As AI assistants and autonomous agents become more capable of reading websites, developers are looking for cleaner ways to make online information easier for machines to understand. One emerging solution is llms.txt, a proposed website standard designed to give AI systems a concise, structured path to important content.
In August 2026, the proposal received its first major revision with the release of llms.txt V2. The updated specification adds formal methods for connecting normal web pages to Markdown versions of the same content, helping AI agents locate cleaner, more machine-readable information.
What Is LLMS.txt?
The llms.txt concept was originally proposed by Jeremy Howard in September 2024. It suggests that websites publish an /llms.txt file containing a concise overview of the website along with links to useful resources that an AI system may need.
Unlike a traditional XML sitemap, which primarily helps search engines discover indexable URLs, llms.txt is intended to give language models and AI agents a curated understanding of a site’s important information.
The file itself uses Markdown because Markdown is relatively simple for both humans and language models to read. A typical llms.txt file can contain a website or project name, a short description, additional context and categorized links to important resources.
The goal is not necessarily to replace existing web standards such as robots.txt or sitemap.xml. Instead, llms.txt is designed to provide another layer of machine-readable context specifically for AI-powered systems.
What Changed in LLMS.txt V2?
The most important improvement in V2 is better communication between standard HTML pages and their cleaner Markdown equivalents.
Previously, the specification suggested that websites could make Markdown versions available by adding .md to an existing page URL.
For example:
/docs/guide.html.md
With V2, websites can also replace the original file extension entirely:
/docs/guide.md
Both approaches are now supported by the proposal.
More importantly, V2 introduces formal linking mechanisms that tell an AI agent exactly where these alternative resources exist.
New “Alternate” Markdown Linking
One of the key additions is the use of the standard HTML relationship:
rel="alternate"
combined with:
type="text/markdown"
This allows a web page to explicitly identify its Markdown version.
For example, an HTML documentation page could tell an AI agent that a cleaner Markdown copy of the same information is available elsewhere.
This is significant because agents no longer need to guess whether /page.md, /page.html.md or another URL contains the Markdown version.
The relationship can be declared in the HTML <head> or delivered using an HTTP Link response header.
The New “Describedby” Relationship
V2 also introduces another relationship:
rel="describedby"
This can point an AI system toward the llms.txt file that describes a particular page or website section.
For example, a website could have:
/llms.txt
covering the entire website while a large documentation area could potentially have:
/docs/llms.txt
providing more specific information for documentation content.
The proposal states that an llms.txt file can describe pages underneath the path where it is located, allowing websites to create more focused AI-readable structures when necessary.
Why Markdown Matters for AI Agents
Modern websites contain far more than the primary article or documentation content.
Navigation menus, scripts, advertisements, cookie banners, widgets and dynamic page elements can all increase the amount of information an AI agent needs to process.
Markdown provides a much simpler representation.
Instead of interpreting complicated HTML structures, an AI agent can potentially retrieve a clean document containing headings, paragraphs, links and essential information.
The llms.txt proposal argues that this can reduce unnecessary processing and provide agents with more concise, expert-level information.
This approach may be particularly useful for documentation websites where coding agents frequently need to find API references, tutorials and implementation instructions.
LLMS.txt Is Already Appearing Across the Web
Although llms.txt remains a proposal rather than a universal web standard, its adoption has been growing.
The specification notes that documentation platforms and CMS providers are already generating llms.txt files, while developer documentation from organizations including OpenAI, Anthropic and Google’s Gemini team also publishes them.
The V2 revision was partly created because practical adoption exposed a limitation in the original proposal.
An llms.txt file could point an agent toward useful pages, but there was no standardized mechanism allowing individual web pages to clearly tell the agent where their Markdown equivalents were located.
V2 attempts to solve that discovery problem.
Does LLMS.txt Improve Google Rankings?
This is where website owners need to be careful.
Creating an llms.txt file should not currently be treated as a Google ranking tactic.
Google Search has stated that it does not currently use llms.txt for Search visibility, and Search Engine Journal reports that Google’s position remains unchanged after the V2 update.
Therefore, adding an llms.txt file does not mean a website will suddenly rank higher in Google Search, AI Overviews or AI Mode.
There is also currently no solid evidence showing that an llms.txt implementation directly increases citations from major AI search platforms.
Businesses should avoid presenting it as a guaranteed AI SEO ranking factor.
Then Why Should Websites Consider LLMS.txt?
The potential value is broader than traditional SEO.
AI systems are increasingly becoming website users themselves.
Coding assistants may retrieve documentation. Research agents may gather product information. Customer-service agents may inspect policies, pricing or technical guides. AI browsers may navigate websites on behalf of users.
Providing those systems with a clearly structured machine-readable path could improve how efficiently they access and interpret important information.
For documentation-heavy companies, SaaS platforms and technology businesses, the benefits may therefore appear sooner than they do for ordinary marketing websites.
LLMS.txt and AI SEO Are Not the Same Thing
There is growing interest in terms such as AI SEO, AEO and Generative Engine Optimization.
Because of that interest, it is easy to assume that every technology intended for LLMs automatically improves visibility inside ChatGPT, Gemini, Perplexity or Google AI search.
That assumption is premature.
LLMS.txt primarily addresses content discovery and accessibility for AI agents.
AI visibility remains dependent on many other signals, including the quality and accuracy of a website’s content, crawlability, authority, entity clarity, structured information, citations and whether an AI platform can actually access and trust the information.
Website owners should therefore view llms.txt as one possible component of an AI-ready technical infrastructure rather than a complete AI SEO solution.
Should You Implement LLMS.txt V2?
For most websites, llms.txt should currently be considered an experimental enhancement rather than an urgent technical SEO requirement.
However, implementation may be worthwhile for websites containing extensive documentation, APIs, developer resources, research libraries or other structured information frequently accessed by AI tools.
If you already publish an llms.txt file, V2 provides a clearer framework for connecting it with Markdown versions of your pages.
The specification says existing implementations generally require relatively minor changes rather than rebuilding the entire setup.
Businesses starting from scratch should first make sure their fundamental website infrastructure is strong.
Clean HTML, accessible content, useful internal linking, structured data, correct canonicalization, XML sitemaps and technically crawlable pages remain more established priorities.
The Bigger Shift: Websites Are Being Built for Humans and Agents
The most interesting part of llms.txt V2 may not be the file itself.
It represents a broader change in how the web is being consumed.
Historically, websites were created primarily for two audiences: people and search-engine crawlers.
Now a third audience is becoming increasingly important: AI agents.
These systems may research products, compare companies, read technical documentation, answer customer questions or perform tasks without a person manually visiting every page.
That could eventually change how websites structure and expose information.
Traditional SEO made websites easier for search engines to understand. The next stage of technical optimization may increasingly involve making websites easier for AI systems to navigate, interpret and use.
Final Thoughts
The introduction of llms.txt V2 is an important development in the evolving relationship between websites and AI agents.
The August 10, 2026 revision adds formal relationships for connecting HTML pages with Markdown alternatives and the llms.txt files that describe them.
It does not currently provide a proven Google ranking advantage, and businesses should not treat it as an instant shortcut to AI search visibility.
Its real importance is architectural.
As AI agents increasingly interact directly with websites, providing clean, structured and discoverable information may become an important part of building an AI-ready website.
LLMS.txt is still evolving, but V2 gives developers and website owners a much clearer framework for experimenting with that future.