Repricing the Open Web

Where programmatic is actually winning in 2026

Two numbers from the past year refuse to sit next to each other comfortably.

The first: page views from Google Search to publishers fell 34% between December 2024 and December 2025, with Google Discover down another 15% over the same window, according to Chartbeat data provided to Axios. Chartbeat tracks thousands of client sites globally and has been measuring this network for close to two decades. This is not a sampling artifact.

The second: programmatic is on track to reach roughly 90% penetration of worldwide display ad spending, per eMarketer, and will account for approximately 96% of all new display ad dollars in 2026. Automated buying is not just growing. It is absorbing effectively all of the growth.

So the referral layer of the open web collapsed, and the money kept coming. That combination is usually explained away as a lag — advertisers haven't noticed yet, the correction is coming. We think that reading is wrong, and the third number explains why.

Across all global publishers Chartbeat measures, average weekly page views dropped only 6% between 2024 and 2025 — a decline Chartbeat attributes to ordinary factors like an off year for elections and a shifting news cycle, not to AI.

Read those three together and the story changes. Audiences did not leave the web. They stopped arriving through Google. The open web didn't shrink. It got repriced — and the buyers who are winning in 2026 are the ones who understood that distinction early.

The distribution shock was real, and it was regressive

None of this diminishes what happened to publishers. It was severe, and it was unevenly distributed in a way that matters for how you buy.

Over the past two years, per Chartbeat, referral traffic from traditional search declined 60% for small publishers (1,000–10,000 daily page views), 47% for medium publishers (10,000–100,000), and 22% for large publishers (100,000+). The gap between the smallest and largest sites is nearly three times.

That asymmetry is the whole game. Large publishers were less exposed because search was never as dominant a share of their traffic to begin with — they have brand recognition, apps, newsletters, and direct audience relationships to fall back on. Small sites built on search arbitrage had no second channel. When the first one closed, there was nothing underneath it.

The consequence for media buyers is straightforward: the long tail of the open web got structurally weaker, and the quality gap between premium supply and everything else widened. Inventory that was already marginal in 2023 is now being propped up by traffic sources you would not choose to buy against if you could see them clearly.

The LLM traffic story is a rounding error

Here is where most 2026 commentary goes wrong. The expectation was that AI assistants would become the new referral engine — that traffic lost to Google would resurface as citations in ChatGPT, Perplexity, and Gemini.

It didn't. Page views from ChatGPT referrals grew more than 200% during the same period, per Chartbeat. Chatbots still account for less than 1% of all publisher page view referrals.

Both facts are true simultaneously, and the second one is the one that should drive planning. A channel growing 200% off a base under 1% is not a distribution channel yet. It is a signal about where attention is moving, not a place to move budget.

This is the single most useful correction we can offer a client right now. LLM assistants are not redistributing the traffic that search lost. They are absorbing the intent — answering the question so the click never has to happen. Those are different problems requiring different responses. You cannot win back a click that no longer exists by optimizing for a referral that isn't coming.

Where the dollars actually went

If the traffic didn't move to AI assistants and the total audience barely moved, then the money went somewhere structural. It did. Three consolidations account for most of it.

1. Search itself fragmented

The most under-discussed number of the year: Google will earn 48.5% of search ad spending in 2026 — the first time in more than twenty years that figure has fallen below half, according to eMarketer. Amazon is taking the biggest share of what Google gave up.

Forward-looking, it accelerates. eMarketer projects Amazon will account for 43.4% of all new US search ad spending between 2026 and 2028 — more incremental dollars than Google will add over the same period.

Search is not dying. Search is decentralizing toward the surfaces where purchase intent is most legible. That is a very different diagnosis, and it points at a very different budget response.

2. Commerce data became the premium signal

US retail media ad spend is forecast to reach $69.33 billion in 2026, up from $58.79 billion in 2025, per eMarketer, with much of the incremental spend flowing to Amazon Ads and Walmart Connect.

More telling than the headline number is where inside retail media the growth is concentrated. eMarketer finds retail media CTV spend growing roughly three times faster than retail media search. Retailers are no longer just monetizing their own search results. They are extending first-party purchase data outward — into streaming, into the open web, into every environment where that signal can be applied.

This is the actual answer to signal loss. As third-party identifiers became unreliable, the market did not find a universal replacement. It found pockets of durable first-party signal and bid them up. Commerce data won because it is the highest-fidelity signal still legally and technically available at scale.

3. The open exchange lost share to structured supply

eMarketer's H2 2026 programmatic forecast puts programmatic direct at 76.3% of US programmatic spend. Strip out social, and direct's share falls to 50.4% — meaning the genuinely open, unstructured auction now accounts for well under half of non-social programmatic dollars.

