On May 19, Google announced a new AI Search experience and called the redesigned Search box its biggest upgrade in over 25 years.

The new Search accepts text, images, files, videos, and browser tabs as input. It builds custom interfaces on the fly: tables, graphs, simulations, interactive tools, all assembled in real time by Gemini 3.5 Flash.

It deploys information agents that run in the background 24/7, scanning the web on your behalf and pushing synthesised updates when something relevant surfaces. It expands agentic booking, too: you describe what you want — a private karaoke room for six on a Friday night that serves food late — and Search brings together pricing, availability, and direct links to finish booking through the provider. In some categories, Google can even call businesses on your behalf.

And it ships a Universal Cart for shopping across merchants and Google services.

The scale is hard to overstate.

AI Overviews now reach over 2.5 billion monthly active users. AI Mode, launched just a year ago, has surpassed 1 billion monthly active users, with queries more than doubling every quarter. The Gemini app itself has surpassed 900 million monthly active users — up from 400 million at last year's I/O. Google now processes over 3.2 quadrillion tokens per month across its surfaces, a sevenfold increase year over year.

For context, ChatGPT sits at more than 900 million weekly active users.

Google is not losing the AI race. It is running it on a different track: the one where 2.5 billion people already show up every month out of habit.

The consequences are not theoretical. They are measurable, and they are accelerating.

That last row is worth pausing on.

Gartner made that prediction in February 2024. At the time, it felt aggressive. Looking at the numbers above, it now looks conservative.

Individual cases tell the story more sharply.

HubSpot will be fine. HubSpot is a CRM, not a blog. The content marketing machine that made them famous was an acquisition channel, not the business itself. The business is the product people pay for and log into every day.

That distinction matters. We will come back to it.

Chegg was a publicly traded online education platform where students paid for homework help, study guides, and tutoring. It peaked at a $14 billion market cap in 2021. Then AI learned to answer homework questions. The content was the product, and AI Overviews can now deliver it for free without sending anyone to Chegg at all.

The stock is down roughly 99 percent from its highs. Whatever exact market cap you use on publication day, what remains is a fraction of the company it used to be.

News publishers are seeing the same pattern. Some have documented drops as steep as 89 percent on queries where AI Overviews appear. Helen Havlak, publisher of The Verge, called it an "extinction-level event" that has already put smaller publishers out of business.

SISTRIX's 59 percent CTR collapse at position one deserves its own line:

And that was before the I/O announcements.

The generative UI features, always-on agents, agentic booking, and custom mini-apps built inside Search all reduce the need to click through to a source.

Here is what I think happened, stripped of the SEO jargon.

For twenty years, Google sent traffic to the open web.

Not out of generosity. Because it needed the web.

Google's product was the organization of other people's information. The more useful the web was, the more useful Google was. Websites created content, Google indexed it, users clicked through, and everyone benefited.

Google no longer needs to send you to a website to answer your question. It can synthesise the answer itself, build an interactive interface around it, and handle the transaction without you ever leaving.

The shift did not happen overnight, but the I/O 2026 announcements make the architecture explicit: Search is becoming an AI operating surface, not a results page.

If this sounds familiar, it should.

Last week I wrote about the subsidy playbook: how companies burn cash on below-cost products to capture a market, then monetize once the competition is dead. Google is running the same play, right now, on itself.

Every AI Overview, every generative UI, every 24/7 agent, every custom mini-app built on the fly — that is a cost center. It destroys the click-through model that funds Google's $77 billion quarterly ad revenue.

Google is subsidising the destruction of its own business model.

And unlike most companies running a subsidy play, Google is rich enough to subsidize itself.

From a game theory perspective, there is no other move. If Google does not cannibalize its own search, OpenAI, Anthropic, and Perplexity will do it for them. Users migrate to AI-first interfaces, Google loses the traffic anyway — but slower, and without owning the replacement.

In every scenario where AI answers become viable, self-cannibalization is the dominant strategy. There is no equilibrium where the old model survives.

It has been done before.

Netflix killed its own DVD-by-mail business — at the time, its entire revenue — to become a streaming company. Reed Hastings kicked the DVD executives out of the main management meeting because they were not adding value to the conversation about where the company needed to go.

The DVD team was responsible for all the revenue and all the profit, and they were still shown the door.

Netflix went from a mail logistics operation to a $200 billion streaming platform. Blockbuster, which could not let go, went to zero.

That is what Google is doing right now: the incumbent that sees the disruption coming and kills itself to become its own successor.

Here is the part nobody is talking about enough: the ad model may not need to disappear. It may need to adapt to an interface where answers, recommendations, and transactions all happen in the same place.

