If you have heard of llms.txt at all, it was probably described as “robots.txt for AI”. That comparison is wrong in the way that matters, and it is why so many of these files end up useless.
robots.txt is a set of instructions: here is what you may and may not crawl. llms.txt is not instructions. It is a plain-English summary of your site, written for something that is going to answer a question about your business without sending the person to your website first.
We had one. It was ten lines, written to tick a box, and it said almost nothing a machine could use. In late September we replaced it across five of our sites. The new ones run between 28 and 44 lines. Here is what went in them and why.
What the file is actually for
When someone asks ChatGPT, Perplexity or Google’s AI results “who does web design in Castle Hill”, the system answering does not read your whole site. It works from whatever it can retrieve quickly and summarise confidently.
Your homepage is written to persuade a human. It has a hero image, a tagline, some social proof and a call to action. Almost none of that is a fact a machine can state on your behalf. Your services are probably in a mega-menu. Your suburbs are probably on separate pages it may never reach. Your phone number is in the footer, possibly as an image.
llms.txt is where you write those facts down in order, in Markdown, at one predictable address: yoursite.com.au/llms.txt.
It is not an official standard and no search engine has committed to using it. It takes an hour, it costs nothing, and the exercise of writing it surfaces gaps in your actual site whether or not anything reads the file. That is a reasonable bet. Treating it as a ranking factor is not.
The structure we used
Every one of ours follows the same shape.
An H1 with the business name. Just the name. Not a slogan.
A blockquote, one sentence, saying what the business does and where. This is the single most important line in the file, because it is the sentence most likely to be quoted back at someone. Ours reads:
Web design, SEO, e-commerce and Google Ads for Western Sydney and Hills District small businesses. A Ganda Tech Services sub-brand, based in Bella Vista, NSW.
Services, geography, parent company, suburb. Four facts, one sentence, no adjectives. Compare that to whatever is currently in your hero banner.
An About section with the facts as a list. Location, services, contact, phone. Not prose — a list a machine cannot misread.
A Services section, with one line each. Every line is a link with a short description after a colon:
- [Web Design](https://cosmoswebtech.com.au/services/web-design/): custom and template website builds for local small businesses.
- [E-Commerce](https://cosmoswebtech.com.au/services/ecommerce/): WooCommerce and Shopify builds.
The description matters as much as the link. “Web Design” on its own tells a machine nothing it could not guess. “Custom and template website builds for local small businesses” tells it who you are for.
A geography section. We list our suburb service pages, then a line saying how many exist in total and pointing at the sitemap. For a business serving the Hills District and Western Sydney this is the section that decides whether you show up in an answer about a specific suburb.
A section of your own work. Ours is headed “Engineering and case-study posts (grounded in our own work, not generic advice)” and lists posts about real defects we found and fixed. This is the section most businesses leave out, and it is the one that separates you from every competitor whose blog is generic advice.
An Optional section at the end. Full blog archive, FAQ, sitemap. The convention is that a machine short on budget may skip anything under “Optional”, which makes it the right place for the things that are big rather than important.
★ Insight ─────────────────────────────────────
The reason the “Optional” heading exists is a budget problem, not a formatting one. Anything reading your file has a limit on how much it will pull in, and if you lead with a 400-item blog archive, the limit gets spent before it reaches your services. Ordering is the whole design: most-quotable sentence first, biggest list last.
─────────────────────────────────────────────────
Three mistakes worth avoiding
Dumping your sitemap into it. This is the most common version of the file and it is the least useful. A list of 400 URLs with no descriptions is a sitemap, and you already have one of those in a format machines prefer. The value of llms.txt is the curation and the one-line descriptions, which is exactly what a dump removes.
Marketing language. “We are passionate about delivering world-class digital solutions” contains no facts. It cannot be quoted, checked, or used to answer anything. Write the way you would answer a direct question from a stranger.
Letting it go stale. A file listing a service you stopped offering is worse than no file, because it is confidently wrong. Ours are checked when services change, and any link in them that does not return a 200 gets fixed before anything else.
Check it is actually being served
Two things to verify after you publish it, because both fail quietly.
Visit yoursite.com.au/llms.txt in a browser. You should see plain text. If you see your 404 page, your site is not serving files from the right directory — on most static site builders the file belongs in public/.
Then check it in a private window, because a cached copy will show you a file the rest of the world is not getting. We verified all five of ours return a live 200 before calling the job done, and that check is not optional — a file that exists in your repository and is not served is a file that does not exist.
Is it worth doing
For a local business in Western Sydney, yes, on a cost basis alone. An hour of work, no ongoing cost, and the writing exercise reliably turns up something wrong with your actual site — a service with no page, a suburb you thought you covered, a phone number that appears nowhere as text.
Just be clear-eyed about what you are buying. Nobody has proven this file lifts anything. What it definitely does is make the facts about your business retrievable in one place, in one request, in a format designed to be quoted. If AI answers keep growing as a way people find local suppliers, that is a cheap hedge.
The companion file, llms-full.txt, is where the longer index goes — and deciding what does not belong in it turned out to be the harder half. That is the next post.
Cosmos Web Tech builds and optimises websites for Hills District and Western Sydney small businesses. Cloud and IT support comes from our sister brand Cloud Geeks, apps from Awesome Apps, and we are part of Ganda Tech Services.
