We’ve added LLMs.txt files to your website not because it’s the latest “Get Ranked Quick™” scheme, but because we optimize for end users, and agentic browsers and LLMs are now at least a fraction of yours.
Though, like Google, they’re only intermediaries, we want them to want you, not your competitors.
ChatGPT reportedly accounts for 12% of search-style traffic in the U.S., according to Graphite.io, April 20, 2026.
Not much, really, but we’re also seeing double-digit percentage increases in valid inbound leads from users on LLMs (“would you like me to generate a version you can send directly to Company X?” is, well, quite revealing), and even Google has gone full-AI, as announced May 19th at Google I/O 2026.
We’re not keen on AI content creation to begin with, and have found your lived experience and expertise do a much better job of communicating reality than AI does.
But LLMs.txt, as a simple fact-based document, isn’t AI-generated. Rather it’s human written, and intended to be ingested by large language models and agentic browsers. These are good things and, if we’re being honest, these pages are now concise summaries of facts that didn’t previously exist in any one location onsite.
An exercise we’re already seeing pay off positively—despite a larger study last year that suggests adding these files won’t increase AI citations.
Far different than churning out AI content at scale (bad), there’s nothing fundamentally wrong with an LLMs.txt file, even if never used by LLMs. Though in our build out process we found these files can harm your site if done wrong (these are competitor sites, but because you typically rank higher and more consistently than they do, we have to assume that any file with irrelevant and incorrect information is going to harm not help. That seems pretty basic. Specifically, the files contained links that don’t lead anywhere, restaurant menu mentions for service providers that don’t sell food, and non-interpretable contact and other key information.)
Core Data Chronology
| Who | What | When |
|---|---|---|
| Jeremy Howard | proposes llms.txt as a website-standardization doc intended to help LLMs use a website. | 2024-09-03 |
| SE Ranking | conducts study of nearly 300,000 domains, finds 10.13% have an llms.txt file, and that the existence of one “didn’t make a domain more likely to be cited by AI models.” | 2025-11-07 |
| Google for Developers (“Lighthouse agentic browsing scoring”) | lists the presence of an llms.txt among its scoring metrics for agentic browsing, in its documentation for developers: “llms.txt: Checks for the presence of a machine-readable summary at the domain root.” | 2026-05-05 |
| Google for Developers (“llms.txt”) | suggests the llms.txt should “provide a concise Markdown summary of your site’s purpose and key links.” Calls the file type an “emerging convention” “used to provide a machine-readable summary of a website’s content, specifically designed for LLMs and AI agents.” Without which, “agents may spend more time crawling the site to understand its high-level structure and primary content.” | 2026-05-05 |
| Google Search Central | publishes its first-ever AIO documentation guide for websites, and writes that, “For Google Search, you can ignore tactics like “chunking” content, creating unnecessary AI text files (like llms.txt), or pursuing inauthentic mentions.” | 2026-05-15 |
| Google I/O 2026 | announces a massive number of new AI features and integrations, but, contrary to a fabricated image [“verify llms.txt files…] and understandably so, makes no mention of llms.txt in speeches or in onscreen graphics. | 2026-05-19 |
How we Implemented
Like all we do, we strive to push the envelope as far as possible while remaining within the parameters of best-practices. Search is a highly opinionated subject. We keep abreast of it, invite opinions, and aggressively look for better ways. Aggressively statisticize.
When we make a change, large or small, we record the date we made it and watch to see what happens. Google suggests that, and we’ve found it best to being able to revert something gone wrong, or to account for what happens next.
So, we created the most substantial llms.txt files we could, kept exclusively to provable facts related to your company, and provided links (internal and external) to where these facts could be verified. We wrote them in standard markdown language, and hosted them on the root domain, which you’ll find at “https://example.com/llms.txt”
We’re also aware of a rising concern by LLM providers of the type that Microsoft security researchers called AI Recommendation Poisoning (February 10, 2026), so we steered entirely clear of anything interpretable as “suggesting or directing LLMs”, promotional, or any opinion or claim.
What we Implemented
That means these files are simply:
- A small text-only file written in standard markdown language.
- Our best interpretation of all that has been published regarding the llms.txt file by its creator, the major search engines, and LLM companies.
- A summary of the core subject, structure and content on your website.
- A concise and complete list of verifiable facts related to your business.
- In a manner we believe would be most helpful to agentic browsers and large language models.
“While websites serve both human readers and LLMs, the latter benefit from more concise, expert-level information gathered in a single, accessible location.” – Jeremy Howard, “The /llms.txt file,” September 3, 2024