Your Next Customer Will Never See Your Website

The number that got my attention came from Rand Fishkin, the guy who founded Moz and has spent twenty years inside search data. His company SparkToro pulled clickstream data from Similarweb for the first four months of 2026: 68 percent of Google searches in the US ended without a click. Not without a click to your site. Without a click to anything. No website, no ad, not even Google's own pages. The person typed something in, got an answer on the results page itself, and left.

In 2019 that number was about 50 percent. The trend line only goes one direction.

I run an AI consultancy in Dallas. Most of my clients are home services companies: HVAC, plumbing, roofing, the trades. And I can tell you what happens when I put that stat in front of an owner. First they don't believe it. Then they check their own Google Analytics, see sessions drifting down while their Google Business Profile calls hold steady or climb, and go quiet for a second. That quiet moment is the sound of somebody recalculating.

Here's the part I want to walk through carefully, because there's a lot of breathless nonsense written about this and I'd rather give you the straight version.

What actually changed

For a decade, local service marketing ran on one loop. Customer has a problem, customer searches, Google shows your website, your website converts them into a lead. Everything we all did, the landing pages, the chat widgets, the sticky mobile headers, the conversion rate obsession, existed to squeeze more leads out of that loop.

Two things broke the loop at the same time.

First, Google started answering questions itself. Featured snippets, knowledge panels, People Also Ask boxes, and now AI Overviews, which show up on roughly a fifth of all searches and, when they appear, cut click-through by about 60 percent according to Semrush's 2025 data. When Google showed its own AI Mode results to users, 93 percent of those sessions ended without a click. Google is not sending people to your website anymore. It is keeping them.

Second, and this is the bigger one that most owners haven't felt yet, people started asking AI assistants instead of searching at all. ChatGPT has hundreds of millions of users. Gartner put out a press release in February 2024 predicting traditional search volume would drop 25 percent by 2026 due to "AI chatbots and other virtual agents." Plenty of people argued with the number. Google's response to the argument tells you who won: they replaced the button in their own search box. Type something into Google now and what you're invited to press is AI Mode, which routes you to Gemini. That's not a prediction. That's the product.

Sundar Pichai has said publicly that 2027 is the year of the agents. I'll come back to what that word means, because it's the part that actually matters for a plumbing company in McKinney.

What it means for a ten-truck operation

Let me make this concrete, because "zero-click search" as a phrase is abstract and I hate abstract.

A homeowner's AC dies at 6pm on a July Saturday in Frisco. Here's the old journey. She Googles "ac repair near me." She gets three ads, a map pack, and some blog posts. She reads maybe two of them, picks whoever looks legit, calls, and gets voicemail because it's Saturday. Maybe she leaves a message. Maybe she calls the next company instead. Your entire marketing spend existed to win that moment: the ranking, the reviews, the website that loaded fast on her phone.

Here's the journey that's forming right now. She tells her phone, or ChatGPT, or Gemini, "my upstairs is hot and the AC fan runs but no cold air is coming out." The AI walks her through it. Bad capacitor or a refrigerant leak, here's what each means, here's what a fair price looks like in Collin County, here's how to tell an honest company from one that will sell her a compressor she doesn't need. Then she says "find me someone." And the AI recommends, maybe two or three businesses, maybe just one.

Sit with the difference. In the old journey, she visited five websites and you had five chances to win her. In the new one, you get zero visits and you're either on the shortlist or you don't exist. There is no second page of an AI answer. There's no "scroll past the ads."

Now, the honest caveat, because I promised you the straight version: agents doing the whole job, booking the appointment, negotiating the slot, showing up in her calendar, are early. When I say the journey is "forming," I mean exactly that. ChatGPT can already buy things through Stripe's agentic commerce protocol. Google's Gemini Spark is a 24/7 personal agent that books restaurant reservations through OpenTable and is rolling out to subscribers now. Google, Microsoft, OpenAI, and more than 250 other organizations have joined an open protocol literally called A2A, agent to agent, so that your customer's agent can talk to a business's agent. The plumbing is being installed while we argue about whether the water is coming.

What's already true today, not 2027, today: the recommendation step. People ask ChatGPT and Gemini and Perplexity "who's the best plumber near me" right now, and those answers get formed from what the models can find and trust about you. That's the game as it stands.

Why "just do good work" won't save you

Every time I explain this, some version of this pushback comes up: my reviews are great, my reputation speaks for itself. And it's half right. Reputation is exactly what the AI systems are trying to evaluate. But they can't evaluate what they can't read.

