Ask ChatGPT, Claude, or Perplexity a question your business should own the answer to. Watch who it names. For most brands the answer is a competitor, or no one at all.
That is the new distribution layer. Not the search result. The answer. Call it the answer economy. In it, you are named or you are invisible. And most of the web is being locked out of the layer that names you, right at the moment it starts to matter.
The platforms locked AI out. It happened in under two years.
Reddit overhauled its API in 2023 citing AI training, then locked its public data behind licensing in 2024 and started charging for access, with a reported sixty million dollar deal with Google and a separate arrangement with OpenAI. Stack Overflow signed a data-licensing partnership with OpenAI in May 2024, after cutting an earlier deal with Google. LinkedIn has never meaningfully opened to AI crawlers and now actively blocks scraping. X shut its API to everyone but paying enterprise customers.
Then the infrastructure layer moved. On July 1, 2025, Cloudflare became the first major internet infrastructure company to block AI crawlers by default. That matters because Cloudflare sits in front of roughly twenty percent of the web. Overnight, a fifth of the internet flipped from open to permission-required for AI.
And it is still tightening. As of a Cloudflare announcement this month, new sites will soon default to allowing search but blocking AI training and agent use on any page that carries ads, with mixed-use crawlers blocked alongside them. The direction is one way.
The crawl-for-traffic bargain is dead.
Here is the number that reframes everything. Bots are now the majority of web traffic. Cloudflare's own data put automated requests at 57.5 percent of all web traffic as of early June 2026, up from around thirty percent a year earlier. There are more machine readers on the web now than human ones, and the gap is widening.
For two decades the deal was simple. Crawlers indexed your content, search sent visitors back, visitors paid the bills. Both sides profited. Generative AI broke that contract, and the data shows how badly.
Cloudflare measures a crawl-to-referral ratio: pages an AI takes versus visitors it sends back. Google's traditional search runs about five pages crawled per visitor returned. Workable. The AI assistants run in the thousands to one. By May 2026, every AI chatbot combined sent about 0.29 percent of measurable referral traffic on the web. Google sent roughly 88 percent.
The machines are reading everything and sending almost no one back. That is the quiet crisis. Not that AI reads your work. That it reads your work thousands of times and returns you nothing.
AI answers from whatever it can still reach.
AI systems need fresh, authentic, human expertise to work well. The training-data problem is real. The retrieval problem is worse. When you ask a question, the model reaches for recent, credible, structured content. When that content is behind a wall, it reaches for whatever is left.
What is left is two categories. Major-publisher content, negotiated into the models through licensing deals. And the slowly rotting archive of the pre-lockout web: old Reddit threads, YouTube comments, SEO spam.
Meanwhile the world is full of experts with genuine knowledge. The solo therapist who has seen two thousand cases. The indie game dev who shipped six titles. The recipe blogger who has tested every variation of a dish. The MACH architect who has rebuilt seven enterprise commerce platforms. They have written things. Almost none of it shows up in AI answers, because it lives on personal blogs with no structure, or behind walls that now block crawlers, or nowhere at all.
Search stopped sending traffic. AI started keeping it.
While AI starves for real expert voices, those same experts are losing their distribution.
Google's AI Overviews are the clearest case. Pew Research found that when an AI summary appears, people click a traditional result about 8 percent of the time, down from 15 percent when there is no summary. That is roughly half the clicks, gone. Zoom out and it is worse. SparkToro's June 2026 analysis found that fewer than one in three Google searches now sends a click to the open web at all, with US zero-click rates sitting near 69 percent. The search-to-traffic engine that funded independent expertise for twenty years is winding down.
So two groups face each other. AI systems that need human expertise and cannot reach it. Human experts who need to be read and cannot get there. Nobody building the connection between them.
The fix is not to reopen the old web.
The obvious move, the one most tech people reach for, is to force the old web open. Demand that LinkedIn and Reddit and X let the bots back in.
That is a losing fight, and it should be. Those platforms walled off because scraping was destroying their economics and their users never consented to feeding AI training. The walls are not coming down.
