Where We’re Going, We Don’t Need Roads

THE NUMBER: 43%. That’s the share of Google searches that now resolve with an AI Overview right there on the page — no click, no visit, no trip to anyone’s website — up from 15% a year ago, per Similarweb’s 2026 report. Google’s AI Mode went from 126 million visits last June to 279 million by May and crossed a billion monthly users. Hold that 43% next to a second number that looks like its opposite: Alphabet’s Search revenue rose 17% last quarter to $63.3 billion. One of those numbers says the business is dying and the other says it’s thriving. They’re the same story. The trick is knowing which one is the tell.

At the end of Back to the Future, Doc Brown fires up the DeLorean in the McFly driveway, Marty says they don’t have enough road to hit 88, and Doc pushes those gullwing doors down with the best last line in the movie: “Roads? Where we’re going, we don’t need roads.” Then the car lifts off and the credits roll.

Keep that image, because it’s the whole issue. For twenty-five years the roads were the ten blue links. You had a question, Google laid down a path of clickable destinations, and you drove it — to a recipe site, a review, a store, a guy’s blog about golf clubs. Every mile of that road had a toll booth, and every toll paid somebody. That road is what’s disappearing. And the only businesses that make it to the other side are the ones who figured out how to fly.

🛣️ The Toll Booth That Ran the Web

Start with what Google actually was, because we’ve all gotten sloppy about it. Google didn’t sell ads. Newspapers sold ads. Google sold intent — the purest thing a business can buy. When you typed “best mirrorless camera under $1,500,” you weren’t a demographic Google was guessing at. You were a person announcing, in your own words, exactly what you wanted to buy at the exact second you wanted to buy it. Google auctioned the answer to that announcement to the highest bidder and took a cut of the handoff. Bill Gross figured out the model at GoTo.com a quarter century ago; Google perfected it into what is, by operating margin, one of the best businesses the world has ever produced. Buffett called not owning it one of his bigger mistakes.

And here’s the part everyone forgets: difficulty was the feature. A simple query — bakery hours, the address of the golf course you’re playing Friday, where The Odyssey is showing near you — cost you one click and Google one ad impression. But a hard query, the kind with real money behind it, didn’t resolve in one link. “Should I move to milled wedges at my handicap if I’m playing elastomer faces on the long irons” sent you back to the search box five, six times, because the answer lived scattered across five different links and you had to assemble it yourself. One search, six clicks, six impressions. The harder you thought, the more Google got paid. The company had built a toll road where the difficult trips were the profitable ones. That is a machine of terrifying elegance, and it ran the commercial internet for two decades.

🔀 The Pincer

Now watch the answer engine take that machine apart from both ends at once.

The simple query goes first, and it’s already mostly gone. When you ask Google what time the bakery closes, it no longer hands you a link to the bakery’s page. It reads the page for you and prints the hours in an AI Overview at the top. You got your answer. The bakery got nothing — no visit, no chance to show you the new croissant, no data that you were even looking. Similarweb’s 43% is this, at scale: nearly half of all searches now end on Google’s results page. The toll booth is still there. The road behind it goes nowhere.

The complex query is the one that should scare Mountain View, because it doesn’t just lose the click. It leaves. When I have a real question now, say the wedges, or a tax wrinkle, or how to think about a business decision, I don’t go to Google and click six links. I open a $20-a-month research assistant and ask it, and it does the work those six links used to do and organizes it into one clean narrative. I don’t want a reference to a website. I want an answer from a system that already read every relevant website so I don’t have to. And that system — ChatGPT, Gemini, Claude, Grok, some open-weight model running who-knows-where — is built by design to never send me anywhere. Referring me out would be admitting it failed.

Sit with the inversion, because it’s the cruel center of this whole thing. Complexity used to be Google’s cash register — the hard questions minted the most clicks. Now complexity is the LLM’s cash register — the hard questions are exactly the ones worth paying a subscription to collapse into a single answer. The trait that made Google the best business ever built is the same trait that now routes its most valuable customers straight out the door. The pincer isn’t symmetric. It’s worst precisely where the money was best.

