robots.txt told crawlers where they may go. llms.txt tells AI systems what they are looking at. Here is what the file is, what it honestly can and cannot do, and how to write one — from an agency that runs one on this very site.
The proposal was published in September 2024 by Jeremy Howard of Answer.AI, with a simple observation: language models increasingly read websites, but websites are written for humans — navigation, banners, scripts, layout. The proposal: give sites a single, predictable location where the essentials are stated in clean Markdown that a model can ingest directly. Adoption since has been voluntary and uneven — strongest among documentation and developer sites, growing among businesses that care about how AI systems describe them.
It is deliberately simple: an H1 with the site or business name, a blockquote summary, then sections of links with one-line descriptions. A minimal example for a Dubai business:
# Al Noor Dental Clinic > Family dental clinic in Jumeirah, Dubai. General dentistry, > orthodontics and hygiene, serving patients in English and Arabic. ## Services - [Services & prices](https://example.ae/services): full treatment list - [Book an appointment](https://example.ae/book): online booking ## Key facts - Location: Jumeirah, Dubai, UAE - Hours: Sat-Thu 9:00-21:00 - Contact: +971 X XXX XXXX
A companion file, llms-full.txt, can carry the full content of key pages inline for systems that want everything in one request. The index file alone is the sensible starting point for most SMEs.
| Claim | Verdict |
|---|---|
| "It guarantees AI systems recommend you" | False. No file guarantees inclusion in any AI answer. |
| "It improves Google rankings" | No evidence. It is not an SEO tactic and should not be sold as one. |
| "AI systems are required to read it" | False. Consumption is voluntary and varies by system. |
| "It ensures AI gets your facts right when it does read your site" | Its real job. A clean, unambiguous statement beats layout-guessing every time. |
| "It costs almost nothing and carries no risk" | True. One text file; no downside beyond keeping it current. |
The honest frame: llms.txt is one layer of a machine-readability stack. The layer with broader, documented uptake is schema.org structured data; llms.txt complements it by adding narrative context no schema field carries. Doing the second without the first is fine; doing neither leaves AI systems guessing — and in our July 2026 Dubai pilot, 71% of testable restaurant sites had no structured data at all.
What the business is, where it operates, who it serves, in 2–4 sentences. This blockquote is what a model will quote — write it the way you want to be described.
Services, prices, booking or contact, location. One line of description each. Leave out campaign pages and clutter — curation is the point.
Location, hours, contact, languages served. These are the facts assistants most often get wrong when inferring from page layouts.
The file must live at yourdomain.com/llms.txt — not in a subfolder. Whenever hours, services or contact
details change, update it in the same edit as the website. A stale llms.txt teaches AI systems wrong facts with
perfect clarity, which is worse than none.
Add LocalBusiness or Organization schema.org markup to the website itself. The two together — structured fields plus narrative file — cover both ways machines read a business.
This site's own file is public: miraireach.marketing/llms.txt (and the expanded llms-full.txt). It states what the company is, lists every service and research page with a one-line description, and carries the key facts — the same structure recommended above, running in production. We add the same layer for clients as part of AI search optimization work, where it ships alongside schema.org markup, crawler-access checks and indexing.
llms.txt is a plain-text Markdown file placed at the root of a website (yourdomain.com/llms.txt) that describes, in a format written for AI systems, what the site and the business are, with links to the pages that matter — so language models can understand the business without having to interpret its web design.
No. It is a public proposal from September 2024 that has seen growing voluntary adoption, particularly among documentation and developer sites. No AI company is obliged to read it, and consumption varies by system. It is best treated as low-cost machine-readability insurance alongside schema.org structured data — not as a placement mechanism.
There is no evidence llms.txt affects classic Google rankings, and it should not be sold as an SEO tactic. Its purpose is different: giving AI systems a clean, unambiguous statement of what the business is, so that when those systems do retrieve the site, they extract correct facts instead of guessing from page layout.
llms.txt is the concise index: what the business is, plus curated links with one-line descriptions. llms-full.txt is the expanded companion carrying the full content inline, so a system can ingest everything in one request. A small business site usually starts with llms.txt alone; both files should agree with the website and each other.
If AI visibility matters to the business, yes — because it costs almost nothing and removes ambiguity. In mirAIreach's July 2026 Dubai pilot, 71% of testable restaurant websites had no structured data of any kind; an llms.txt plus schema.org markup puts a business in the small minority whose facts AI systems can read directly rather than infer.
We check your structured data, llms.txt, AI mentions, Maps presence and website health, scored 0–100 with the gaps identified. Private to you. No obligation.
Get your free AI visibility scan