Blog · AI
llms.txt v2: useful for AI agents, irrelevant for Google ranking
llms.txt v2 makes it easier for AI agents to access Markdown content, without improving Google ranking. Here’s what to really expect.
llms.txt v2: useful for AI agents, irrelevant for Google ranking
The llms.txt file has just been updated to version 2. This evolution brings something tangible: a cleaner way for AI agents to find the Markdown versions of documentation.
But it changes nothing about Google ranking.
Google is very clear on this point. llms.txt does not improve ranking in standard results, nor visibility in AI Overviews or AI Mode. Google Search states that it ignores this file for its search systems.
So why discuss llms.txt v2?
Because its value lies elsewhere. It primarily concerns AI agents, development assistants, and tools that need to crawl extensive documentation. For a SaaS, API, or service with hundreds of technical pages, this distinction is far from trivial.
llms.txt v2 is not a new SEO trick. It’s a technical building block to make certain content easier for machines to discover and use.
What llms.txt v2 actually changes
The first version of llms.txt already allowed placing a Markdown file on a site to present its key content to language models.
The idea was simple. Instead of asking an AI agent to crawl an entire site, its HTML, navigation, and many pages, you provided a more readable entry point.
The version 2 of llms.txt, published by Jeremy Howard in August 2026, goes further. It primarily formalizes the relationship between an HTML page, its Markdown version, and the llms.txt file describing the documentation.
Two mechanisms become particularly interesting.
rel="alternate" indicates the Markdown version of a page
An HTML page can now explicitly signal that an equivalent Markdown version exists.
For example:
<link
rel="alternate"
type="text/markdown"
href="/docs/authentification.md">
An agent browsing the HTML page can thus immediately detect a version designed to be easier to read and process.
The principle is similar to alternative relationships already used elsewhere on the web. Here, the type text/markdown specifies the nature of the resource provided.
The llms.txt v2 specification accepts two URL conventions. A page like /docs/tutorial.html may have an equivalent /docs/tutorial.html.md or /docs/tutorial.md.
This update addresses a simple issue from the first version: knowing that documentation exists was not always enough to know where to find its Markdown version.
rel="describedby" links the page to the llms.txt file
The second new feature allows specifying which llms.txt file describes the resource being accessed.
<link
rel="describedby"
href="/docs/llms.txt">
This detail becomes useful on complex sites.
An llms.txt file can now be placed at the domain root, but also in a subdirectory of the site. A file /docs/llms.txt can thus cover the documentation located under /docs/. When multiple applicable files exist, v2 stipulates that the agent uses the most specific one.
For a company with multiple products or documentation sets, this approach is cleaner than a single file containing hundreds of links.
llms.txt v2 can also work without modifying the HTML
Another key point for technical teams: this information does not necessarily need to be added to the <head> of each page.
Version 2 also allows its use in the HTTP header Link.
Link: </docs/page.md>; rel="alternate"; type="text/markdown", </docs/llms.txt>; rel="describedby"
This method can be managed at the server or CDN level. It also works with non-HTML resources.
For substantial documentation, this is a real advantage. A team can gradually expose Markdown versions without modifying the entire template system.
We then begin to understand the true purpose of llms.txt v2: to simplify AI agent navigation through structured documentation.
Not to artificially improve SEO.
No, llms.txt v2 is still not a Google ranking factor
This is the key point to remember before any project.
Google has issued a particularly explicit guideline regarding llms.txt. To appear in Google Search, AI Overviews, or AI Mode, it is not necessary to create a new machine-readable file, an AI-dedicated file, or a specific Markdown version.
Google even states that maintaining an llms.txt file does not help or harm visibility or rankings, since Google Search ignores it.
The following shortcut should therefore be avoided:
“My site has a llms.txt file, so it will be better understood and ranked by Google.”
This cause-and-effect link is not established.
The same caution applies to GEO. There is currently no evidence to suggest that adding llms.txt will automatically result in a brand being cited more often in ChatGPT, Gemini, Claude, or Perplexity.
The file may make it easier for certain tools to access information if they choose to use it. This is different from a ranking signal or a guarantee of citation.
The distinction between SEO and GEO becomes essential as generative engine usage grows.
Why does Google Lighthouse check for llms.txt, then?
This is where the topic can become confusing.
On one hand, Google Search states that llms.txt is not necessary.
On the other, Chrome Lighthouse has included an llms.txt audit in its agentic navigation category since May 2026.
There isn’t really a contradiction.
Google Search and Chrome Lighthouse address two different needs.
Lighthouse presents llms.txt as an emerging convention to provide AI agents with a machine-readable summary. Its audit mainly checks that an existing file does not return a server error. If no llms.txt exists and the server simply responds with a 404, the audit is marked as not applicable. The file remains optional.
Google Search, however, focuses on ranking and displaying content in its results.
The presence of llms.txt in Lighthouse should therefore not be interpreted as SEO validation.
