- [ Home ](/) / - Methodology # Methodology Every llmtxt.info measurement has a named scope, a dated snapshot and a published limitation. This page describes the current process so results can be checked or reused without guessing. Last updated: August 12, 2026 ## Measurement principles - **Primary source first.** Provider documentation and the served public file take precedence over summaries. - **Date every observation.** A result describes the checked response at a moment in time, not a permanent property of a domain. - **Separate publication from use.** A reachable file shows that it was published. It does not establish automatic discovery, AI traffic, ranking, or answer inclusion. - **Publish the method with the number.** The panel, result files and release manifest remain downloadable alongside the pages that summarize them. ## Fixed-panel tracker The tracker tests a curated fixed panel of 219 public hosts. On the snapshot dated 2026-08-03, 218 hosts were testable and 113 returned a qualifying `llms.txt` response. A host qualifies only when its requested file is public, returns HTTP 200, is not an HTML app shell, and begins with a Markdown H1. The panel is intentionally developer-heavy. It is useful for comparing this same cohort over time, not for estimating all websites. The full host list and raw metrics are available at [/adoption-panel.json](/adoption-panel.json) and [/adoption-stats.json](/adoption-stats.json). ## V2 discovery checks A separate snapshot checks observable discovery signals on verified files: linked-page reachability, Markdown alternates, the `describedby` relation, and path-scoped files. The latest release checked 294 linked pages. It found 31 pages with a Markdown alternate, 4 pages with `describedby`, and 21 hosts with a qualifying subpath file. These are implementation observations only. They do not claim that any client consumes a relation automatically. The raw snapshot is available at [/v2-adoption-stats.json](/v2-adoption-stats.json). ## What the data cannot prove - It cannot show that an AI product read a file, used it in retrieval, or cited it in an answer. - It cannot establish a Google ranking effect. Google documents that it does not use `llms.txt` for Search or its generative search features. - It cannot turn a selected panel into a global adoption estimate. - It cannot infer a relationship, partnership, customer status or endorsement from a public file. ## Downloads and reuse Cite the snapshot date and its identifier when reusing a figure. The machine-readable release manifest contains the source hash, scope and limits: [JSON](/observatory.json), [CSV](/observatory.csv), and [status](/observatory-status.json). For wording and commercial-disclosure rules, see the [editorial policy](/editorial-policy/). ## Sources - [ llmstxt.org, specification proposal ](https://llmstxt.org/) On this page - [ Measurement principles ](#principles) - [ Fixed-panel tracker ](#panel) - [ V2 discovery checks ](#discovery) - [ What the data cannot prove ](#limits) - [ Downloads and reuse ](#reuse)