The Bounce-Click Argument

Google has a name for the clicks that AI Overviews replaced. It calls them bounce clicks. The framing, articulated by a senior Google search executive last year and repeated in various forms across every Google statement on the subject since, is that the clicks AI Overviews removed were the ones that were not actually working in the first place. A user clicks a link, the page does not have what they wanted, the user clicks back to search and tries again. That is a bounce. Bounces, the argument goes, are wasted clicks. AI Overviews answer the query directly, the user gets the fact, no bounce, no wasted click, everyone wins. Users who actually want to read the article still click through. The publisher only loses the click that was never valuable to them.

The argument is technically defensible. The argument is also, almost line for line, the same one regional sports network executives offered in 2018 about the cable bundle.

I want to write about why those two arguments belong in the same essay.

What the Data Says

Start with the numbers, because the numbers are not in dispute. The disagreement is about what they mean.

A field study published last year tracked 42% of organic queries triggering AI Overviews and measured outbound clicks at 0.38 per search; with AI Overviews removed, the same set of queries produced 0.61 clicks per search. That is a 38% reduction in clicks attributable to the Overview. Other studies have reported organic CTR drops between 34% and 61% on AIO-triggered keywords.

Publisher-level data tells the same story at higher resolution. Wired’s Google traffic is down 62% over the AI Overviews rollout window. The Verge, HowToGeek, and ZDNet have each lost more than 85%. Mashable is down 30%. Digital Trends went from 8.5 million monthly Google clicks in March 2024 to 264,861 in January 2026, which is a number that takes a moment to process. Chartbeat reported that small publishers — sites in the 1,000 to 10,000 daily-pageview range — saw Google referral traffic drop by roughly 60% over two years. The IAB Tech Lab estimated AI summaries reduce publisher traffic 20%–60% on average and translate to roughly $2 billion in annual ad revenue losses across the publishing sector.

These are not edge cases. These are not contested numbers. Even Google does not really dispute them in aggregate; what Google disputes is the claim that the clicks lost were the kind of clicks that ever mattered.

Hold that disagreement in mind, because it is the entire game.

The Defense, Stated Plainly

The Google argument, in its strongest form, runs like this.

Search has always had a long tail of low-value clicks. A query like “what time does the Super Bowl start” used to send the user to a page that took eight seconds to load, served a banner ad, and answered the question in one sentence the user could have read in a knowledge panel. The user got nothing valuable from the visit. The publisher got a click that was not going to convert, was not going to retain, was not going to do anything for the business beyond produce one CPM impression at the bottom of the page. AI Overviews remove that click. The publisher loses the impression. The user saves five seconds. Net welfare goes up.

The clicks that survive, in this telling, are the clicks the publisher actually wanted — readers who want to spend time on the page, sign up for the newsletter, return for the next article, become a subscriber. Those readers are the publisher’s real audience. The bounce traffic was always noise.

Google has separately argued that the methodologies used in the public studies are flawed and that the company “consistently directs billions of clicks to websites daily” without observing significant drops in aggregate web traffic. More recently, the company has pointed to AI Overviews CTR rising from 1.3% in late 2025 to 2.4% by February 2026 as evidence that the click-through rate is recovering as the product matures.

I want to be careful here. The defense is not nothing. There genuinely is a category of click that the publisher never monetized well. There genuinely is a recovery underway in raw AIO click-through rates. There genuinely are publishers — direct-traffic newsletters, paid Substacks, vertical specialists — who never depended on the bounce-click economy and are not seeing the decline.

The defense is true. That is the part of the defense that is worth saying out loud, because if it were obviously false, this essay would not be necessary.

The problem with the defense is not that it is false. The problem is that it is the same shape, almost word for word, as a defense I have already watched run its course in my own industry.

"The new product is cannibalizing some of the old product, but the old product still has revenue, and the cannibalized portion was the marginal portion." That sentence describes the AI Overviews defense. It also describes every regional sports network deck I saw between 2017 and 2022.

