The Day the Internet Stopped Being a Library

In April of this year, an attorney I do work for told me about a phone call he had gotten. A man needed a defense attorney. The man did not search for one. He did not scroll a results page, did not compare websites, did not click a single link. He asked ChatGPT who he should call. ChatGPT gave him a name. He called the name.

No search. No click. No website. A question went in and a person came out.

I have spent more than twenty years building websites in the Tri-Cities, including thirteen of them running the websites of our local newspapers. When he told me that story, I realized the future I had been watching slowly arrive had finally reached my own hometown. The way people use the internet has changed for the third time in my working life, and most of us are still building for the version that just ended.

The library

For almost thirty years, we treated the internet like a library. You needed something, you asked, and the system pointed you to a shelf of books. Your job was to pick one. Every website I have built since the early 2000s, and every search decision we made at the newspapers, served one goal: get our book onto the first shelf.

But before the library had a search desk, it had a card catalog. And people kept that catalog by hand.

The Directory era: I browse (1994 to 1998)

Yahoo began in January 1994 as a hand-built list of websites kept by two Stanford graduate students. It was literally called Jerry and David's Guide to the World Wide Web before it got a better name. The Open Directory Project, better known as DMOZ, launched in June 1998 and did the same job with an army of volunteers, more than 90,000 editors at its peak. If you wanted your website found, you submitted it to a human being and waited for a human being to approve it.

I lived this at street level. My first paid website build came around 2002, when a man walked into the little PC repair shop I was running and asked if I could build a site for his Oriental rug company. I said yes, quoted him way too little, and built it in HTML with frames, leaning on Dreamweaver to keep the design and the code talking to each other. When it was done, I submitted it to the Yahoo Directory and to a local Tri-Cities directory that no longer exists. That was the marketing plan. Around here in 2000, a website was a brochure: here is what we do, here is how to reach us. Nobody talked about leads or funnels. People found their way to the web through Yahoo and Ask Jeeves, and through this newer one, Google, that everybody kept recommending.

Then the language changed. Somewhere around 2000 I started hearing "just Google it" where a web address used to go, and by 2002 or 2003 it was simply how people talked. When the verb changes, the era has changed. The directories hung around as ghosts for years afterward. Yahoo finally closed its directory at the end of 2014. DMOZ went dark in March 2017. By then, almost nobody noticed.

The Search era: I search (1998 to 2024)

Here is my favorite detail from that first turnover, because it rhymes with right now. In 2000, Yahoo hired Google to power the search box on Yahoo's own pages.

For roughly four years, Yahoo trained hundreds of millions of its own users to trust the results of the company that would replace it.

By the time Yahoo cut Google loose in 2004 and launched its own crawler, it was too late. Google had passed Yahoo somewhere in the 2002 to 2004 window and never looked back. Incumbents have a habit of inviting their replacement inside.

I joined our local newspapers in 2012 to run their websites, and I spent the next thirteen years inside the Search era's engine room. In the newsroom, the homepage was the front page. It was the product. But the analytics told a different story. By the Libercus years, around 2015 and 2016, I could see that the article pages consistently outdrew the homepage, and I started saying out loud that the homepage was not our main point of entry anymore. I remember a sales rep asking me why the homepage numbers were down, and explaining that thanks to Google, Google News, and social media, readers were arriving through individual article pages. They were coming in through the side doors, thousands of them, and skipping the front door entirely. We still treated the homepage as the product through the replatform around 2018 and again through the redesign around 2021, while the data said otherwise both times.

The numbers I can offer here are recalled, not cited, because I no longer have access to those analytics accounts. As best I remember, search alone easily accounted for 35 percent of our traffic at its peak. Facebook was our single biggest referrer for years, first in the clickbait era, when readers poured in for one headline and bounced right back out, and then it collapsed when Facebook changed its algorithm to keep people on its own platform. The drops were enormous. Even then, the lesson was sitting in plain sight: the platforms owned the front door, and they could close it whenever they wanted.

We did not notice it happening because traffic was still growing. The numbers looked healthy. But the source of that traffic had quietly changed. Fewer people woke up and typed our homepage into their browser each year. More arrived through a single article, read exactly what they came for, and left. We thought we were publishing newspapers online. In reality, we had become one stop on somebody else's route through the web.

For the news business, that closing is now measurable. An analysis of more than 400 news publishers worldwide by NewzDash found that traditional Google web search fell from 51 percent of Google-driven traffic to news sites in 2023 to 27 percent by late 2025, while Google Discover, the algorithmic feed on your phone, grew to nearly 68 percent. The front door I spent thirteen years optimizing was closing while I watched.

The Conversation era: I ask (2025 and on)

Which brings us to the phone call, and to the reason the library metaphor no longer fits.

A library hands you a shelf of books and lets you decide. What is emerging now works more like a city. You do not ask a city where the plumbing books are kept. You ask who can fix your leaking pipe today, and the city routes you to someone.

AI assistants move through information the way people move through a city.

Libraries are built around shelves. Cities are built around relationships. Restaurants gather near hotels. Law offices cluster around the courthouse. Coffee shops open where people already walk. Nobody experiences a city one building at a time. They experience it as connected places. AI reads the web the same way. It understands a business partly by what the business says about itself, and partly by how everything around it connects: who mentions it, who links to it, what its neighbors say. It crosses between sources, compares, summarizes, and hands back one synthesized answer.