The same pattern shows up in video. IAB projects CTV growth near 13.8% year over year, second-fastest of any channel behind social. The structural milestone: CTV upfront commitments of $17.73 billion are forecast to exceed primetime linear TV upfronts of $16.98 billion for the first time, with US CTV spend landing around $38 billion for the year.

Upfronts are a commitment structure. Streaming inventory crossing that threshold means the most valuable digital video supply is being locked up in advance, by relationship, at negotiated terms. That is the opposite of open-auction dynamics — and it is where the premium sits.

The through-line across all three shifts is identical: scarce, verifiable signal commands a premium, and buyers are paying it. Retail media won on purchase data. CTV won on attention and content quality. Curated and direct paths won on inventory transparency. None of that requires publisher referral traffic to hold up. It only requires the audience to still be there — and the audience is still there.

The AI surface: real budget, wrong mental model

The AI advertising numbers are large enough that they deserve their own treatment, and specific enough that most of the commentary around them is misleading.

US AI ad spending will reach $32.03 billion in 2026 — nearly triple the prior year — and exceed $68 billion by 2030, per eMarketer's May forecast. Those are serious numbers.

Now the detail that reframes them: more than 80% of that spending flows through traditional paid search listings appearing alongside AI-generated results, not into chatbot conversations.

In other words, most "AI advertising" in 2026 is being bought and managed by the paid search teams that already exist, through the platforms they already use, against inventory that sits next to an AI answer rather than inside one. The revolution arrived, and it showed up in the accounts you were already running.

The genuinely native piece — sponsored placements inside conversational responses — is real but early. eMarketer projects AI search ads reaching 13.6% of all US search ad spending by 2029, up from 0.7% in 2025. That is a steep curve from a very small base, which is exactly the shape of a channel worth testing with learning budgets and exactly the wrong shape for a major reallocation.

Meanwhile the surface itself is already ubiquitous. More than 60% of US commercial Google queries returned an AI Overview as of May, per eMarketer's reporting. The answer layer is fully deployed. The buying motion around it simply hasn't caught up.

The strategic error we see most often is treating this as a content problem — a scramble to get cited, restructure pages, chase visibility in model outputs. That work has value, but it is being asked to carry weight it cannot bear. The near-term AI advertising opportunity is a paid media problem sitting in existing search and programmatic budgets, and it should be staffed and measured accordingly.

What we'd tell a CMO heading into Q4 planning

Budget follows signal, not sessions. Stop using publisher traffic health as a proxy for open web viability. The two decoupled this year. Evaluate supply on the quality of the signal it carries and the transparency of the path to it, not on whether its referral chart is up or down.

Buy structured supply rather than blocklisting your way to quality. With unstructured open-auction inventory now a minority of non-social programmatic spend, exclusion lists are a rear-guard action. Curated and direct paths give you inventory quality as a property of the buy, not something you claw back after the fact.

Treat GEO as visibility infrastructure with a known paid timeline. Getting surfaced in AI answers is earned-media work with a real return. But a paid layer is arriving on a forecastable curve, and it will be bought through search and programmatic channels. Build the organic visibility now; plan the media response as a paid discipline, not a content one.

Move one cycle early on the emerging surfaces, sized honestly. AI-native placements deserve test budgets proportional to their current share of spend — which is small — not to their headline growth rate. The advertisers who built creative, targeting, and measurement capability in CTV before it was crowded are the ones buying it efficiently today. The same window is open now, and it will close the same way.

The open web spent fifteen years being priced as a reach channel, valued on the traffic Google sent it. That pricing model broke this year. What replaced it is narrower, more structured, more expensive per impression, and considerably more accountable — a performance channel wearing the old channel's clothes.

The buyers still pricing it as reach are the ones losing. The ones who repriced it are quietly having their best year in a decade.

Sources

• Chartbeat data provided exclusively to Axios, March 2026 (publisher referral traffic by size; Google Search and Discover page view declines; chatbot referral share; aggregate page view change)
• eMarketer, Worldwide Programmatic Ad Spending (programmatic share of display; share of incremental display dollars)
• eMarketer, US Programmatic Advertising Forecast and Ad Tech Trends H2 2026 (programmatic direct share, total and ex-social)
• eMarketer, US Search Advertising Forecast 2026 (Google share of search ad spend; Amazon share of incremental spend)
• eMarketer retail media forecasts (US retail media spend 2025–2026; retail media CTV vs. search growth)
• eMarketer AI ad spending forecast, May 2026, and AI search ad spending forecast (total AI ad spend; share via traditional paid search; AI search share of search spend through 2029; AI Overview query coverage)
• IAB / eMarketer CTV forecasts (CTV growth rate; CTV vs. primetime linear upfront commitments; US CTV ad spend)