{"id":2641,"date":"2026-08-20T12:39:34","date_gmt":"2026-08-20T10:39:34","guid":{"rendered":"https:\/\/mairateam.com\/?post_type=glossary&p=2641"},"modified":"2026-08-20T15:55:19","modified_gmt":"2026-08-20T13:55:19","slug":"llms-txt","status":"publish","type":"glossary","link":"https:\/\/mairateam.com\/en\/glossary\/llms-txt\/","title":{"rendered":"llms.txt"},"content":{"rendered":"

llms.txt<\/strong> is a text file at the root of a website that tells language models, in plain language, who you are, what you do and which content matters for understanding your brand. Unlike robots.txt it forbids nothing, it offers a curated index instead. No search engine guarantees it is read.<\/p>\n

In short<\/h2>\n\n\n\n\n\n\n\n
What it is<\/strong><\/td>\nA \/llms.txt<\/code> file in markdown: who you are, what you do, and links to the content that matters<\/td>\n<\/tr>\n
What it is not<\/strong><\/td>\nNot a mandatory standard and not a replacement for a sitemap. Nobody guarantees a model will read it<\/td>\n<\/tr>\n
Why do it anyway<\/strong><\/td>\nIt is the one place where you set the interpretation of your own brand, numbers and facts included<\/td>\n<\/tr>\n
How many links<\/strong><\/td>\nDozens of links. Ten is more of a declaration, a dump of the whole site is no longer a curated selection<\/td>\n<\/tr>\n
When to skip it<\/strong><\/td>\nWhen the site has nothing to offer. Without content and without original numbers the file is an empty gesture<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

What belongs in it<\/h2>\n

The structure is free, but this order works: a short brand summary, hard facts as bullets (team size, markets, volumes, years), an overview of services, then links to key pages and results. The first two paragraphs decide the outcome, because that is where a model takes the company's identity from.<\/p>\n

The facts have to match the rest of the site<\/h3>\n

When llms.txt states different numbers than your about page or a directory profile, the model resolves the contradiction on its own terms, usually not in your favor. Consistent numbers across sources are cheaper than any optimization.<\/p>\n

Links should be selected, not exhaustive<\/h3>\n

The point of the file is curation. Dumping the whole site merely repeats the sitemap and makes nothing easier for the model.<\/p>\n

From our own practice: a curated selection, not a site dump<\/h2>\n

We wrote our own llms.txt by hand as a curated index: who we are, which pages carry the services, which articles hold our own data and numbers, and in which language. We switched off the automatically generated version, because it mixed languages and listed important pages next to technical ones.<\/p>\n

The practical range is a few dozen to roughly a hundred links. A machine dump of the whole site is not a curated selection; a file with ten links, on the other hand, is a declaration rather than a useful index. llms-full.txt is not worth treating as a priority.<\/p>\n

Common mistakes<\/h2>\n