For twenty years, Google took the content and sent back the audience. That exchange paid for newsrooms, product reviews, niche blogs, and a huge amount of the internet people still call the open web.
AI search is breaking that deal. The traffic doesn't disappear because people stop asking questions. It disappears because the answer arrives before the click.
The numbers point in opposite directions
The latest figures are hard to reconcile if you still think search growth and publisher growth are the same thing. On July 22, Alphabet reported second-quarter revenue of $119.8 billion, with Search advertising up 17% to $63.3 billion, according to reporting compiled by PPC Land.
The same week, Reach plc, Britain's largest commercial news publisher, said Google referral volumes had fallen 55% year over year and that it was no longer modelling a recovery. One side of the ecosystem is growing its search revenue. The other side is losing the visits that made the ecosystem valuable.
That isn't a temporary mismatch. It's a transfer of value.
The supply data makes the shift uglier. Ozone benchmarking shared with Digiday tracked roughly 20 billion impressions and found publisher ad request volumes down between 32% and 37% year over year in the United States, and between 39% and 41% in the United Kingdom, during the second quarter.
Fewer impressions, higher prices, lower total spend. That is not a healthy growth market. It's a market compressing around the inventory that still has direct access to people.
AI search traffic changes the product
Most search strategy still assumes the product is a visit. Earn a ranking, win the click, turn the session into a subscription, lead, or purchase. That model is already too narrow for an answer layer that can summarize several sources and keep the user inside the interface.
Google's own 2026 Marketing Live updates show where the ad business is heading. Ads are becoming part of conversational discovery rather than a separate list of links. The winning placement is no longer just the blue link at the end of a query. It is the recommendation, comparison, or product mention that shapes the answer.
That creates a brutal distinction between being found and being used.
A publisher can rank and still lose the visit. A brand can be mentioned and still lose the relationship. A creator can provide the information that makes an answer trustworthy and receive none of the downstream behaviour that proves it worked.
I've written before about the measurement crisis in AI search. The problem is bigger than missing referral data. The commercial unit itself is changing. A citation, an extracted claim, a product inclusion, and a click are not interchangeable outcomes, but most marketing dashboards still treat the click as the only event worth counting.
Publishers are starting to ask the wrong question
The obvious response is to threaten Google. Block the crawler. Raise licensing demands. Put the archive behind a paywall. Refuse to feed the machine that is summarizing the work and keeping the audience.
The anger is justified. The strategy is incomplete.
A publisher that blocks every AI crawler may protect a slice of content, but it can also disappear from the new discovery layer. A publisher that allows every crawler without negotiating usage, provenance, or attribution becomes raw material. Neither choice recreates the old referral bargain.
The better question is not, "How do we get Google to send the old traffic back?" It is, "Which parts of our value should be discoverable, which should be licensed, and which should only be available through a direct relationship?"
That is a product question, not an SEO setting.
The archive can be open while the analysis is paid. A data set can be cited while the interpretation sits inside a membership. A publisher can let an answer engine verify a fact, then make the original reporting materially better for people who follow through. The point is to separate the evidence layer from the relationship layer.
This is also where the lessons from the GEO blind spot become practical. Writing for AI discovery isn't about stuffing a page with phrases that sound machine-readable. It is about making claims specific, attributable, well-supported, and useful enough that a system has a reason to select them.
Marketers are about to buy less certainty
The open web's contraction will not make advertising disappear. It will make the clean-looking version of advertising less believable.
Budgets are already moving toward environments with first-party data, closed-loop purchase signals, and control over the full customer path. Retail media, apps, gaming, and physical screens all benefit from the same pitch: we can tell you who saw the message and what happened next.
That pitch is not always true. It is simply easier to package.
A shrinking open web also creates more incentive for bad supply. When legitimate inventory contracts, made-for-advertising sites can look more attractive because they manufacture volume cheaply. The same reporting noted that made-for-advertising spend rose this year after declining for several years. Scarcity does not automatically improve quality. Sometimes it gives manufactured supply more room to hide.
For marketers, the response should be a harder definition of incrementality. Don't ask only whether a channel produced a visible click. Ask whether it created a qualified action that another channel could not have created, whether the audience can be reached again without renting access, and whether the brand's evidence survives outside the platform that delivered it.
That last question is the one most teams skip. They build a campaign inside a walled garden, measure the garden's version of success, then discover they don't own much of the audience or the proof.
The pattern is familiar in agentic systems too. Vendor lock-in begins as convenience and becomes dependency once the workflow is too expensive to move. Distribution lock-in follows the same curve. At first, Google is a channel. Eventually, it becomes the place where the brand's discoverability, measurement, and audience assumptions all live.
The next advantage is direct evidence
The brands best positioned for this shift will not be the ones with the most content. They will be the ones with the clearest evidence and the strongest reason for someone to come back.
That means publishing original data instead of endless commentary. It means attaching named expertise to claims. It means building tools, communities, products, and services that cannot be compressed into a two-paragraph answer. It means treating every AI citation as a distribution event, not a conversion report.
The open web is still useful. It may become more valuable precisely because generic content is getting squeezed out. But its economics are moving away from infinite pageview supply and toward scarce, trusted, directly reachable attention.
Google can keep growing its search business while the web around it gets poorer. Those facts are not in conflict anymore. They are the business model.
The uncomfortable part for marketers is that there may be no replacement channel that restores the old math. The next version will require better evidence, more direct relationships, and a willingness to measure influence before the click, after the click, and sometimes without one at all.
The web did not lose its audience in one dramatic moment. It is being quietly edited out of the transaction.