The better answer is to build for the reader that is actually arriving. A platform designed from the start for both human readers and machine readers. One where experts want to publish because the structure serves their work. One where AI crawlers are welcomed, not tolerated. One where attribution is built in, so experts get credit when machines cite them.
That is what I built. It is called Markwright.
What Markwright is
A publishing platform where verified human experts post structured content designed to be found, read, and cited by AI systems.
Creators sign in with LinkedIn. They write in Markdown. The platform structures each post with the metadata AI systems actually use: clean semantic HTML, schema.org markup, claim-type annotations, citation-ready formatting, explicit expertise declarations. Every post lives at a stable URL, served both as human-readable HTML and as machine-readable Markdown. The platform publishes an llms.txt index, RSS feeds, and a clean JSON API. AI crawlers are explicitly welcomed in robots.txt, not grudgingly allowed.
There is no paywall. Bots access everything free. The platform is funded by the creators, not by the machines reading them. First five posts a month are free. Beyond that, posts cost a dollar each from a wallet top-up.
It is the inverse of every content-monetization play from the last two years. TollBit charges bots for access across more than three thousand publisher sites. ProRata splits ad revenue with over five hundred publishers, including The Atlantic, TIME, and Fortune, based on how often their work is cited. Cloudflare and now Microsoft are building whole marketplaces to make bots pay. They are all trying to charge the machines. I am giving machines clean access and charging creators for distribution. The unit economics only work because the creators want this more than the bots do.
This has a name now. Agent Experience.
When I started building Markwright, treating bots as the primary audience felt like a fringe bet. It is not fringe anymore. It has a name and a discipline.
Netlify coined the term Agent Experience, or AX, in early 2025. The definition is clean: AX is the experience an AI agent has when it uses a product on behalf of a human. Developer Experience optimizes for the people building on your platform. Agent Experience optimizes for the agents those people increasingly send in their place. When a customer's agent hits your API and fails silently, that is not a DX problem. It is an AX problem.
Netlify took it further and built AXIS, an open-source framework at axis.run that scores how well any service serves agents, across goal achievement, service quality, environment, and agent behavior. They even launched a product entry point built for agents rather than humans. Their argument is that autonomous agents are becoming a whole new user base, quite possibly the most important one.
That is the exact thesis Markwright is built on, aimed at expert content instead of developer tooling. Designing for agents is not theory for me. I already run a nine-agent company on a refurbished mini PC. Markwright is the same instinct pointed at expert knowledge. The machine reader is a first-class user. You design for it deliberately. You structure the page so an agent can find the passage, read it cleanly, and cite it correctly, with the human still getting a beautiful article. The dual-layer pattern, a human-optimized surface plus a machine-optimized feed, is becoming a standard architecture across publishing. Markwright is that pattern, built for individual experts, with the economics flipped so the bots ride free.
If the web is splitting into a human web and an agent web, expert knowledge needs a front door on both. That is what this is.
The web is splitting in two. There is a window to claim position.
There is a larger bet under Markwright, and it is worth saying out loud.
The web is sorting into two systems. One is locked, licensed, and negotiated between big publishers and big AI companies. The other has to be built deliberately, with machine-readability baked in from the structure layer up. Not scraped. Not legacy-indexed. Purposefully published for the new reader.
I think there will be multiple platforms doing this, and they will look different by vertical. There is room for Markwright to serve solo experts and small publishers while enterprise-scale platforms serve the giants. Those are different problems.
I also think the window to establish position is eighteen to twenty-four months. After that, the big AI companies either build this natively or sign enough exclusive deals that the default citation set is locked in for a decade. That is not a forever window. That is a now window.
Where to start
Markwright is live at markwright.app. Signup is free. First five posts a month are free.
If you are a founder, a consultant, a creator coach, a newsletter writer, an indie expert, a technical practitioner, a solo therapist, or a professional with real knowledge and no good place to put it where machines will find it, you are who this is for.
The first contributors set what this platform becomes. In a category where citations compound, your work shapes what AI learns to name.
The answer layer is being written right now, one citation at a time. The experts who show up early are the ones AI learns to name. Show up.
