The open internet has a new traffic problem: the visitor may never arrive. Axios reported this week that executives at its Cannes event framed the shift bluntly: bots now outnumber people online, and AI systems are turning the web from a destination network into a raw material layer.
That matters because the old bargain was simple. Search engines indexed pages, sent readers to publishers and merchants, and those visits could become ad revenue, subscriptions, sales, or leads. AI breaks that chain. A model can read, summarize, compare, and recommend without producing a meaningful click. Cloudflare co-founder and CEO Matthew Prince said at the event that users increasingly trust AI answers and skip the cited links, cutting publishers out of the path where money used to move.
The numbers Axios cited are the useful shock. Prince said getting a click from Google is already far harder than it was four years ago, and that AI referrals are dramatically worse: OpenAI and Anthropic send far fewer visitors back for each crawl or answer than traditional search. Even if the exact ratios vary by site and category, the direction is hard to ignore. AI discovery is becoming high-consumption, low-referral distribution.
For publishers, this pushes licensing from a legal side quest into core revenue strategy. If models are going to absorb work without reliably sending readers back, then content owners need terms, measurement, and enforcement. Robots.txt was built for a different internet. The next layer needs crawler identity, usage reporting, opt-out rules that do not punish search visibility, and payment rails that make lawful access easier than scraping around the edges.
For brands and ecommerce operators, the same shift changes the job of marketing. AI agents do not browse like humans. They compare thousands of options without loyalty, visual merchandising, or emotional attachment. That makes product data, pricing clarity, availability, reviews, warranty terms, and trust signals more important than page design alone. If an agent is the first buyer interface, your site has to be legible to machines without surrendering the whole relationship.
Spotify's example in the Axios report points to one possible pattern: models that give users more control and compensate creators when their work is used. That is not a complete answer for news, retail, software documentation, or independent media, but it is the shape of the fight. The web needs commercial rules for AI intermediaries, not just better prompts and more badges.
The practical takeaway is sharp: every content and commerce business should audit how AI systems see it. What can be crawled? What is licensed? What structured data exists? Which pages still earn clicks, and which are being mined without return? The answer will determine who has leverage as AI agents become the new front door to the internet.