The man who needed a defense attorney was not a library patron. He was a resident of the new city asking for directions, and the city answered.

The data says most people are already living there part time, even inside Google itself. The Pew Research Center tracked the real browsing behavior of 900 American adults across nearly 69,000 Google searches in March 2025. When Google showed an AI summary at the top of the results, users clicked through to a website only 8 percent of the time, versus 15 percent when no summary appeared. Only 1 percent ever clicked a source link inside the summary itself. The answer sat there, and the visit never happened.

The bigger picture is the same trend on a longer clock. SparkToro, working with Similarweb clickstream data, found that just over 68 percent of US Google searches in early 2026 ended without a single click to any website, up from about 60 percent in 2024 and roughly 45 percent a decade ago. Two out of three visits to the library now end at the front desk without anyone taking a book off the shelf.

Here is the honest part, and it matters. The AI assistants themselves send almost no traffic anywhere. Similarweb measured about 1.13 billion referral visits from AI platforms to the top 1,000 websites in a single month, against 191 billion from Google search in the same month. That is not people clicking from ChatGPT to websites instead of from Google to websites. That is answers getting resolved with no click at all.

The referral economy did not move. It evaporated.

So no, AI is not replacing websites. AI is becoming the new front door.

The website is still the building. The new doorman reads everything and points at almost nothing, and it decides which buildings deserve visitors.

I should say that I am not the first to slice the web into eras. Microsoft published its own three-eras framing this spring, aimed at advertisers deciding where to sell, and plenty of search histories have traced the road from directories to crawlers to AI answers. What I am describing is the view from the other side of the counter. I built websites through all three eras, and from where I sat, each shift moved more of the thinking from the person to the machine. First we navigated. Then we searched. Now we ask.

What this means if you own a building

In the Search era, I watched local businesses learn the library's tricks. A landscaping company would hire an SEO outfit that churned out a page for every town: Landscaping Kingsport, Landscaping Johnson City, Landscaping Bristol, the same 300 words with the city name swapped. It was built to impress a filing system, and for a while it worked. Eventually it cost them credibility twice, once with search engines that learned to spot it, and once with actual customers who noticed the copy and paste. It was a book written for the card catalog instead of the reader.

Compare that to what a page looks like when it is written for a reader. Not "Landscaping Kingsport" repeated five ways, but "How we regraded a backyard in Colonial Heights that flooded every spring." Not another copied services list, but "What a retaining wall actually costs in Northeast Tennessee, and where the money goes." One is inventory for a filing system. The other is evidence that real people solved a real problem, and evidence is exactly what a reasoning system can confidently hand to the next person who asks.

A reasoning system is a different reader. It does not count keywords. It follows relationships, weighs evidence, and looks for signs that a real person solved a real problem. This is not just intuition. A peer-reviewed study out of Princeton and Georgia Tech found that structural changes like quoting real experts, including concrete numbers, and writing clearly organized answers measurably increased how often AI systems cited a page, by as much as 40 percent in their tests. The things that make a page trustworthy to a careful human turn out to make it citable to a machine. Clarity finally beats volume.

And the stakes are already local. BrightLocal's 2026 consumer survey found that 45 percent of consumers had used an AI tool to find local business recommendations in the past year, making AI the third most-used discovery channel, behind only Google and Facebook. People do still verify. Only 18 percent said they would contact an AI-recommended business without checking it out elsewhere first. The assistant builds the shortlist, and the website closes the deal. Both halves matter.

Now the other side of that coin. Industry measurements suggest the city has barely started drawing its map. One analysis of more than 350,000 business locations found ChatGPT recommending only about 1 percent of them, against roughly 36 percent of the same businesses showing up in Google's local map results. Those numbers come from vendors with something to sell, so hold them loosely. But the direction is consistent everywhere I look: people's willingness to ask AI is racing far ahead of AI's actual coverage of local businesses.

Business owners ask me some version of the same question all the time now: should I even be in AI? There is real anxiety behind it, and I understand where it comes from. But the honest answer is short. Your competitor is going to be. AI is not going away, the bubble is not going to pop and take it with it, and neither is your website. The only real question is whether the city knows your building exists.

Building for a different reader

Back to that phone call one more time. A man needed help, asked a question in plain English, and got routed to a person who could help him. Strip away the technology and that is the most natural act in the world. It is how you would find a plumber in a small town in 1950: you would ask somebody you trusted. The Search era, with its keywords and its ten blue links and its shelf of results, may end up looking like the strange middle chapter, the decades when we all learned to talk like filing systems.

We spent thirty years building websites for search engines. Now we are building them for reasoning systems, and that is a very different reader. It cannot be impressed by volume, and it is hard to fool with tricks. What it rewards is the thing good businesses had all along: real work, clearly explained, in a place it can find.

For thirty years, success meant writing the best book on the shelf. The next thirty may belong to the businesses that become the buildings everyone trusts.

Libraries reward organization. Cities reward reputation. The internet is becoming a city.

Most businesses are still arranging books.