📉 The Best Business Ever Built, Giving It Away

So put yourself in Sundar Pichai’s chair, because this is a genuine tragedy in the classical sense — a great figure undone by its own defining strength.

Google monetized intent better than any company in history. AI now forces it to give that intent away in three flavors, none of which pay like the old machine did. The simple intent, it hands over free — here, take the answer, no charge. The complex intent, the crown jewel, it can only hope to be chosen to fulfill, as one $20 assistant among four, where the same query used to pay it automatically on all six clicks. And to stay in the conversation at all, it’s giving away an open-source version — Gemma just crossed 900 million downloads, with Pichai posting “1B next” like it’s a win, which it is, except a download isn’t a click and a free model isn’t an auction. Three roads out of the old business, and every one of them tolls at a fraction of what the blue links did.

Now add what it costs to even play. To be the chosen research assistant, Google has to spend hundreds of billions on data centers and chips — yes, it designs its own TPUs, but designing them and buying the fabs, the power, the land, and the networking to run them at planetary scale is a capital bonfire either way. It’s spending that money against OpenAI and Anthropic and xAI, companies with no legacy margin to protect and a stated willingness to light investor billions on fire for a foothold. Google is being asked to disrupt a search business that throws off something close to a 70% operating margin and replace it with a capital-intensive, lower-margin, maybe-you-get-picked business — while the insurgents have nothing to cannibalize and nothing to lose. That is the Innovator’s Dilemma in its purest published form, and we flagged the same knife on July 24 in The Money’s Not Here: Google’s winning as a fund and stumbling as a builder, its record profit a piece of misdirection over the fact that it has to eat the greatest business ever built before someone else does.

Which is why that 17% revenue “growth” is the tell, not the reassurance. Where an AI Overview appears, click-through to the underlying site is down better than half, while the price advertisers pay per click keeps climbing through 2026. Read that carefully: Google is charging more for an inventory that delivers less. That’s not a healthy market clearing. That’s the concession stand jacking up beer prices while the crowd files toward the exits. We’ve seen this movie — it’s the late innings of every supercycle, Lucent booking bigger sales to customers it was quietly financing, right up until it wasn’t. Rising revenue on collapsing delivery is what the milking phase looks like from the outside. The number going up is not the same as the machine being healthy.

🍽️ The Beast Eats the Farmers

And there’s a back wall to this room that makes “the click is dying” more than a marketing problem. The answer engine only knows what the web taught it. Every crisp synthesis Gemini or ChatGPT hands you is a distillation of the six links it ate — links written by people who used to get paid in clicks for writing them.

Take away the clicks and you take away the reason to write. No traffic, no ad revenue, no affiliate cut, no reason for the bakery or the reviewer or the golf-forum obsessive to publish the thing the model needs to stay current. The machine that killed the click still needs a live click economy to have anything fresh to synthesize. Nobody — not Google, not OpenAI, not the open-weight crowd — has solved who feeds the beast once it’s finished eating the farmers. The pincer doesn’t just close on the publishers. Eventually it closes on the answer engines too, and they know it. That’s why you’re starting to see the licensing deals and the “we’ll cite our sources” pledges: not generosity, but a dawning realization that you can’t strip-mine the field forever and still expect a harvest.

📺 If I Ran Mountain View

Here’s where it gets interesting, and where Google’s tragedy might have a third act. There’s one surface the answer engine hasn’t cracked, and Google happens to own it outright.

YouTube. When you want to actually see how to reface a cabinet or fix a putting grip or install a dishwasher, there is no AI system that replaces it, because “answer” and “video” haven’t fused yet. You still go to YouTube directly, by name, as a verb — which is exactly the escape hatch we’re about to prescribe for everyone else. It’s a noun, not a ranking, and no Overview sits between it and its audience.

