- Dwg
- VTA-T4
- Title
- llms.txt generator
Dwg VTA-T4Free · runs in your browser
Write your llms.txt.
A plain-text file at /llms.txt that tells assistants what you do, in your own words. Fill in the facts. The generator formats them the way the llms.txt proposal expects. Nothing leaves your browser.
llms.txt
Save it as /llms.txt at your site root, served as text/plain, then run the readiness check.
What llms.txt is for
llms.txt is a plain-text file at /llms.txt that tells AI assistants what you do, in your own words. That's it. It's optional, and it isn't a ranking switch. Nothing in it makes ChatGPT, Claude, Gemini or Perplexity name you, and anyone selling it as a shortcut to visibility is overselling.
So why bother? Because it gives a model a clean, short, unambiguous statement to read instead of guessing from a homepage full of slogans. It's a summary you control. The generator above formats it the way the llms.txt proposal expects, and it runs in the browser, so nothing you type leaves your machine.
Think of it as a fact sheet, not a pitch. If an assistant ever reads it, you want it to come away with the right idea of who you are and who you serve. Whether it gets used at all is out of your hands.
What to put in yours
The fields are simple: name, website, a one-sentence description, who it is for, facts (one per line), key pages (title, URL, one-line note) and contact. Most of the value sits in three of them.
The description should say what the business does, plainly. "Payroll software for UK construction firms" works. Anything you'd hear in an advert doesn't.
Facts come one per line, and each should be checkable. A few examples of the shape:
- Founded as a limited company and based in Manchester
- Supports CIS returns and weekly pay runs
- Integrates with the main accounting packages
Those are illustrations, so swap in your own. The test is simple: could someone verify it in under a minute? If yes, keep it. If it needs a caveat, cut it or fix it first.
Key pages are where people overdo it. Pick the handful a buyer would actually need: what you sell, how it works, who it's for, how to get in touch. Give each a title, the URL and one line saying what the page covers. "Pricing: plans and what each includes" is enough. You're pointing, not summarising the whole site.
Common mistakes
The first one is marketing adjectives. "Leading", "innovative", "best-in-class": a model can't check any of it, and neither can a reader. Strip them out. Plain nouns and verbs carry more weight here.
Second, stale facts. A file that says you support something you dropped last year is worse than no file. The llms.txt is written once and then forgotten, which is exactly how it goes wrong. Put a reminder in whatever calendar your team actually looks at, and update it when the business changes.
Third, listing every page. It's tempting, since the generator will happily take as many as you give it. But a file with two hundred links stops being a summary and turns into a sitemap, and you already have one of those. Choose the pages that matter and leave the rest.
Fourth, serving it as HTML. The file has to come back as text/plain. If your CMS wraps it in a page template, or your server sends it with the wrong content type, what an assistant receives is markup and navigation, not your clean statement. This one catches a lot of people, because the file looks fine in a browser.
After you publish it
Save the file at the site root, so it lives at /llms.txt and not in a subfolder. Then check it.
Open the URL yourself first. You should see plain text, no styling, no surrounding page. Then run the AI readiness check. It fetches your homepage, robots.txt and llms.txt once, as VisibilityToAI-ReadinessCheck, and runs six deterministic checks with no LLM involved. It'll tell you whether the file is there and readable, which is the boring but necessary part.
Also check robots.txt. A perfectly formed llms.txt does nothing useful if you've blocked the crawlers that would read it. The AI-crawler robots.txt check evaluates each AI crawler token the way that crawler reads the file: its own group if it has one, otherwise the * group, with the longest rule winning.
Then set expectations. The file describes you; it doesn't make any engine cite you. Whether assistants name and cite your company for the questions buyers ask is a separate thing to measure, and the AI citation check is a fair place to start. The generator is free, needs no account, and stores nothing.