In the old Search, you bought a keyword and got a labeled ad placement. The user knew it was an ad.

In the new Search, Google is already testing Gemini-built ad formats in AI Mode and AI-powered Shopping ads that sit much closer to the recommendation layer. Google says these formats will continue to be clearly labeled as "Sponsored," which matters: there is no public evidence that Google is secretly turning paid advertisers into organic-looking AI citations.

But the direction is still important. Ads no longer need to sit beside the answer. They can become part of the conversational flow around the answer.

The auction does not necessarily become invisible. It moves closer to the moment of decision.

The broader bet is that once Search owns more of the transaction — shopping via Universal Cart, booking via agentic search, services via mini-apps — monetisation can shift from pay per click toward pay per action, lead, or transaction.

Their content trained the models. Their traffic funded the transition. And now their users are being absorbed into a closed loop where Google handles discovery, comparison, and purchase without ever sending anyone out.

Every dominant platform eventually runs the same playbook:

Zynga built its business on Facebook's platform in 2009–2012: free distribution, viral mechanics, hundreds of millions of users. Then Facebook changed its feed algorithm, and Zynga's stock declined 80 percent from its peak.

The game had not changed. The landlord had.

Restaurants that built their customer base through delivery apps discovered the same thing. DoorDash and Uber Eats owned the customer relationship, the data, and the margin. The restaurant owned the kitchen.

Musicians with millions of Spotify streams cannot email their own listeners to sell a concert ticket.

Google's version of this was free traffic.

Billions of clicks a month, delivered at no cost, to anyone who could rank. It was the greatest demand-generation subsidy in the history of business.

And like every subsidy, it was temporary.

The businesses that understood this used the window to build something that did not depend on it. The rest kept optimising for the algorithm and called it a strategy.

Which brings me to what I did — and, more importantly, why.

I built ELECTE's newsletter early.

I want to be honest about why: it was not because I predicted that Google would one day cannibalise the open web. I had no thesis about AI Overviews in 2023. I did not run a scenario analysis on zero-click search trends.

I built a newsletter because talking directly to the people who cared about what we were doing seemed like the obvious thing to do.

It was not strategic foresight. It was, in retrospect, a side effect of doing something right.

It is a direct channel to people who chose to hear from us. No algorithm. No intermediary. No one between us and the inbox.

When Google's traffic numbers started collapsing across the industry, it did not collapse for us, because our acquisition model was never built on Google in the first place.

I did not see this coming. I got lucky.

But the luck had a structure to it.

They are the ones that were already doing the right thing before the disruption made it necessary.

The company that invests in direct relationships with its customers does not do it because it foresees a specific platform collapse. It does it because owning the relationship is better than renting it.

That is true whether Google thrives or dies, whether Facebook sends traffic or does not, whether interest rates are zero or five.

The company that builds a product people actually log into every day does not do it as a hedge against algorithm changes. It does it because that is what a good product looks like.

The company that invests in brand — in being the name people type into the search bar rather than the result they scroll past — does not do it because of an analysis of zero-click trends.

It does it because being known is better than being found.

None of these decisions require a prediction.

They require clarity about what matters.

Thousands of businesses are now scrambling to "adapt to AI search."

They are hiring consultants. They are restructuring their content strategies. They are learning new acronyms — AEO, GEO — and optimising for a new set of rules they do not control and did not write.

They are, in other words, doing exactly what got them here.

The problem was never Google's algorithm.

The problem was building an entire acquisition model on a channel someone else owns.

The specific mechanism — whether it is an AI Overview, a core update, or a complete platform redesign — does not matter. If you do not control the channel, the channel will eventually be taken from you.

The only question was when.

The businesses now asking "how do we adapt to AI search?" are asking the wrong question.

The right question is:

This is what I did. None of it was clever. You do not need a consultant for this.

If more than half of your pipeline comes from a channel you do not control, you do not have a growth strategy. You have a vulnerability with good metrics.

An email address. A subscription. A login. If the only way to reach your customers is by paying Google or Meta, you are renting your own audience.

A product people log into, a tool they rely on, a community they belong to - those are not summarisable by an AI Overview. HubSpot lost its blog traffic and barely flinched. The blog was the funnel. The CRM was the business.

What replaces algorithmically distributed traffic is not a new algorithm to optimize for. It is the old, boring work of building direct relationships with people who care about what you do.

If you do not have a good answer to any of the above, the algorithm is not your problem.

Fabio Lauria

CEO & Founder, ELECTE

Every week, we explore AI without the hype—using data, analysis, and an independent perspective.

Ps. That was always the real strategy. Most businesses just never needed it badly enough to do it.The competence was never in the code. It was in knowing what to build with it.