What 2018 Sounded Like Inside an RSN

I have spent enough of my career in sports media to know what the conversation around regional sports networks sounded like in the back half of the last decade. By 2018, cable subscriptions had been declining for six straight years. The DTC narrative was already in motion. ESPN+ had launched. DAZN was building out its U.S. footprint. Cord-cutting headlines ran weekly.

What the RSN side said, repeatedly, was a structurally specific version of the bounce-click argument.

The cable bundle was still, by an enormous margin, the largest revenue source for the franchise. Local rights inside the RSN system paid more per fan than any direct-to-consumer alternative could plausibly replicate. The fans who were leaving cable were, the argument went, the marginal fans — the ones who did not watch enough games to justify the bundle anyway. The fans who actually loved the team would always pay for the bundle, because the bundle was where the games lived. The cord-cutters were leaving for reasons that had nothing to do with sports — they were leaving for Game of Thrones and House of Cards and Netflix originals. Sports was the anchor of the bundle. The anchor would hold.

Every line in that paragraph was true at the time it was spoken. The cable bundle was the largest revenue source. RSN economics did pay more per fan than DTC could match in the late 2010s. Many of the cord-cutters were leaving for reasons unrelated to sports. The structural arguments were directionally correct. The math worked, in the particular accounting frame the RSN executives were operating in.

Diamond Sports Group, the entity that held the largest portfolio of regional sports networks in the country, filed for Chapter 11 bankruptcy in March 2023. It re-emerged as FanDuel Sports Network in early 2025 after a rebrand via FanDuel naming rights, and is now in a second wind-down. Thirteen NBA teams and seven NHL teams and most of the affected MLB clubs spent late 2025 and early 2026 watching their primary local-rights infrastructure dissolve in real time, with most of them now scrambling to assemble local-rights stacks outside the cable bundle.

That is not what an industry that is losing the marginal fan looks like. That is what an industry that priced its model against the marginal fan and found out, too late, that the marginal fan was load-bearing looks like.

What the RSN Defense Got Wrong, Structurally

The RSN argument’s logical error was not in the math. The math was correct. The error was in the assumption that the categories were stable.

“Marginal fan” is a categorization that depends on what you compare it against. A fan who watches twenty games a year, in the era when the cable bundle is the only delivery system, is a marginal fan — relative to the fan who watches eighty. In a world where DTC streaming exists, that same twenty-game fan becomes the core of a different product. They are someone who would happily pay $15 a month for twenty games, who would not pay $90 a month for the bundle to access the same twenty games. The category of fan did not change. The product changed underneath the category.

When the product changes, the category re-prices.

The RSN executives who used the “marginal fan” frame were not wrong about the marginal fan. They were wrong about what would happen when a different product taught the marginal fan that they were actually a core customer of that product. Every cord-cutter who picked up an NBA League Pass, an MLB.tv subscription, an ESPN+ trial, a YouTube TV vMVPD package, was a fan who had been mis-categorized as marginal in the bundle frame. They were not low-value; they were waiting for a product priced for the way they actually wanted to consume.

The cable bundle did not lose its marginal fans. It lost the category — and discovered, in the loss, that the category had been propping up the entire pricing structure.

Why the AI Overviews Defense Is the Same Shape

Now read the AI Overviews defense again with that pattern in mind.

“The clicks AI Overviews replaced were bounce clicks. The user did not want them. The publisher did not monetize them. The clicks that survive are the high-value ones — the readers who actually wanted the article.”

The mathematical part is, as with the RSN defense, correct. There genuinely is a category of click that bounces. There genuinely is a category of reader that converts. The two categories are real, and they are different.

The structural error is in the assumption that the categories are stable.

