What llms.txt Is, Why It Matters, and What Goes in Yours

The content mill around llms.txt has two modes.

Mode one: "Implement llms.txt immediately or be invisible to AI forever."

Mode two: "Major AI platforms don't use llms.txt. It's useless. Skip it."

Both are wrong in the ways that content mills are usually wrong — oversimplified into a prescription that generates a click, then a conversion.

The actual situation is more nuanced. llms.txt is a proposed standard with real structural value, limited current platform adoption, and meaningful indirect benefits that the evangelists don't explain well and the dismissers don't see at all.

Here's what llms.txt actually is, why you should probably ship one, and what belongs in it.

What llms.txt Actually Is

llms.txt is a proposed standard — introduced by Jeremy Howard of fast.ai in September 2024 — for a markdown file placed at the root of your domain. The file summarizes what your site is about and provides a structured index of your most important content, formatted for large language models to consume efficiently.

The naming convention parallels existing web standards: robots.txt for search crawlers, sitemap.xml for search engines, security.txt for security researchers, now llms.txt for AI systems.

The file lives at yoursite.com/llms.txt and is typically paired with llms-full.txt, a longer version containing your full canonical content in markdown.

The structural template:

# Your Company Name

> One-sentence description of what you do and who you serve.

Additional context paragraph explaining your product,

audience, and the kind of content available on the site.

## Core Documentation

- [Link to page](url): Description of what's there

## Product

- [Link to page](url): Description

## Content

- [Link to key article](url): What it argues

Simple. Low-friction. Designed to be readable by both humans and machines.

The Honest Current State

Here's where most content on llms.txt gets inaccurate.

As of mid-2026, no major AI platform has officially committed to using llms.txt as a ranking or retrieval input. Google has publicly stated their systems don't currently use it. Anthropic, OpenAI, and Perplexity haven't published commitments either.

The adoption data reflects this ambiguity. Empirical studies have found no measurable citation uplift from llms.txt adoption. Adoption rates among top websites are near zero. The proposed standard exists in the uncomfortable middle ground of being widely discussed and narrowly used.

So if llms.txt doesn't currently move AI visibility metrics, why implement it?

Two reasons. Neither is what the evangelists say.

Why llms.txt Matters Anyway

Reason one: the exercise forces strategic clarity.

To write a good llms.txt, you have to answer three questions that most B2B companies have never answered cleanly:

• In one sentence, what does your company actually do?

• What are the most important pieces of content about you, in what order?

• How do your product, services, and content map to the categories a buyer would use to find you?
  

Most marketing teams spend months on positioning exercises and still can't answer question one in fewer than 40 words. The constraint of llms.txt — short, structured, machine-parseable — forces compression. That compression is strategically useful regardless of whether AI platforms ever consume the file.

If you can't write your llms.txt, you don't have clarity. AI reads your whole site the same way it would read your llms.txt — as a collective answer to "what is this company?"

A website whose owner couldn't write the llms.txt version is a website AI can't confidently describe.

Reason two: low-cost forward compatibility.

llms.txt is cheap to implement. An hour of work for most sites. Low maintenance overhead. Zero downside if AI platforms never adopt it. Meaningful upside if adoption accelerates in 2026-2027, which is plausible but not certain.

This is the stance the intellectually honest practitioners hold: llms.txt is insurance against a plausible future state, at a cost low enough that the expected value is positive even if the probability of adoption is under 50%.

What Actually Goes in a Useful llms.txt

If you're going to ship one, ship a good one.

The mediocre auto-generated versions are worse than no file at all — they signal to any AI that does read them that you don't understand what you do.

Start with a clean one-sentence summary. Not your tagline. Not your mission statement. The actual sentence a reasonable stranger would use to describe what your company does and who for. Example: "Acme helps mid-market finance teams automate month-end close."

Add a short context paragraph. Two to three sentences elaborating on the audience, the category, and the kind of content hosted on the site. This is where you land the positioning AI systems should remember.

Organize content by category, not by chronology. Auto-generated llms.txt files dump blog posts in date order. That's wrong. Organize by topic — Product, Pricing, Documentation, Case Studies, Thought Leadership — with your strongest content in each category linked first.

Link only your strongest content. Every link you include is a signal to AI about what you consider representative. If you include twenty mediocre blog posts and one flagship research piece, you've diluted the research piece by nineteen other signals. Be selective. The goal isn't completeness; it's signal quality.

Include third-party coverage. Most implementations omit this. Include links to your best analyst coverage, notable press mentions, podcasts your founder has appeared on, and other authoritative third-party content. AI systems treat third-party coverage as higher-weight signal than first-party content.

Write descriptions that are informative, not promotional. "How [category] works in 2026" is useful. "The definitive guide to [category]" is noise. Descriptions should tell the AI what's in the linked content, not why you're proud of it.

Include a short positioning statement. Most llms.txt templates don't have a dedicated section for positioning. Add one. Two or three sentences about what you do differently from competitors in your category. This is the text with the highest chance of being directly quoted by an AI summarizing your brand.

Keep it under 200 lines. If your llms.txt is longer than 200 lines, you're hedging. The file is meant to represent the 20% of your site that matters most. More than 200 lines means you haven't done the prioritization work.

What Doesn't Belong

A few things to specifically exclude:

• Your entire blog archive. Selection is the point.

• Boilerplate about the company. "Founded in 2015, we help businesses..." — every company has this, and AI systems pattern-match it as noise.

• Links to gated content. AI can't read gated content. Linking to it from llms.txt just exposes the gating.

• Outdated content you haven't retired. If it's on the site but isn't current, don't promote it here.

• Marketing calls-to-action. "Book a demo!" in an llms.txt file signals that you don't understand the format. The file is for machines, not funnel conversions.

The Pragmatic Recommendation

Here's the measured take.

Ship an llms.txt. Take the hour. Do it well.

Don't expect it to move your AI visibility metrics in the short term, because the data suggests it won't. Do expect the exercise to reveal positioning gaps you didn't know you had. Do treat it as cheap insurance against a future where AI platforms commit to reading it.

And when you write it, treat it like any other piece of authoritative content: write the version you'd be proud to have AI systems quote directly, because the day may come when they do.

llms.txt is one of the few marketing artifacts you'll produce this year that forces genuine strategic clarity in under 200 lines. That's worth an hour of someone's time, regardless of whether it shows up in an analytics dashboard.

The Real Question Is Upstream

llms.txt isn't the AI visibility silver bullet. It also isn't pointless.

It's a structured exercise in compressing your brand into the shape an AI system can efficiently consume, at the cost of about an hour of work, with meaningful indirect benefits and plausible long-term upside.

Most marketing teams should ship one.

Most marketing teams should also spend the hour before shipping it thinking about whether they can actually write the one-sentence description honestly.

If the answer to that is no, fix the underlying problem first.

No file will compensate for not knowing what your company is.