SEO was never generous. But it was a contract: you optimize, Google sends traffic. That contract is broken, and the industry is still honoring it.

Thomas Claburn at The Register documents it with precision: Google has converted search from a traffic distribution system into a retention platform, consuming the open web to feed AI Mode while returning proportionally less of the traffic that content created in the first place.

How Google AI Mode Is Draining Search Traffic

Ahrefs research published in February 2026, analyzing 300,000 keywords, found that Google AI Overviews correlate with a 58% reduction in click-through rates for top-ranking pages. Eight months earlier the same figure was 34.5%. The acceleration matters as much as the absolute number.

AI Mode is worse. Seer Interactive analyzed 25.1 million impressions and found that 93% of AI Mode queries produce zero outbound clicks. One out of a hundred readers of an AI answer clicks through to a source. Pew Research confirmed in July 2025 that when AI Overviews appear, only 1% of users click the embedded source links.

Google AI Mode now has more than one billion monthly users, with query volume doubling every quarter. The traffic that isn’t arriving is not statistical noise. It’s the business model of thousands of publishers, quietly draining.

Citation as Alibi

Google responded to cannibalization complaints by introducing inline citations: numbered footnotes in AI answers, visible source chips. The implicit argument: sources are acknowledged, the moral contract holds.

It doesn’t.

A citation without a click is not attribution. It’s free brand awareness for Google and zero-value exposure for the publisher. The reader sees the source name, perhaps trusts the information more because of it, and never visits. The publisher gets no traffic, no data, no conversions. They get a link that 99% of users won’t follow.

ALM Corp’s analysis of 1.3 million AI Mode citations reveals something sharper: Google.com accounts for 17.42% of all AI Mode citations as of February 2026, tripling its share since June 2025. Including YouTube, Google-controlled properties appear in roughly 20% of all citations. The system cites itself more than any independent publisher.

These citations are not an attribution mechanism. They’re a legitimization mechanism: they make it publicly acceptable that Google uses the web as a corpus while not redistributing the traffic that corpus generates. The word “referral” has undergone a quiet semantic mutation. Nobody has yet formalized what it actually means in this context.

What Changes for Developers

For anyone building content distribution tools, analytics systems, or editorial platforms, the implications are concrete.

The metrics we use to measure content distribution are built on a model that no longer exists. “Organic clicks,” “sessions from search,” “Google referrals”: these numbers still measure something, but they measure far less than before. A brand cited ten times a day in AI Mode with zero clicks shows nothing in any analytics dashboard. It exists in the AI answer. It’s invisible in the numbers we use to make editorial decisions.

Building analytics dashboards now means thinking about new columns: AI Mode impressions, citation frequency, brand mention rate in AI answers. Google does not expose these in Search Console with the same granularity as traditional clicks. The GEO (Generative Engine Optimization) tooling market is forming to fill this gap, but it remains fragmented and, for the most part, unverifiable.

As I wrote about AI billing as structural dependency, markets tend to build structural reliances toward whoever controls the infrastructure. SEO for Google has become the clearest form of that dependency: we optimize for a channel that has already stopped reciprocating proportionally, because the alternatives offer even less guaranteed reach.

TechMonk’s Take

Claburn’s piece documents the damage well. But it skips a question worth asking out loud: why does the industry keep investing in traditional SEO with the same logic it used five years ago?

The honest answer is that there’s no easy alternative. The channels that route around Google exist but require years of audience building that most publishers never did, because while Google sent free traffic, it didn’t seem necessary. Those channels (newsletters, direct community, social, referral networks) are still there. Now that the traffic has largely gone, building a direct audience takes years, not quarters.

This is the part The Register doesn’t develop: Google’s cannibalization goes beyond predatory behavior by a dominant company: it forces a reckoning with a dependency the industry chose deliberately for two decades. Nobody forced publishers to build their entire distribution model on a free referral controlled by a single commercial platform. But nearly everyone did, because it was the path of least resistance.

SEO in 2026 is no longer a growth strategy. It’s a defensive posture with diminishing returns: you optimize for Google not to grow, but to lose ground more slowly than your competitors. Teams investing in technical SEO today are spending to preserve residual visibility in a system that has already decided to keep the traffic.

The citation shift is the subtle part. “Attribution” in the AI citation context means your name appears. It means Google can say it cited you. It means the content you produced has a visible reference. It does not mean you’ll receive visitors. It does not mean your editorial investment translates into measurable traffic. Standard analytics won’t capture this kind of presence, and publishers measuring success in organic sessions have already lost the war before noticing.

Building for the open web in 2026 means accepting a specific reality: Google is still the dominant distribution infrastructure, but it is no longer a partner in the content economy. It’s a client that uses your work without paying, offers visibility in exchange for traffic, and has redefined “visibility” to exclude visits to your site.

The Question Left Open

Whether Google is cannibalizing the web is not a debate anymore. The data doesn’t leave room for it. The open question is how long it takes for the industry to stop optimizing for a channel that is replacing it with an internal competitor, and to start building the alternative it should have built ten years ago.

Or more precisely: who will still be around when that moment arrives.