Here's an example. A few months back I ran a test, asking ChatGPT for price expectations on a water heater replacement in the DFW area. The answer it gave was generic and honestly kind of useless, because almost no local plumbing company publishes pricing content. Not exact price books. Just honest ranges, what pushes a quote up, what pulls it down. The model had nothing local to work with, so it defaulted to national averages and hedging. Every plumber in North Texas had the same opportunity to be the source that answer cited. Essentially none of them were.

That's the quiet tragedy of this shift for our trades. The knowledge exists. It lives in the heads of estimators and in ten thousand phone calls about capacitors and expansion tanks and whether that stain on the ceiling is a roof problem or a supply line. And almost none of it is written down anywhere a machine can find it.

There's a second problem nobody talks about. Google put out guidance in 2025 on what they want to see from content in the AI search era, and they said it plainly: commodity content, generic listicles any chatbot could write, will not just be ignored, it will hurt your brand. What they reward is hyper-specific, real material. The catch, and every agency owner who's tried to extract war stories from a field crew knows this catch, is that the real material is the hardest to collect. Your tech doesn't want to write a blog post after a ten-hour August day. That's not a character flaw, that's physics.

Which is why, when I started building this stuff for clients, I stopped thinking about content and started thinking about capture. Every phone call your company answers already contains the content. The question, the price discussion, the "should I fix the section or replace the whole roof." If you can capture those conversations and publish the answers, you're generating material no competitor can copy, because they weren't on your calls. I use AI voice recorders and call transcription for exactly this, and I'll write more about the field-capture side separately, because it deserves its own piece.

The list I actually give owners

When an HVAC or roofing client asks me what to do this quarter, I keep it short, because the honest answer is that most of this is not complicated. It's just unfamiliar.

Publish your pricing. Ranges, not a price book. Low end, high end, and what moves the number. People are asking AI what things cost right now, in numbers that would surprise you, and AI can only cite who published. The first honest company in a market to publish pricing becomes the default source for every model answering that question. You will hate this idea for a week and then you'll notice your competitor's truck at the house where your estimate never even got called.

Make sure machines can read you. There are now standard files, agent-card.json, llms.txt, ai-catalog.json, that let AI systems discover what your business does, where it works, and what it sells — see the three files explained. Almost no small business has them. I built a free tool that generates all three for home service trades, it's on this site — Agent Stack Builder — and it takes about two minutes. This isn't a gimmick, it's the same "get listed where customers look" work you've always done, except where customers look is changing.

Start capturing your real conversations. Record walk-throughs, transcribe calls, turn the questions into published answers. This is the moat. Generic content is free for your competitor to generate. Your calls are not.

Fix your booking path. Online scheduling that an AI can eventually transact with. The major field service CRMs, Jobber included, are building agent booking integrations right now. If your scheduler is a phone and a voicemail box, you're handing the agentic era to whoever isn't.

Track your AI share of voice. Ask ChatGPT who's the best at your trade in your city every couple of weeks. Screenshot it. Ask Gemini too. If you're not showing up, you now know, and you know what to fix, because the inputs are the items above.

The part worth being honest about

I don't want to sell you a panic. Websites are not going to zero tomorrow. Plenty of my clients still get solid leads from organic search today, and they will next year. The world doesn't flip like a switch, it drains like a bathtub, and the drain is already open, 50 percent zero-click in 2019, 60 in 2024, 68 now.

But here's what I'd say if you and I were sitting across a table. Every platform shift pays the people who moved early, and it bills the people who moved late. The businesses that were doing local SEO in 2012, before their competitors believed in it, printed money for a decade. The ones that started in 2019 paid for every year of delay with rankings they never got back. The agents are the next one of those. Satya Nadella stood on stage at Microsoft Build and said "we've entered the era of AI agents," Bill Gates wrote that agents are the next platform, and Marc Benioff is out there calling it a three-to-twelve-trillion-dollar shift. When the people who built the last two eras agree this loudly about the next one, the debate about "if" is over. The only question left is whether you're early or late.

The data says late is expensive. The tooling to be early is sitting right there — the Knowledge Catalog Building service handles it end to end. I'd start this quarter.

Nathan Brown runs HiVergent AI in Dallas, TX. He builds AI phone agents and agent-readiness systems for small and home services businesses. Try the free Agent Stack Builder linked below, or book a free AI readiness assessment.