Le Monde blocked the bots. Now it’s working out what to do about paying readers showing up as agents - Digiday
AI Revenue Generation
Le Monde blocked the bots. Now it’s working out what to do about paying readers showing up as agents
By Jessica Davies • June 17, 2026 •
Ivy Liu
Le Monde blocks almost every bot that tries to hit its site. Now it’s starting to think about what happens when its paying readers show up via AI agents instead of a browser.
The French news publisher now blocks almost all non-human traffic unless there’s a licensing deal in place, including Google’s AI training crawler Google Extended.
Speaking to Digiday at Fastly’s Xcelerate event in London last week, Le Monde CTO Paul Laleu, said the company is “figuring out” how to maintain its subscription partnership with readers who use AI agents rather than its homepage or app. He’s interested in early technical standards that would let an AI assistant tell Le Monde, in effect, “I’m fetching this article for a user who already pays you,” so the site can decide what to show. “I can recognize that user and be able to say: ‘this one is a subscriber, I can give him the content because he’s paying money for it,’” said Laleu. “I see that as part of the future of the web.”
Laleu expects that kind of agent-mediated access to stretch beyond chatbots into on-device assistants. He pointed to Apple’s June 8 announcement of its Intelligence features and Google’s Gemini assistant as early signs that software like Siri or Gemini will increasingly sit between users and publisher apps. Inside the phone, he said, the interaction between an agent and Le Monde’s app will become “a new area to work on” if the publisher wants paying readers to be recognized as subscribers even when they never visit the homepage.
To get there, Laleu is watching early technical standards as potential plumbing to let AI agents pass a simple message back to Le Monde: this user is logged in and already pays. In practice, that means things like Model Context Protocol (MCP) apps — a way for AI models to plug into external tools and data sources — and OAuth‑style extensions: the same login flows many sites already use, adapted so agents can prove who a user is and what they’re allowed to see. Then the publisher can apply its paywall and access rules even when requests don’t come directly from a browser.
Publishers aren’t just reacting to today’s crawlers; they’re trying to get ahead of a world where most requests to their servers won’t be from people at all. Fastly’s most recent analysis of billions of requests across its global network shows AI-driven traffic is growing at 6.5 times the rate of human traffic, pointing to a fundamental change in how people are using the internet and why publishers are starting to design for “agent” visitors alongside human ones. Tollbit’s latest report points in the same direction, estimating that for every 31 human visits to a site, one is from a bot.
The Fastly findings point to a new reality for businesses: blocking malicious bots still matters, but automated traffic is becoming too important to treat only as a threat. “We’re just about to tick over into 50% of all traffic is bots… and that percentage of crawler and fetcher AI bots is creeping up and up,” said Simon Wistow, co-founder of Fastly. The job now, he argued, is not just to stop the wrong bots, but to work out what kinds of automated interactions you do want and build the tools to support them. “No two publishers want the same thing,” he added.
Le Monde is not the only publisher starting to plan for an agent-driven future. The Economist and others are also experimenting with ways for AI assistants to pull from their reporting while still respecting paywalls and subscriber status, treating agents as a new distribution channel rather than just another crawler to block.
Le Monde’s thinking on agents sits on top of a much harder line it has already taken on bots. Since 2023, the publisher has moved from simply filtering out bad traffic to blocking almost all non‑human visitors by default, and only making exceptions for companies that sign licensing deals. So far that includes OpenAI, Meta and Perplexity, though Laleu is confident more will flow. “We have open discussions with all those companies, and we strongly believe that we will manage to get deals from all of them,” he said.
Even with licensing deals in place, Lelau said the usage data coming back from AI partners is “not complete” compared with what publishers see in Google Search Console for SEO. He expects AI firms to build more console‑style tools for their products over time, noting that Google has already said it will add AI Overview reporting into Search Console, which he believes others may copy.
Laleu credits a long‑running partnership with Fastly with helping build what is now effectively a technical moat around its “no deal, no crawl” policy as AI crawlers and...