For thirty years, web publishing optimized for exactly one reader: a human with a browser, arriving from a search engine. That assumption is visibly breaking. Language-model agents fetch pages to answer questions, compare products, monitor changes and complete tasks, and they experience a modern webpage very differently than people do. Navigation chrome, consent banners and script payloads consume an agent's context budget before the substance begins.
The early response is a quiet dual-audience rebuild. Sites are publishing llms.txt indexes, root-level markdown maps of their most important content, and serving clean markdown mirrors beside their HTML. A June 2026 scan found llms.txt on 87 of the web's top 1,000 domains, or 8.7 percent, roughly twenty-one months after the proposal.
Why this is early, not passing
The economics favor legibility. Serving an agent a markdown mirror instead of full HTML can sharply reduce token consumption, giving agent operators a practical reason to prefer sources that are cheaper to read. Preference can compound into defaults, and defaults into distribution. The parallel with early search-engine optimization is useful, except this time the crawler is also the reader and the buyer's assistant.
The visible triggers to watch: whether major model providers formalize llms.txt or a successor into their retrieval stacks; whether agent traffic becomes a reported line in analytics suites; and whether any large publisher begins charging agents differently than humans.
Preference compounds into defaults, and defaults into distribution.
What it means for anyone publishing information
Being agent-readable is becoming a developer-experience and access strategy, not only a technical courtesy. A site that exposes structured indexes, stable markdown mirrors and honest metadata is easier for machine readers to parse. A site that hides its substance behind scripts and interstitials is, from an agent's point of view, barely there at all.
The consequence cuts both ways: content built for machine reading is also easier to verify, cite and correct. The practical aim is not only to be found, but to be quoted more accurately.