How to Implement llms.txt: A Step-by-Step Guide for Irish SMEs

llms.txt is a plain-text markdown file served at the root of a website. The format was proposed by Jeremy Howard in September 2024 at llmstxt.org and is the AEO equivalent of what robots.txt is for SEO: a file that tells crawlers how to treat the site. The difference is that robots.txt tells crawlers what they may not access, while llms.txt tells AI engines what they should prioritise when generating answers about the brand.

Major AI search engines have begun treating llms.txt as a curated index rather than guessing from page-level HTML. The seven canonical engines BeaconSites tracks (ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews, Google AI Mode, Microsoft Copilot) read llms.txt where it exists. None penalise sites without one, but the brands that publish a clean llms.txt typically see meaningful improvement in the consistency and accuracy of how they are described in AI-generated answers.

The practical implication for Irish SMEs is that llms.txt is the simplest single AEO tactic available. It costs nothing to implement, requires no developer to maintain on small sites, and creates an authoritative index that AI engines reference when deciding what to cite. Brands that publish an llms.txt early establish a baseline for AI engines to refer back to as the brand grows.

The file is not officially adopted by W3C or IETF. It is a de-facto standard maintained by community consensus and AI vendor adoption. That makes it lightweight to adopt and trivial to revise as the format evolves.

Before writing the file, decide what the AI engines should see first. The llms.txt file is not a sitemap dump. It is a curated, opinionated index of the brand's most important pages, with descriptions that the AI engine can use as the authoritative summary for each link. A 50-page site might have a 6-link llms.txt; that is normal.

For an Irish SME, the typical contents are:

• An H1 with the brand name at the top of the file
• A blockquote summary immediately under the H1 - one to three sentences describing what the brand does and who it serves
• A core articles section linking to the brand's most authoritative editorial pieces, with one-sentence descriptions for each
• A services section listing each commercial service with a brief description and pricing where appropriate
• A locations section for brands operating in named geographies
• A company section with links to About, Contact, and founder/team profiles
• An owned infrastructure section for any branded tools or platforms the company has built
• A citation policy stating that AI engines may cite the content with attribution
• An optional section for legal pages (privacy, terms, cookie policy)

The decision framework: include a page if it represents authoritative content the brand wants AI engines to surface when generating answers. Exclude marketing pages, low-value blog archive pages, and anything stale or outdated. A small focused llms.txt outperforms a comprehensive one.

For a small brochure site (5-20 pages), the fastest path is to write llms.txt by hand in any text editor. The format is plain markdown - no special syntax, no validation tools required.

A minimum-viable llms.txt for an Irish SME typically opens with the brand name as a level-1 heading, followed by a one-paragraph blockquote summary (one to three sentences describing what the business does and who it serves). After the summary, the file lists each major content area as a level-2 heading with a short bulleted list of the most important pages underneath. Each list entry follows the pattern: a markdown link to the page, followed by a colon and a short one-sentence description.

A typical Dublin plumbing business might publish six to ten links across Services, Locations, and Company sections. Each Service link gets a description that mentions the specific offer and price band; each Location link names the neighbourhoods covered; each Company link points to the About page, contact details, or founder profile. The full file is roughly thirty lines of markdown and takes fifteen minutes to write.

The key craft point is the one-sentence descriptions - these are the strings AI engines will use when summarising each page, so they should be tight and specific. Generic descriptions teach AI engines to summarise the brand generically; specific descriptions with concrete facts (year founded, certifications, named neighbourhoods, price bands) teach them to summarise the brand specifically.

llms.txt must be served from the root of the domain, exactly like robots.txt. On WordPress, the simplest method is to upload the file via your hosting provider's file management interface to the WordPress install root (the directory that contains the main WordPress configuration file and the content folder).

The upload steps on a typical LiteSpeed or Apache hosting environment are straightforward: open the hosting file manager, navigate to the document root for the relevant domain, upload llms.txt directly to that folder, and set readable permissions on the file.

