llms.txt standard: should your SaaS blog have one?

The llms.txt standard now has an actual paper trail. There are on-the-record Google quotes, dated adoption studies, and a clear map of which crawlers fetch it and which ignore it completely. That's new. For most of 2025, the honest answer to "does llms.txt work" was a shrug. In 2026, you can check.

This is the evidence-based companion to our llms.txt guide, which covers the mechanics of writing and hosting one. That post assumes you've decided to add a file and shows you how. This one answers the question that comes first: is it worth doing at all, and for whom.

What is the llms.txt standard, and who proposed it?

The llms.txt standard is a proposed convention for a plain Markdown file, served at /llms.txt on a site's root domain, that gives language models a curated map of a site's most important pages. Jeremy Howard published the proposal on behalf of Answer.AI on September 3, 2024, reasoning that "site authors know best, and can provide a list of content that an LLM should use."

The spec is looser than most people assume. The only required element is a single H1 with the site or project name. Everything else, the blockquote summary, free-text notes, and the H2-delimited sections of links, is optional. That looseness is part of why adoption looks inconsistent in the wild: two sites can both claim a "valid" llms.txt while looking nothing alike.

It's worth saying plainly that this is a proposal, not a ratified web standard. No browser vendor, standards body, or major AI lab has adopted it as a requirement. It spread the way robots.txt originally did, informally, because enough people agreed it was a reasonable idea. Two years later, that's still roughly where it sits.

Does any AI crawler actually honor the llms.txt standard? The 2026 evidence

Not for the mainstream use case most teams assume. Google has said on the record that it isn't pursuing llms.txt for Search or its AI features. Where it does get used is narrower and more technical: developer-documentation sites and coding agents, not the consumer chat interfaces most SaaS teams picture when they add one.

What Google says on the record

Google's position is now unambiguous, and it comes from two separate teams saying the same thing in different ways. At a Search Central Live Deep Dive Asia-Pacific session, Google's Gary Illyes and Amir Taboul confirmed that Google is not pursuing llms.txt, as Search Engine Journal reported. Google's AI optimization guide, updated June 15, 2026, goes further and explicitly lists llms.txt among the tactics not needed for AI Overviews, AI Mode, or its other generative Search features.

John Mueller has made the same point more colorfully. Search Engine Journal reports that Mueller compared llms.txt to the keywords meta tag, the metadata field spammers stuffed in the 1990s until Google stopped reading it entirely, and noted that no AI services actually use the file to inform their answers. That's a pointed comparison from someone at Google: a tag any site can add, that once mattered enough to spam, and now sits in your source unread.

There's one genuine crack in that consensus, and it comes from a different part of Google. Chrome's Lighthouse 13.3 shipped an experimental "Agentic Browsing" audit category that checks whether a site serves an llms.txt file, on the theory that without one, an autonomous browsing agent burns extra steps just orienting itself on a page. That's not Google Search endorsing llms.txt for ranking or AI Overviews. It's the Chrome team hedging for a different, still-emerging use case: agents that browse on a user's behalf, not the AI Overview that answers a search query directly. Same company, two products, two different bets.

What OpenAI, Anthropic, and other labs actually do with it

The honest pattern is a split between developer tooling and consumer products, and it holds across every major lab. OpenAI's own Agents SDK documentation site serves an llms.txt index of its docs pages, and markdown versions of individual pages are available by appending .md to the URL, a convenience aimed squarely at coding agents and RAG pipelines that want clean text without HTML to parse.

Anthropic's developer documentation does the same. We checked the live file directly: docs.claude.com/llms.txt now redirects to platform.claude.com/llms.txt, which serves a real, current index covering the API reference, SDKs, admin and compliance docs, and more, in eleven languages. It's exactly the kind of machine-readable sitemap the original proposal described, built for an AI assistant or coding tool trying to understand the shape of a large docs site fast.

What neither company does is put one on its consumer-facing property. Adoption tracking from caseyrb.com found that none of the major consumer-facing AI apps, ChatGPT, Claude.ai, Gemini, OpenAI.com, Anthropic.com, or Character.ai, serve their own llms.txt, while several of those same companies' developer-documentation sites do. That's the pattern in one sentence: the file lives where coding agents and API integrators go looking, and it's absent where ordinary users show up.

How many sites have actually shipped one

Two independent 2026 trackers both land in the same range, which is reassuring given how differently they sample. caseyrb.com's adoption study found 5.61% of the top 10,000 websites had a valid llms.txt as of June 2026, up from 1.04% in July 2025, a 5.4x increase in twelve months. The rate is close to flat across size tiers: 6.28% for the top 1,000, 5.17% for the top 100,000, and 5.07% for the top 1,000,000, which suggests the file is spreading roughly independent of how large or well-resourced a site is.

Rankability's tracker, scanning the Tranco top 1,000 domains, found 8.7% had a valid llms.txt as of the July 17, 2026 update, up from near zero a year earlier. Because a chunk of any top-1,000 domain list is backend infrastructure, CDNs, DNS providers, cloud APIs, with no public website at the root, Rankability also isolated the 549 domains that were reachable as an actual public site. Among just those, the rate was 15.8%.

