In the SEO world “new mandatory files” that supposedly change the rules of the game show up every now and then. llms.txt is exactly such a case: it is a proposed, experimental standard meant to make it easier for large language models (LLMs) to understand which resources on your site are key and worth reading first. In practice it is a text file in Markdown format, placed in the root directory of the domain (for example /llms.txt), which works more like a “curated map of the most important content” than like a classic file for controlling indexing.
Important: some industry materials stress that there is no confirmation that the leading models use llms.txt as an official signal when generating answers. In other words, it is not a “ranking factor” and there is no guarantee of any effect. Even so, companies in Poland and worldwide are interested in the topic, because the importance of search and recommendations “through AI” is growing, and brands want to increase the chance that AI will land on the right, up-to-date and unambiguous sources (for example the FAQ, policies, specifications, documentation, key landing pages).
Table of contents
llms.txt versus robots.txt and sitemap.xml: similar names, different jobs
The comparison with robots.txt and sitemap.xml is tempting, but it is worth breaking it down into basics:
- robots.txt: tells bots where they may or may not go (access and crawling control).
- sitemap.xml: lists URLs so that search engines can discover the structure of the site more easily.
- llms.txt: it is not there to “block” anything or to “list everything”, but to point to the most important, best prepared sources and give them context (what this is, what it is for, where to look for answers).
This distinction matters especially in e-commerce and on service sites: the structure there tends to be extensive (categories, filters, variants, pagination, language versions), while the “most important things” for customers and AI are usually a narrow set of pages: the offer, the process, the FAQ, integrations, terms and conditions, returns, payments, delivery, API documentation, pricing, contact.
What should be in llms.txt for it to make business sense
The description of the standard in practical guides points to a fairly consistent core: the name of the site, a short description and a set of links grouped by topic (sections), preferably with short annotations. Except that “links” are not yet a strategy. A sensible llms.txt should answer the question: which content has the greatest value as a source of truth when someone (a user or an AI system) asks about your company, your offer or your process?
For Web Systems, as an agency that builds and develops e-commerce, websites and applications and supports online marketing, the logical set usually looks like this:
- Offer pages (e-commerce, web and mobile applications, websites, online marketing) - that is “what you do” and for whom.
- Process pages - what the rollout, maintenance, further development, SLA, audit and roadmap look like.
- FAQ / knowledge base - answers to questions about costs, deadlines, technologies, integrations, responsibilities.
- Portfolio (with no “made-up clients”) - as proof of competence and of the context of the projects delivered (case studies, description of scope).
- Formal documents (privacy policy, cookies, terms and conditions, legal information) - because AI and users often ask about compliance matters.
Does llms.txt work: facts, skepticism and honest expectations
One thought runs through most descriptions: the implementation is cheap, but the effect is not certain. Some sources say outright that there is no confirmation that the top AI models use this file as a standard mechanism, and that the topic is partly “noise” around a novelty. That matters, because a common decision-making mistake looks like this: “we will add llms.txt and AI will start quoting us”. Nobody can promise that honestly today.
On the other hand, there is also a rational “engineering” argument: llms.txt forces order and curation of content. If you treat it not as a magic SEO file but as a list of canonical sources (a single source of truth), then even if the models themselves do not parse it directly, you still improve the quality of your content ecosystem: internal linking, consistency of messaging, freshness, unambiguous service definitions. And that already supports both classic SEO and conversion.
How to implement llms.txt on a website: a practical step-by-step process
Below is a process that at Web Systems makes sense in the context of service sites and e-commerce (WordPress/WooCommerce, web applications, integrations, online marketing), with no promises of miracles but with the emphasis on quality and maintenance.
- Define the goal and the scope
Do you want to: (a) make it easier for AI and users to find “the truth about the offer”, (b) put the company documentation and FAQ in order, (c) support visibility in the AI ecosystem? If the goal is unclear, llms.txt will turn into a random list of links. - Choose the canonical subpages (10-30 links to start with)
Do not put everything in. llms.txt is meant to direct people to the “best” resources. Sources describe it as a map of the most important content, not as an index of the entire site.
For e-commerce the typical set is: delivery, returns, payments, contact, FAQ, top categories, buying guides, key landing pages. For an agency: services, process, portfolio, FAQ, contact. - Make sure those pages are “readable” and up to date
llms.txt will not fix weak content. If the offer page is vague and the FAQ does not answer real questions, then pointing AI there will only cement a weak message. - Save the file in Markdown and add thematic sections
Descriptions of the structure in practice usually come down to: an H1 with the name of the site, a short description and lists of links in sections (H2), sometimes with an “optional” part.
From a usability perspective, the sections should match intents: “Offer”, “E-commerce”, “Applications”, “Marketing”, “Support and maintenance”, “Portfolio”, “Contact and formalities”. - Put the file in the root directory and make sure it is served correctly
The standard expectation is that it is available athttps://yourdomain.com/llms.txt(with no redirects and no blocks). - Set a rhythm for updates
The biggest mistake: implement it and forget it. If you change your offer, technologies or process, or add important content (for example a new case study), llms.txt should reflect that. Sources point out that the file can be short or extensive - but the extensive version requires maintenance.
The most common mistakes that ruin the point of the implementation
- A list of everything: llms.txt then stops being a “curation” and becomes a second sitemap (and loses its value).
- Links to low quality pages: thin service descriptions, old posts, outdated price lists - both AI and people will get the wrong signal.
- No consistency with the information architecture: when llms.txt promotes different subpages than the ones you have in your navigation and internal linking.
- No measurement: if after the rollout you do not watch traffic on the pages you listed and do not monitor lead inquiries, you will not know whether the change has any impact at all.
How to approach llms.txt: a sensible proposal for e-commerce and B2B services
If your goal is a real impact on business results (site traffic, leads, sales), then llms.txt should be an addition to the three foundations that Web Systems usually takes care of in development projects anyway:
- Strong offer and informational content
Well described services for building and developing e-commerce, WooCommerce rollouts, web and mobile applications, integrations and online marketing - with specific processes, scope and responsibilities. This is the basis both for SEO and for generative answers. - Technical SEO and order in indexing
robots.txt, sitemap, canonical URLs, pagination, structured data, speed, Core Web Vitals, correct 404/301 responses - without those, every “novelty” will only be an add-on on top. - GEO/AEO as a practice, not as a slogan
If you want to be quoted in AI answers, you need content that AI wants to quote: unambiguous definitions, comparisons of options, checklists, FAQs, procedures, current information about the offer and the process. llms.txt can help here as a “table of contents for the model”, but it will not replace the work on content.
