impressions across members including the Guardian, News UK and the Wall Street Journal, shows first-half 2026 programmatic spend down 30.6% year over year. Ozone COO Danny Spears attributes the drop to platforms, chiefly Google, answering in place rather than redirecting users to the source.

The proprietary figure lines up with independent data. Similarweb and Chartbeat, reported by Press Gazette, put Google referral traffic to publishers down about a third in the year to November 2025; SparkToro's 2026 analysis found fewer than a third of Google searches now end in a click, and Pew Research recorded an 8% click rate when an AI summary appears versus 15% without. Different datasets, same direction.

This is without doubt a race to the bottom.

— Gabe Dorosz, Advertising Lead, INMA

Why rising eCPMs are not the recovery they look like

The strongest case for a yield-led open web

The optimistic read is credible. Scarcity raises price: U.K. buyers paid roughly 30% more per thousand impressions even as volume fell, and AI-referred visitors convert several times higher than organic ones, per Ahrefs and Semrush 2026 data. Luke Stillman of Madison and Wall points to television, where marketers pay multiples for smaller audiences. On this view, premium publishers trade volume for value and survive.

Why scarcity pricing has a floor the open web can't hold

The analogy breaks on one point. Television kept its audience inside the walled garden it monetized; the open web is losing its audience to a surface — the AI answer — that the publisher does not monetize at all. Yield on a shrinking base is a smaller business each year. Stillman himself concedes buyers will migrate to walled gardens for data and measurement regardless of price. Higher eCPMs buy time to re-architect; they do not reverse the shift.

What this means for brands and for publishers

For CMOs, media buyers, and agencies: budget for the answer, not just the page

Treat AI visibility as a line item, not a rounding error. Open-web display inventory is repricing and shrinking, while the audience sits inside AI answers where an ad layer is already forming — OpenAI began showing ads in ChatGPT in February 2026, and Google now serves ads in roughly a quarter of AI Mode answers. Audit how much of your funnel depends on open-web referrals, and ringfence a test budget for placement inside AI answers before a competitor is cited and you are not.

For publishers: monetize the citation, not the traffic rebound

Stop pricing the future on traffic recovery. Subscriptions, apps and logged-in products are sound defense, but they concede the open web rather than reclaim it. The durable asset is being the source AI engines cite. The revenue model that fits that role is compensation for the citation itself — being paid when your content grounds the answer, not only when someone lands on your page.

Three shifts that will decide who gets paid by 2027

Three signals point to where this lands. First, a payment rail is forming: on June 3, 2026, Cloudflare's Matthew Prince reported that bots now generate 57.5% of HTML web traffic, and from September 15 Cloudflare will block training and agent bots by default on ad-supported pages while opening a gateway to charge them. Second, the ad layer inside answers is real but engine-controlled — the AI search ad market is estimated at $15–25 billion in 2026, yet the cited publisher is not in that value chain. Third, brands are hiring for AI visibility. Whoever standardizes payment to the cited source first will set the price for everyone else.

Conclusion

Hold on to one thing: the open web's ad call is being replaced by the AI answer, and the content still flows even as the payment stops. Yield-led survival and subscription pivots are defense; they do not restore the publisher ad revenue that discovery carried away. The structural fix is an ad layer built for AI Search — Generative Engine Advertising that places native ads inside AI answers and pays the publishers whose content is cited. That is the category Smalk AI is building: a bridge between brands that need to appear in AI answers and the media sources those answers depend on. What to watch next: who standardizes payment to the cited source first — that is the moment the answer economy prices in.