At Google I/O 2026, Sundar Pichai announced that AI Mode in Search had crossed 1 billion monthly users. Queries have more than doubled every quarter since launch. People are asking questions that are 3x longer than traditional searches. Follow-up questions are up 40% per month in the US. By every metric Google tracks, the product is working.
Here's the other number: 92–94% of searches in AI Mode end without a click — Semrush measured it across 69 million sessions.
That's not a bug. That's the product.
What Actually Changed at I/O
AI Mode is no longer experimental. Gemini 3.5 Flash is now the global default for AI answers in Search. AI Overviews — the box that answers your question before listing any links — now appears on roughly 48% of tracked queries and serves 2.5 billion monthly active users. It spans 200+ countries and 40+ languages. This isn't a test anymore.
The product demo was genuinely impressive. You can ask a complicated question, get a structured answer, follow up with a more specific question, and keep going without ever reformulating from scratch. Google is betting that this is what search should have always been — a conversation, not a keyword box.
The traffic data is telling a different story elsewhere.
HubSpot, one of the largest content marketing companies in the world, lost somewhere between 70 and 80% of its organic traffic between November 2024 and Q2 2025. CEO Yamini Rangan said it plainly on the Q1 2025 earnings call: "AI Overviews are giving answers, and fewer people are clicking through to websites." Business Insider lost 55%. Global publishers saw Google traffic drop roughly one-third in 2025. A Seer Interactive analysis measured organic click-through rates dropping 61% on queries where AI Overviews appeared.
Sixty percent of all Google searches now end without a click. In AI Mode specifically, it's more than nine in ten.
The Deal, and What Broke It
For most of the internet's history, the arrangement was simple: you make content, Google indexes it, users click through to you, you get traffic. That traffic pays the bills — ads, subscriptions, leads — so the incentive to keep making content exists. Imperfect, but functional.
AI Overviews broke step three. The AI reads your content, assembles an answer, delivers it directly. The user is satisfied. The publisher gets nothing. And the AI's answer is probably pretty good, which makes it hard to argue with from the user's perspective.
It's a bit like a library that used to send visitors to your bookshop for the books they wanted. And now prints a summary at the front desk.
The reader is happy. The bookshop is not.
The thing that doesn't sit right with me: AI Overviews are built from content. That content was made by people who had an economic reason to make it — because search traffic existed and paid. If that traffic disappears, the incentive to make new content shrinks. New content dries up. AI answers start citing older and older material. Eventually you're synthesizing a web that stopped growing.
Nobody at the I/O keynote mentioned this.
The Take
I'll be honest about where I'm standing when I say this. I'm a year into writing consistently on the internet. I have a portfolio site. I've been reading about SEO, about how to structure content so search engines can understand it, about building domain authority slowly. Everything I read assumed that if you make something worth reading, Google will send people to it.
Google I/O 2026 made me feel like I walked into a cooking class for a restaurant that had just closed.
The industry response has a name now: Answer Engine Optimization — AEO. The pitch is that instead of optimizing for Google to rank you, you optimize to be cited inside an AI answer. Structure content into 200–400 word self-contained sections. Answer questions directly. Be authoritative enough that the model includes you rather than a competitor.
I want to be honest about what I don't know: nobody has really proven this works yet. The people writing AEO playbooks are, many of them, the same people who wrote SEO playbooks — and some of them are improvising as they go. The stat that "brands cited in AI Overviews earn 35% more clicks than uncited brands" sounds encouraging until you realize AI Mode users still don't click more than nine times in ten. Getting cited means your information travels. You might not.
And I should grant the other side its due: the content SEO maximally incentivized at full optimization was not always content worth reading. Content farms, listicles padded to 2,000 words, exact-match keyword stuffing — that ecosystem was genuinely bad. If AI Overviews defund the garbage, that's a real improvement.
But the personal-finance explainers, the independent journalism, the careful how-to writeups from people who figured something out — that content was funded by traffic too. The stuff worth keeping was inside the same economic model as the stuff worth losing.
The uncomfortable thing is that Google has 2.5 billion people using AI Overviews. That's a product decision that reshapes the economics of almost everything people publish online, made by one company, with no real mechanism for the web to push back. I'm not saying it was the wrong call. I'm saying it's a big one, and the announcement at I/O felt more like a product launch than a reckoning.
Maybe the web reorganizes. Maybe newsletters and paywalls and direct distribution replace what search traffic used to fund. Maybe AEO eventually works and the new game has rules.
Quiet Part
The funny thing is that this post might still get found through Google.
Or it might not. An AI might summarize it for someone who never visits.
I'm not sure the distinction matters the way it used to. And I think that's kind of the whole point.