Getting Found Is the New Craft
5 min read
Building got cheap. Being found got expensive. In the zero-click era, getting found is the craft that decides the first year.
The most common failure mode of an early AI product is not churn. Churn would require users.
Watch how founders respond to that sentence and you can see the pattern that decides first years. The product is genuinely good. Onboarding is smooth, the pricing page answers every objection, the upgrade path is tested. Every surface a visitor could touch has been polished. Ask about visitors, and the answer is a roadmap: one more feature, one more redesign, one more month in the workshop. The machine gets another coat of lacquer. The street outside stays empty.
I have evaluated hundreds of startups over 26 years, first from inside Fortune 500 innovation programs, now at Aberlay, where I advise AI-first founders and put skin in the game when conviction is total. The companies that died rarely died of bad product. They died unvisited.
Polish is the comfortable procrastination
Pieter Levels named the pattern in June: indie builders assembling elaborate AI factories, agent pipelines generating code and landing pages and dashboards, while almost none of them have customers, revenue, or traffic. Marketing waits until the super system is finished. The super system is never finished.
The bias has a structure. Product work is measurable, private, and always yields a visible improvement. Distribution work is public, repetitive, and most days yields nothing. One feels like craft. The other feels like standing in the street handing out flyers. So the roadmap fills with craft, and the calendar quietly runs out.
For a decade that bias was survivable, because attention was cheap and search sent clicks. Rank or pay for traffic, optimize the funnel, and arithmetic did the rest. An entire generation of founders was trained on that arithmetic, raised on conversion rates and A/B tests, taught that the funnel is the business. The funnel made sense as the center of gravity in a world where the top of it filled by default.
That world is gone. The belief survived it.
The doors closed while everyone was building
The numbers are stark. In SparkToro’s June 2026 measurement, 68 percent of US Google searches between January and April 2026 ended without a single click. Of every 1,000 searches, 276 clicks now reach the open web, down from 374 two years earlier, a 26 percent drop. AI Overviews appear on more than a fifth of searches and cut click-through by roughly 60 percent when they do. The direction is not seriously in dispute: in sworn testimony in May 2025, Apple’s services chief said Google searches in Safari had fallen for the first time in 22 years, and Alphabet lost about seven percent of its market value that day.
The attention did not evaporate. It moved behind model answers. ChatGPT alone fields more than 2.5 billion prompts a day. And what the new layer does send is not junk traffic: Adobe’s retail panel, one of the largest first-party measurements of AI-referred commerce, tracked AI-referred visits up nearly 700 percent year over year through the 2025 holidays, converting 31 percent better than other sources.
Discovery did not shrink. It changed owners. The founders still optimizing pages for a stream of clicks that no longer arrives are tuning an engine that is not connected to the wheels.
The repricing already happened
If you want to know where a constraint sits, watch what acquirers pay for. In November 2025, Adobe agreed to pay $1.9 billion in cash for Semrush and said plainly what it was buying: generative engine optimization, visibility inside AI answers. Three months later Profound, a two-year-old company that tells brands how models see them, raised at a $1 billion valuation with more than a tenth of the Fortune 500 already paying for the answer.
Nobody pays those prices for a building problem. Building got cheap. Being found got expensive. The market has already marked the constraint to market; most roadmaps have not.
What the disciplined founder does differently
Treat distribution as an engineering discipline, not a mood. It gets what the codebase gets: a fixed cadence that ships every day whether or not inspiration arrives, instrumentation that says what moved, and kill criteria that retire channels the way you retire dead code. The founders who compound are not louder. They are more regular.
Respect the arithmetic. Conversion multiplies arrival. Improving the conversion of an empty street multiplies zero. Until arrival is a solved, boring, recurring system, every additional hour of funnel polish is spent on the wrong factor of the equation.
Win intent, not volume. The clicks that survive the zero-click shift are branded, local, and high-intent. The broad query is gone; the specific one still lands. Owning one narrow intent completely now beats ranking loosely for a hundred, because the narrow intent is where both the remaining clicks and the model citations concentrate.
And make the product legible to machines, because the next visitor evaluating it is as likely to be a model as a person. Public pricing, structured documentation, claims a system can check. The payment networks are ahead of most founders here: card networks now score merchant sites for whether agents can navigate and complete tasks on them, and maintain directories of verified participants. The storefront of the next decade is machine-readable, and the brands that agents can parse and verify will simply exist more than the ones they cannot.
None of this argues for building less well. The machine still matters; it decides what happens after arrival. But nothing inside the machine decides arrival, and arrival is where the first year is won or lost.
The machine converts what arrives. Arrival is the product now.