There's a new file making the rounds in web development circles, and most business owners have never heard of it. llms.txt is a plain-text file you place at the root of your website — like yoursite.com/llms.txt — that tells AI language models what your site is about, what content matters, and how to represent you accurately. At Nuclear Marmalade, we've started treating this as a baseline requirement for every site we build or audit. If you care about how AI assistants describe your business, this file is not optional anymore.
What exactly is llms.txt and what does it do?
llms.txt is a structured plain-text file that gives AI systems a curated summary of your website's purpose, key pages, and content hierarchy. Think of it as a handshake between your site and any LLM that crawls or references it. Instead of forcing GPT-4, Claude, or Perplexity to guess what you do by scraping random paragraphs, you hand them a clean, intentional brief. The format — proposed by fast.ai founder Jeremy Howard in 2024 — uses simple markdown: a title, a short description, and a list of your most important URLs with context. That's it. No code. No API. Just a text file that takes about 20 minutes to write properly and can meaningfully change how AI tools represent your brand to the people asking about you.
Why does this matter more than traditional SEO right now?
Traditional SEO got you ranked on Google. llms.txt gets you represented accurately in AI-generated answers — and those two things are increasingly different channels. When someone asks ChatGPT "who does AI automation for small businesses in Ireland," Google rankings don't decide that answer. The LLM's training data and live browsing behaviour do. Glen Healy has been watching this shift for the better part of two years, and the pattern is clear: businesses without structured AI-readable content are getting misrepresented or skipped entirely. Our AI consulting work keeps turning up the same problem — companies with great products but zero presence in AI-generated responses because their site gives LLMs nothing useful to grab. llms.txt is one of the simplest fixes on that list.
How is llms.txt different from robots.txt or sitemaps?
robots.txt tells crawlers what they can't access. Sitemaps tell them what pages exist. llms.txt does something neither of those do — it tells AI systems what your site means. A sitemap might list 200 URLs. An LLM doesn't know which three of those actually define your business. llms.txt solves that by letting you say: here's our core service, here's who we serve, here are the three pages that explain us best. It's editorial control over your AI representation. We built something similar in spirit when we designed AI memory systems — the principle is the same. Context given deliberately beats context guessed from scraps. Sitemaps and robots.txt aren't going anywhere, but llms.txt fills a gap they were never designed to close.
What should actually go inside an llms.txt file?
Keep it lean. The file should open with your site or company name as an H1, followed by a one-paragraph description that a smart stranger could read and immediately understand what you do. Then a ## Key pages section listing your most important URLs with a one-line description each. Some implementations also include an ## Optional section linking to more detailed docs or blog content. The whole thing should be readable in under two minutes. Here's the honest version of what we learned building ours at Nuclear Marmalade: our first draft was too long. We listed everything. That's the wrong instinct — LLMs don't reward comprehensiveness the way humans might. They reward clarity. Cut it to the 5-7 URLs that actually define you, write one tight sentence per link, and leave the rest out. Our web design process now includes llms.txt creation as a default deliverable.
Does llms.txt actually get used by AI models yet?
Honestly — adoption is still early. Not every LLM crawler reads llms.txt by default today. Perplexity has shown interest. Some custom GPT browsing setups respect it. The spec is young. But here's the thing: robots.txt wasn't universally respected overnight either, and nobody argues you shouldn't have one. The businesses building this infrastructure now are the ones who won't be scrambling in 18 months when AI-driven discovery is the primary way customers find services. We've already seen early returns in our own SEO and GEO work — sites with clear, structured AI-readable content consistently outperform comparable sites in AI-cited responses. llms.txt is one layer of that. It won't transform your visibility overnight. But it's a 20-minute investment with compounding returns, and there's no downside.
How do you actually create one — what are the steps?
Creating an llms.txt file is genuinely simple. Open a plain text editor. Line one: # Your Company Name. Line two: blank. Lines three to five: a short paragraph explaining what your business does and who it's for — write it like you'd explain yourself to someone at a conference. Then add ## Key pages and list your 5-7 most important URLs with a one-line description each. Save it as llms.txt and drop it in your website's root directory — the same place your robots.txt lives. That's the minimum viable version. If you want to go further, add an ## Optional block linking to your blog or detailed documentation. Some teams also create llms-full.txt — a longer version for models that want richer context. We cover the full setup as part of our AI agent integration services when clients want this handled end-to-end rather than DIY.
What's the one mistake most businesses will make with this?
Writing it for humans instead of for machines. This is where most people go wrong — they treat llms.txt like an About page and fill it with brand language, mission statements, and marketing copy. LLMs don't need inspiration. They need facts. "We help mid-sized logistics companies cut manual data entry using custom AI workflows" is perfect. "We're passionate about transforming the way businesses operate through innovative solutions" is useless — and honestly, it's the kind of thing that makes LLMs trust your content less, not more. The cleaner and more factual your llms.txt, the more likely an AI system is to pull it as a reliable source. Write it like you're briefing a very literal-minded assistant who has no patience for waffle. That's not a bad way to write anything, actually.
Key Takeaways
llms.txtis a plain-text file at your site root that gives AI models a curated, intentional summary of what your business does — think of it as editorial control over your AI representation- Traditional SEO and AI discoverability are increasingly separate channels; ranking on Google doesn't mean you're showing up accurately in ChatGPT or Perplexity answers
- The file itself takes about 20 minutes to write — list your 5-7 most important URLs, one sentence each, and keep the description factual and jargon-free
- Adoption by AI crawlers is still early, but the infrastructure you build now compounds — same logic as having a sitemap before every crawler required one
- Don't write it like a brochure; write it like a briefing note for someone who needs facts, not feelings
If you want Nuclear Marmalade to audit your current AI discoverability and get your llms.txt set up properly alongside the rest of your GEO strategy, see how we work — or browse the blog for more on what's changing in AI-driven search.