# AI Agents Now Outread People on Documentation Sites. Here's What They Receive. > AI agents now make more requests to documentation sites than people do. We tested 40 leading documentation and knowledge sites to see what those agents receive. On many, it is a blank page. **Canonical URL:** https://discovercx.com/blog/ai-agents-now-outread-people-on-documentation-sites **Last updated:** 2026-06-04 --- *By David Hillis — 2026-09-24T17:00:00.000Z* AI agents now read documentation more than people do. In one week this spring, AI agents made up 51.8% of documentation reads on GitBook, ahead of people. In January 2025, AI accounted for less than 10% of its documentation traffic (GitBook, May 2026). No count catches every agent. Some don't identify themselves, and some arrive looking like an ordinary browser. But the direction is clear. The most frequent reader of your documentation is now software working on someone's behalf: a coding assistant, a support bot, or an AI answer engine responding to a customer before they ever reach your site. That raises a question most documentation teams can't answer yet. When an agent reads our docs, what does it receive? We tested 40 documentation sites to find out For the Agent Readiness Index 2026, we scanned 40 leading documentation and knowledge sites, 373 pages in all. They span healthcare and life sciences, enterprise software, security and infrastructure, and medical devices and industrial, plus a control group of developer platforms. Half come from enterprises and half from mid-market companies. Every page was read three ways: as a person sees it in a browser, as an AI crawler receives it, and as an AI tool that doesn't run JavaScript receives it. Each site was scored out of 10 on three measures: Access: can agents get to the content? Content: how much of what a person sees does the agent receive? Trust: does the page say which date or version it covers? The median score across the 40 sites was 5.7 out of 10. Only 7 scored 8.0 or higher. What we found Agents are let in, then handed an empty page 38 of the 40 sites allow all six major AI crawlers in robots.txt. Getting in is not the problem. What agents receive once they're in is. Outside developer platforms, 15 of 32 documentation homepages return almost no text to AI crawlers. Release notes, security notices and "start here" links load with JavaScript, and crawlers that don't run it receive a blank page. That means the newest and most important content on the page is the part agents miss. All 8 developer platform homepages in the study returned their full text. Enterprise software documentation fared worst, with a median Content score of 1.9 out of 10. Trust is the weakest score in every industry The median Trust score is 2.5, against 8.0 for Access and 7.1 for Content. Only 13 of 40 sites put a machine-readable date on most pages, and only 8 state a last-updated date or version in text an agent can read. For documentation, this matters more than almost anywhere else. An agent that can't tell version 4 from version 6 will answer with whichever page it found first, and your customer will get instructions for software they don't run. Mid-market companies outscore enterprises Companies with fewer than 5,000 employees have a median score of 6.8. Enterprises score 5.0. Six of the seven top-scoring sites come from mid-market companies. Bigger documentation budgets aren't buying agent readiness. Most sites have no usable llms.txt Only 14 of the 40 sites publish a valid llms.txt file. 20 have none. Another 4 return a web page at /llms.txt instead of the file, so tools that check for one get a false positive. Every top-scoring site has a valid llms.txt. What the top sites have in common None of this requires rewriting your documentation. The sites scoring 8.0 or higher share six traits, and most of them are about how pages are delivered, not what they say: The homepage ships as finished HTML, not an empty shell filled in by JavaScript. All six major AI crawlers are allowed in robots.txt. A valid llms.txt file names the site and links to its main sections. Every page has a canonical URL, so agents cite one address instead of duplicates. Almost every page carries a machine-readable date in its markup. At least 80% of the text a person sees on a doc page reaches the crawler without JavaScript. The full report includes heatmaps of all 40 homepages, scores by industry and company size, and a side-by-side teardown of a blank homepage and a top performer. Why we ran this study At DiscoverCX, we build knowledge centers: the documentation portals, help centers and knowledge bases that customers, support teams and field technicians rely on. More and more, those readers reach the content through an AI agent instead of a browser. The question we hear from documentation leaders has changed. It used to be "can people find the answer?" Now it's "what does the AI say when a customer asks about our product?" That answer starts with what the agent received from your site. Our view is that a knowledge center should serve both readers from the same source: finished pages for people, and clean, dated, versioned content for agents. For transparency, we also scored one site we built, the Canadian Cancer Society's cancer information center. It's reported separately and isn't included in the 40-site medians. It scored 8.3. See how your site scores Read the Agent Readiness Index 2026, or request your own score. We'll scan your documentation site with the same method and send you your homepage heatmaps, your Access, Content and Trust scores, and the five fixes with the most impact. --- ## About DiscoverCX DiscoverCX is the headless content delivery platform built on the world's leading CCMS — author in DITA, Markdown, or HTML, deliver to portals, docs sites, Salesforce, and AI assistants from one source of truth. **Talk to sales:** https://discovercx.com/contact **Request a demo:** https://discovercx.com/demo **Pricing:** https://discovercx.com/pricing *This is the LLM-friendly markdown version of https://discovercx.com/blog/ai-agents-now-outread-people-on-documentation-sites. The human-readable page is at the canonical URL above. For a full site index, see https://discovercx.com/llms.txt.*