Something structural is happening to the open web. And it’s showing up in the data so fast that ignoring it in your Q3 planning would be a mistake.
Zero-click searches (queries resolved by an AI summary without the user ever visiting a publisher’s site) now account for roughly 60% of all Google searches, according to Similarweb data. Click-through rates drop between 34% and 61% when an AI Overview appears on the results page. Google search traffic to publishers fell 33% globally in 2025. Individual publishers have reported declines far steeper: some down 50%, others down 70–80%, a few reporting near-total losses on certain content categories. Google’s I/O 2026 announcements made clear this is a deliberate product direction, not a side effect the company is looking to reverse.
For performance marketers running programmatic display as a meaningful share of their acquisition mix, that’s not just an industry headline. It’s a structural change to the inventory environment your campaigns depend on.
What’s Actually Happening to Open-Web Display Inventory
Here’s where it gets more complicated than a simple “supply is shrinking, CPMs are rising” narrative and why the real risk is actually harder to manage than pure price inflation.
Overall web traffic is down only about 6% year-over-year, according to Chartbeat, and total programmatic display spend in the U.S. grew 16.6% in 2025, hitting $187 billion. So the supply isn’t collapsing in aggregate. What’s happening is a quality and composition shift that’s harder to see in headline numbers.
Search referral traffic (the traffic that historically fueled mid-funnel publisher inventory on informational content pages) is declining sharply. Pageviews from Google Search fell 34% year-over-year across Chartbeat’s publisher network between December 2024 and December 2025. That traffic isn’t being replaced by AI referrals: chatbots currently account for less than 1% of all publisher pageview referrals. The gap is being filled partly by direct traffic, internal traffic, and dark social — channels that reflect loyal audiences, not prospecting-stage readers.
For performance marketers, this matters because mid-funnel publisher content is where prospecting audiences live. When AI search absorbs informational queries without a click, the inventory attached to that content shrinks and the remaining open-web supply skews toward direct and branded audiences who were already in your funnel. You’re not necessarily paying more for the same audience. You may be paying similarly for a less useful one.
The CPM picture reflects this mixed reality. Overall programmatic CPMs rose 34% year-over-year in April 2026, but that increase is driven by a combination of strong seasonal demand and a broader shift of budget toward premium and private marketplace inventory as advertisers follow quality. Open-exchange display CPMs remain in the $2–$6 range, while private marketplace deals now average $8–$15. The spread between commodity and curated inventory has widened more than 60% since 2024. Budget is chasing quality, which means open-exchange supply, which is most directly affected by search traffic decline, is getting bid on by a shrinking pool of buyers willing to accept the quality trade-off.
The trajectory matters more than any single quarter’s numbers. AI search products are getting better at resolving queries without clicks. The I/O 2026 redesign turned Google Search into an AI answer engine that keeps users on the results page. This isn’t an algorithm update publishers can optimize around — it’s a fundamental change in how information retrieval works upstream of every publisher’s monetization model.
Three Ways This Changes Your Acquisition Cost Model
Audience quality drift on open-web display. As search-referred traffic declines, the prospecting audiences that programmatic display has historically reached shift toward AI interfaces, not publisher pages. The audience remaining on the open web is more likely to be direct, loyal, or already in your funnel. Prospecting on open-web display becomes harder to justify if the prospecting-stage audiences you’re paying to reach are spending less time on the pages your ads appear on.
Attribution opacity gets worse, not better. As AI intermediaries absorb traffic that once generated measurable site visits, the path between ad exposure and conversion gets harder to trace. A user who sees your display ad, then asks an AI assistant about your category, then searches your brand name and converts, looks like branded search in your attribution model. The display impression that contributed to the decision gets no credit. This problem isn’t new, but AI search accelerates it.
Budget concentration risk in walled gardens. As open-web supply quality and targeting precision erode, media budgets consolidate into walled gardens — platforms with their own authenticated identity and closed-loop measurement. That consolidation gives walled garden platforms more pricing power over time, and it means more of your acquisition infrastructure depends on identity and measurement you don’t control. The more your spend concentrates there, the more exposed you are to their policy changes, their pricing decisions, and their definitions of what counts as a conversion.
Why Fixed, Deterministic Supply Changes the Unit Economics
Programmatic direct mail operates on a supply model that none of these forces affect.
The supply is the USPS delivery network and approximately 130 million verified U.S. household addresses. That number doesn’t decline when AI search improves. There’s no algorithm update that reduces the number of mailboxes. No platform policy shift siphons the impression before it reaches the recipient.
Every piece of mail that enters the mailstream reaches a physical address. The impression is deterministic — you know which household received it, when it was delivered, and whether that household subsequently converted, validated through address-level matchback attribution. There’s no view-through ambiguity, no platform-reported metric to interrogate, and no supply-side variable eroding your unit economics as AI search matures.
Direct mail costs are also fixed before launch. Print and postage don’t float with auction dynamics or shift when competitor budgets rotate into your target segments during peak season. If your display CAC is rising quarter over quarter — whether from CPM inflation, audience quality drift, or attribution erosion — direct mail gives you a channel where the cost denominator doesn’t change mid-campaign.
How to Pressure-Test Your Display Allocation Against the New Supply Reality
If your current mix puts a significant share of acquisition budget on open-web display, here’s a practical framework for assessing your exposure.
Pull your quarter-over-quarter CPMs for in-market audience segments on open-web display. If you’re seeing meaningful increases alongside declining reach against prospecting-stage audiences, the supply composition shift is already showing up in your cost data.
Stress-test your attribution model for AI-influenced conversions. How many of your branded search conversions might have been driven by category awareness built on publisher pages that are now disappearing? If you can’t answer that, your current model may be over-crediting lower-funnel channels and under-crediting the awareness investment that preceded them.
Run a holdout-tested programmatic direct mail pilot against your highest-value prospect segments. Use your first-party CRM data to build ML-powered lookalike audiences, mail a treatment group, hold out a matched control group, and measure incremental conversions through matchback attribution. You’ll have a real CPA number (not a modeled estimate) within a single campaign cycle, on a channel where the measurement methodology doesn’t depend on search traffic patterns changing around you.
Compare true incremental CPA across channels. Because programmatic direct mail supports randomized holdout testing at the household level, you can compare a ground-truth lift number against display metrics that increasingly rely on probabilistic attribution in a shifting supply environment. That comparison should be the basis of allocation decisions, not platform-reported ROAS.
This Isn’t a Fire Drill — But It’s Worth Treating Like One
The brands that will maintain acquisition efficiency through this shift aren’t the ones optimizing display bids most aggressively. They’re the ones that recognized a structural change in the supply environment and diversified into channels where impression economics don’t depend on an AI interface deciding whether to send users to a website.
Channels with deterministic, fixed supply and address-level measurement aren’t an experiment for next year’s planning cycle. They’re the structural counterweight your acquisition model needs while this shift is still in early innings.
If you’re thinking through the impact of AI search on your acquisition costs, Postie’s lookalike modeling and matchback attribution can deliver a real incremental CPA within a single campaign cycle — on a channel where none of this changes the math.