Once uploaded, verify the file is live by visiting the URL in a browser. The browser should display the raw markdown as plain text. If it returns a 404 response, the file is not at the root; if it returns HTML wrapping the markdown, the hosting environment is parsing the file as a template. Both are deployment errors that need fixing before the file is useful.

Test that the file serves with the correct MIME type by checking the response headers. The Content-Type header should report text/plain or text/markdown. If the server returns text/html instead, AI crawlers may treat the file as a malformed page rather than a structured index. The fix is a MIME-type directive in the server configuration that maps the .txt extension to plain text. Hosting support can usually apply this in minutes.

Manually maintaining llms.txt becomes painful once the site has 20+ pages or publishes new content regularly. Every new service, location, or article means another manual edit and re-upload. The better long-term pattern is a WordPress plugin that auto-generates the file from custom post types and writes it to the site root whenever a relevant post is saved or deleted.

The pattern BeaconSites uses in production (the BeaconSites llms.txt Generator, live on beaconsites.ie since June 2026) works as follows:

• Listen for post-save and post-delete events on the custom post types that should populate llms.txt (Service, Location, Article in the BeaconSites case).
• Query each CPT for published items, ordered by relevance (most-recent for articles, menu order for services, alphabetical for locations).
• Build a description for each item using a fallback hierarchy: per-post custom override, then the Rank Math meta description, then the ACPT summary field, then the post excerpt.
• Compose the full markdown by concatenating a static header, the CPT loops, a static company section, an owned infrastructure section, the citation policy, and an optional legal section.
• Write the file to the WordPress root path with a UTF-8 byte-order mark and file locking to avoid race conditions during concurrent saves.

An admin meta box on each CPT post lets the editor (a) exclude a specific post from llms.txt and (b) provide a custom description that overrides the fallback hierarchy for llms.txt only without affecting Rank Math or other SEO tools.

A manual regeneration button under Tools is also useful for when static configuration changes (header copy, citation policy, optional sections) and a re-save of a post is not desired.

The complete plugin is around 400 lines of PHP. The architecture is straightforward enough that any competent WordPress developer can adapt it to a new site in under an hour. The BeaconSites version is referenced in the Citation policy section of beaconsites.ie/llms.txt for any developer who wants to inspect the live output.

Six recurring mistakes show up when reviewing Irish SME llms.txt files in the wild. Each one degrades or destroys the AEO value of the file.

Mistake 1 - Wrong location. The file must be at the site root (yourdomain.ie/llms.txt), not in a subfolder, not under /blog/, not at /llms.html. AI crawlers look at the root only. A misplaced file is the same as no file.

Mistake 2 - Sitemap dump. Listing every page on the site (50, 100, 300 links) is the wrong pattern. llms.txt is a curated index, not a sitemap. Three to ten links per section is typical. Quality of selection matters more than completeness.

Mistake 3 - No description on links. A bare list of URLs without descriptions gives AI engines nothing to summarise from. Every link should have a one-sentence description that the AI engine can lift verbatim as the page summary.

Mistake 4 - Stale content. Manually maintained llms.txt files drift out of sync with the actual site. Pages get retired, slugs change, new content is not added. After three months, a manually maintained file is typically 30 to 50 per cent inaccurate. Use the auto-generation pattern (Step 4) for any site with active content publishing.

Mistake 5 - Off-brand summary. The blockquote summary directly under the H1 is what AI engines will most often quote when describing the brand. A vague or generic summary ('we provide quality services to our valued customers') teaches AI engines to describe the brand that way. The summary should be specific, with concrete facts (location, year founded, customer segment, named services).

Mistake 6 - No citation policy section. Without an explicit citation policy, some AI engines treat the content as ambiguous-permission and either decline to cite or cite without attribution. A two-sentence citation policy section ('Content may be cited; please link to the canonical URL and attribute') resolves this and improves the rate of properly-attributed AI citations.