But the real prize isn’t defending YouTube. It’s building the AI version of it — an answer engine that assembles a video reply on the fly, edited in real time from every clip ever uploaded, narrated, personalized, exactly as long as your question requires. Whoever ships that owns a brand-new intent surface, one that’s far more monetizable than a text answer because video is where the ad dollars have always been richest. And here’s the thing: Google is the only company on earth positioned to build it. It owns the video corpus, the distribution, a billion-plus logged-in users, and the ad machine to charge for it. If I ran Mountain View, that would be the Manhattan Project and everything else would be a rear-guard action. The company that got disrupted out of the text-answer business could reinvent itself as the first to monetize the video answer — the same move it pulled going from search to YouTube the first time. The tragedy has a possible fourth act. It just requires them to cannibalize themselves faster than the insurgents can.

What This Means For You

You don’t run Google. You run a business that, until recently, got found because someone searched and clicked. So take the collapse of the road seriously and start building wings.

Pull your traffic sources today and mark what lives or dies on a Google click. Split your inbound into “relationship we own” — email, direct, app opens, referrals from actual fans — and “rented from search.” That rented number is your exposure to the 43%, and it’s being cancelled a point at a time whether or not anyone sends you a notice. If it’s north of half, you don’t have a marketing plan. You have a lease on a road that’s being torn up.

Stop optimizing for a rank the answer now sits on top of, and become the noun. The brands that fly over the pincer are the ones people open by reflex — TripAdvisor for a hotel, Wirecutter for a toaster, Google Flights for a fare, YouTube for a how-to. Nobody searches their way to those. They’re destinations, and no answer engine gets to stand in the doorway. Pick the one question your business should own the answer to, write the definitive version only you can write, and publish it where you keep the audience, not where you rent the click.

Own a pipe the synthesis engine can’t intermediate. Your list, your feed, your community, your channel — the places where you reach people without asking a gatekeeper for permission and without hoping to be cited. One owned subscriber is worth ten rented impressions that die on the results page. This is the same move on offense and defense: replace discovery you rent from a dying road with a relationship you own outright.

Three Questions We Think You Should Be Asking Yourself

If Google sent me zero clicks next year, what happens to my business? Not a hypothetical anymore — it’s the 43% trend line extended a few quarters. If the honest answer is “I’d be in serious trouble,” you’ve found the single most important project on your desk, and it’s not an SEO project.

Am I structured to be the answer, or just to be a link? For the queries you can’t escape, being cited inside the Overview beats being the eleventh blue link nobody scrolls to. That means clear, authoritative, well-structured content a machine can lift cleanly. But understand that being cited is a consolation prize — it drives far less than a click did. The real game is above it.

What is the one thing people would come to me for, by name, with no search box in between? If you can answer that cleanly, you have a destination and a future. If you can’t, you have a rank, and the road under it is disappearing. Building that answer is the work of the next two years.

Roads? Where we’re going, we don’t need roads.”
— Dr. Emmett Brown, Back to the Future (1985)

— Harry and Anthony

Signal/Noise by CO/AI is published most weeknights from New Canaan, Connecticut. The point is to make you the smartest person in the room without taking more than fifteen minutes of your morning. If we did, forward it to one person. If we didn’t, hit reply and tell us why.

Sources

• Google is Making the Answer the Default — Shelly Palmer, Jul 28, 2026 (Similarweb 2026 report: AI Overviews on 43% of searches, up from 15%; AI Mode 126M→279M visits, 1B+ MAU; “share of prompt”)
• Alphabet Q2 2026 results — Google Search & Other revenue $63.271B, +17% — Alphabet, Jul 2026
• Google AI Overviews & organic CTR in 2026 — SEO industry data, Jul 2026 (click-through down ~50%+ where an Overview appears; rising CPCs through 2026)
• Sundar Pichai on X, Jul 25, 2026 — Gemma family past 900M downloads (“1B next”); Google Cloud ~$100B run-rate, ~40% of Google’s search business (repost of @chetanp)
• Tons of People’s Claude Chats Are Exposed on Google — 404 Media, Jul 2026 (the answer layer absorbing private content)
• CO/AI prior issue this builds on: The Money’s Not Here (Jul 24 — Google winning as a fund, stumbling as a builder; the Innovator’s Dilemma in Search) (verify slug on live index before publish)
• CO/AI prior issue: Your Company Needs a Harness, Not an Upgraded Chatbot (May 28)
• Back to the Future (1985) — Dr. Emmett Brown, the DeLorean, and the road you don’t need