It is rather a sign that the web is beginning to prepare for navigation performed directly by agents.
Where llms.txt v2 becomes truly interesting
Imagine a SaaS platform’s documentation.
It includes an authentication page, an API reference, integration guides, code examples, usage limits, and hundreds of technical pages.
A human can use the internal search engine or navigation.
An AI agent, however, must identify the right resource, fetch its content, and distinguish useful text from all the technical elements surrounding the page.
A clean Markdown version reduces this friction.
The v2 specification then makes this version easier to discover.
The benefit is therefore particularly strong for companies with extensive technical documentation, a public API, a help center, product specifications, or a knowledge base intended for automated tools.
The official proposal also notes that llms.txt is currently primarily used in software documentation and developer agent contexts.
For a ten-page showcase site, the potential gain is much lower.
llms.txt v2 and GEO: a technical building block, not a strategy
This is probably the most important distinction.
The GEO, or Generative Engine Optimization, aims to increase a brand’s or content’s ability to be understood, used, cited, or recommended by generative systems.
This relies on far more than a single file.
A page must provide accurate information. Key facts must be easy to identify. The company must be clearly associated with its business. Its content must be accessible. Its expertise must be verifiable. Its important pages must also be interlinked.
llms.txt v2 can improve part of this technical layer.
It does not create the authority that content lacks.
It does not turn poor documentation into a reliable source.
It does not replace good internal linking.
And it certainly does not replace the fundamentals of organic search.
Google notes that its generative AI features remain tied to its core Search systems. To be eligible as a link in AI Overviews or AI Mode, a page must be indexable and appear in Google Search with a snippet.
This is also why SEO and GEO should not be pitted against each other. We explore this shift in more detail in our analysis of AI Overviews and Google SEO.
Should you upgrade your site to llms.txt v2?
For a company with extensive documentation, yes, the topic is worth exploring.
But priorities matter.
If pages are poorly structured, hard to access, lacking in information, or isolated from the rest of the site, llms.txt v2 won’t fix the issue.
However, if the documentation is already solid, offering Markdown versions and properly linking them with rel="alternate" is a coherent technical evolution.
You must then ensure the Markdown accurately reflects the page’s useful content, that URLs remain stable, that HTTP responses are correct, and that the system can be maintained as the documentation evolves.
The llms.txt file itself should remain concise. Its role is to guide the agent to the right resources, not to duplicate the entire site.
To understand its basic structure before moving to v2, our complete guide to the llms.txt file remains the best starting point.
The real signal behind llms.txt v2
The most interesting part of this update isn’t its impact on Google.
It’s what it reveals about the Web’s evolution.
For years, a website was primarily designed for two visitors: humans and indexing bots.
Now, a third must be considered: the AI agent browsing a site to complete a task.
This agent might search for information, consult documentation, compare options, or use an API. It needs clear, accessible, and easy-to-navigate content.
llms.txt v2 attempts to provide a simple convention for this new use case.
The proposal is still emerging and may evolve further. But it’s becoming more concrete: Markdown URLs are better defined, their discovery is clarified, and the relationships between pages and llms.txt files are easier to automate.
This is a development worth monitoring closely for product and technical teams.
Not a reason to rewrite your SEO roadmap.
Key takeaways for your SEO and GEO strategy
llms.txt v2 primarily helps AI agents discover Markdown content. It is not a Google ranking factor.
For a documentation-heavy site, implementation may make sense as its cost remains low and its technical role becomes clearer.
To improve Google rankings or appear more in Search generative features, priorities remain elsewhere: content quality, accessibility, indexability, architecture, internal linking, expertise, and the ability to precisely address user needs. Google confirms this in its official guidelines.
For better visibility in other engines and AI assistants, llms.txt can be part of a broader GEO strategy. It should then be treated as an infrastructure layer, not a shortcut to citation.
At Scroll, we work precisely on this boundary between SEO, GEO, development, and real-world AI use cases. The challenge is not to add every new file proposed by the ecosystem, but to identify technical evolutions that truly benefit your product, users, and the agents now crawling the Web.
Questions fréquentes
What is llms.txt v2?
llms.txt v2 is the second version of the llms.txt proposal. It notably clarifies how to link a web page to its Markdown version using rel="alternate" and how to indicate the associated llms.txt file using rel="describedby".
Does llms.txt improve Google SEO?
No. Google states that Search does not use llms.txt, and its presence neither improves nor harms a site’s visibility or ranking.
Does llms.txt help appear in AI Overviews?
Not directly. Google states that no specific AI file is required to appear in AI Overviews or AI Mode. SEO fundamentals remain the priority.
Should llms.txt v2 be installed on all sites?
No. Its value is most apparent for sites with documentation, APIs, knowledge bases, or large amounts of structured content. For a small showcase site, other SEO and GEO tasks will often have a far greater impact.