One number is worth reading carefully rather than at face value. The caseyrb.com study notes that Shopify pushed llms.txt to every store on the platform by default in April and May 2026, which is why Shopify's own adoption rate, 78.1%, is so much higher than WordPress's 8.7%. WordPress's number reflects actual site owners choosing to add the file. Shopify's reflects a platform default nobody opted into. Blending the two into one "adoption rate" would flatter the trend; keeping them separate is the more honest read, and it's the reason we cited WordPress, not Shopify, as the comparison point above.

llms.txt vs robots.txt vs sitemap.xml: what it doesn't replace

llms.txt doesn't replace robots.txt or sitemap.xml, because none of the three does the other's job. robots.txt is a permission file that every major crawler is built to check before it fetches anything. sitemap.xml is an exhaustive URL list search engines use for discovery. llms.txt is a curated, voluntary reading list that, per the evidence above, close to nothing currently reads.

The distinction that matters most for a SaaS blog is the one between access and description. Getting your robots.txt rules right for AI bots decides whether a crawler is even allowed to fetch your pages in the first place, training crawlers like GPTBot and ClaudeBot versus search and citation crawlers like OAI-SearchBot and PerplexityBot. That's the file with teeth. llms.txt has none. A crawler that ignores robots.txt disallow rules is doing something adversarial. A crawler that ignores llms.txt is doing nothing wrong at all, because the file was never a request for compliance, just a suggestion nobody is obligated to take.

How to generate an llms.txt file, with a real example

Given the low stakes, the right amount of effort here is small: a few minutes, not a project. Here is a fast path if you want one live today, plus what a real file looks like once it's written.

Free llms.txt generator tools

If your site runs on a documentation framework, a plugin can generate the file automatically from your existing content, which is the least effort option and stays current as pages change. The active options worth knowing:

• vitepress-plugin-llms: generates an llms.txt (and per-page markdown) automatically for a VitePress-built docs site, following the llms.txt specification.
• docusaurus-plugin-llms: the same idea for Docusaurus sites, generating llms.txt from your existing docs structure at build time.
• llmstxtgenerator.org: a web-based generator with a free tier that scans up to 20 URLs, crawls the pages you point it at, and outputs a structured file for sites with no docs framework to plug into.
• llms_txt2ctx: a CLI and Python module from the original proposal's toolset that parses an existing llms.txt and expands it into full LLM context, useful for validating a file you wrote by hand.

A blog on a hand-rolled site or a simple CMS will usually get a cleaner result writing the file by hand, since the whole point is a short, opinionated list, not an automated crawl of everything.

A minimal llms.txt file example

Here is a real, minimal file for a SaaS blog. It covers the required H1, the optional summary, and one link section, which is enough to be useful without becoming a second sitemap.

# Acme Analytics

> Acme Analytics is a privacy-first product analytics tool for small SaaS
> teams. We help founders see which features drive retention without
> shipping user data to third parties.

## Docs

- [Quickstart](https://acme.com/docs/quickstart/): Install and send your first event.
- [API reference](https://acme.com/docs/api/): Endpoints, auth, and rate limits.

## Blog

- [Retention metrics that matter](https://acme.com/blog/retention/): The four numbers we track.

Absolute URLs, a description after each link rather than a bare title, and a summary that says what you do without marketing adjectives. Our llms.txt guide walks through the fuller version with product and pricing sections, plus the rules that keep a longer file useful instead of bloated.

Where to host it and how to verify it's live

Save the file as llms.txt and serve it at your domain root, so it resolves at https://yourdomain.com/llms.txt as plain text or Markdown. Static sites drop it in public/; most frameworks need either a static-file route or a small handler that returns the content with a text/plain content type. If your blog already runs a CI pipeline that gates PRs on broken links and schema, adding an llms.txt reachability check to that same workflow is a natural extension: a curl step that confirms /llms.txt returns a 200 catches the file silently 404ing after a deploy, the same way it already catches a broken canonical.

There's no registry to submit to and no approval step. Open the URL in a browser once it ships. If you see your raw Markdown, you're done.

Should your SaaS blog bother with the llms.txt standard yet? The verdict

Yes, add it, but for the honest reason, not the hoped-for one. It's not worth it because Google, OpenAI, or Anthropic will suddenly rank or cite you differently. The evidence above says the opposite: Google is on the record saying no, and where the file does get read, it's coding agents and developer-docs tooling, not the AI Overview or ChatGPT answer a SaaS blog is usually chasing.

Add it because the cost is a few minutes and the downside is zero. There's no crawl budget to spend, no ranking risk, and no maintenance burden beyond updating the list when you ship or retire a major page. That's a rare trade in SEO, where most tactics carry at least some opportunity cost. This one barely does.

What llms.txt can't do is stand in for the work that actually earns a citation, from either a search engine or a model. A curated map is worthless if the pages it points to are thin, stale, or quietly wrong, which is exactly why fact-checking every claim before it ships matters more than the map itself. If you're building the SaaS content engine llms.txt is a footnote to, SEO for SaaS is the fuller playbook, and answer engine optimization covers the broader shift toward AI-cited answers that llms.txt was supposed to solve and, so far, mostly hasn't.

Lyra fact-checks every claim and link before a post ships, which is the part of AI citation that actually moves the needle, and she serves her own llms.txt at trylyra.ai so the file practices what this post preaches.

Try Lyra → · Talk to the founder