What a “bounce click” is depends on what the alternative is. A click that bounces in a world where Google is the entry point to the entire web is a click that the publisher failed to retain. The publisher had the user on the page; the page did not deliver; the user left. That is a publisher problem. A click that bounces in a world where the user can get the answer without ever leaving Google is a click that the publisher never had access to. The user was never going to be on the page in the first place. That is a platform problem. The publisher’s job has changed without anyone telling the publisher.

The “high-value clicks remain” frame is structurally a “core fan stays in the bundle” frame. It is true at the moment it is spoken. It is true the year after. The thing it does not survive is the way the surviving cohort gets retrained.

A reader who used to find Wired through Google now finds the answer to their immediate question in the AI Overview, and never opens Wired again unless they happen to know it exists by name. The reader who does know Wired exists by name and goes there directly is the reader the bounce-click defense says is the real audience. That reader is also a much smaller cohort than the one Wired used to have, and that cohort is not generative — it does not get replenished by the next college student who Googled their way into the Wired universe in 2014, because that pathway is now answered by Google before it reaches the publisher.

What the Surviving Cohort Has To Do

The publishers that hold up against this — and there are some — share a pattern with the sports franchises that adapted earliest to the post-RSN world. They built a direct relationship with the audience that did not depend on the platform’s referral.

For the franchises, that has meant team-owned streaming products, free-streaming partnerships outside the cable bundle, and first-party data collection that the RSN never permitted. I wrote a piece a few weeks ago on the parallel mechanism in minor-league baseball, where free local streaming has quietly become the load-bearing layer. The structural shift is from “rented audience through a third-party pipe” to “owned audience through a first-party surface.”

For the publishers, the same shift looks like newsletters that subscribers actually open, paid memberships with retention metrics that make sense, podcasts whose listeners came from the host’s own audience, communities that exist in the publisher’s environment rather than someone else’s feed. None of that is novel. What is new is the realization that the publishers that did not start building the direct relationship five years ago are running out of time to start now.

The bounce-click defense is exactly the kind of argument that delays the start. If a publisher believes the clicks they are losing are the bad ones, the publisher does not feel the urgency to rebuild the audience model. The publisher waits for the AIO CTR recovery curve. The publisher trusts Google’s directional language about quality of clicks improving. The publisher reads the 2.4% AIO CTR figure as a turn rather than a ceiling. And the publisher loses another two years.

I do not think most publishers have those years.

What I Am Not Saying

I am not saying AI Overviews are a bad product for users. They are an excellent product for users. The compression of “look up a fact” into a single result is a real welfare gain, and the welfare gain is going to keep showing up in usage.

I am not saying Google is lying. The bounce-click argument is internally consistent. The CTR recovery is real. The category of valuable click that survives the Overview is a real category. None of that is in question.

I am saying that “true and operationally fatal” is a category that exists, and the AI Overviews defense — like the RSN defense before it — sits squarely inside that category. The defense is correct in its premises and wrong in its conclusion, and the wrongness shows up not in the next quarter but in the cohort math three to five years out.

The most useful thing I can offer, having watched the same shape of argument play out once already, is the observation that the inside of the cliff and the bottom of the cliff look the same in the accounting until they do not.

What I Would Do If I Ran a Publisher

I would not be optimizing for AI Overviews citations. I would not be trying to game the new SERP. I would not be spending the next twelve months in the SEO conference circuit waiting for a CTR recovery that is not coming back to its old shape.

I would be doing what the franchises that survived the RSN collapse did. I would be building the direct surface. I would be moving the audience off Google’s rails and onto a relationship the platform cannot disintermediate. I would be giving up some short-term traffic to invest in a model the next product cycle cannot collapse.

That is the part of the answer that none of the defenses, mine or Google’s, makes easier. Building a direct audience is harder than running an SEO factory. Running an SEO factory was supposed to be the business. The business changed.

The bounce-click defense is the sound of an industry that has not yet noticed that the business changed.

That sound is familiar. I have heard it before.