Article summary

There are several ways to help a visiting AI understand what your business is, what it offers and who you help. Some of these, such as JSON-LD, live as invisible text in your website's code. Others are how you structure you sites content.

Then there's an llms.txt file.

An llms.txt file gives AI systems a short, curated guide to the most useful content on a website.

There is currently no strong evidence that publishing one improves AI rankings, citations or recommendations. Major AI platforms have not documented it as a visibility signal.

However, implementation is inexpensive, the practical downside is negligible when the file is accurate and kept up to date, and the possible future benefit could considerably exceed the cost.

For most SMEs, the sensible position is not to prioritise llms.txt over more important GEO work, but to publish a simple, well-maintained file once the fundamentals are in place.

What is an llms.txt file?

An llms.txt file is a page on your website that your human visitor will never know or care about. It's purely for visiting AIs. It will live at a URL such as yourwebsite.com/llms.txt.

It gives those visiting AI systems a concise description of the website and links to the pages the site owner considers most useful or authoritative. Think of it like a site summary and quick-start guide.

What problem does llms.txt solve?

The idea came from Jeremy Howard in September 2024, and the intention was to make websites easier for language models to navigate, especially large, complex sites.

Websites are designed primarily for humans and browsers, and that creates several problems for AI systems:

• Important information may be distributed across many pages.
  
  

• Website navigation does not necessarily reveal which pages are authoritative.
  
  

• HTML pages can contain large amounts of boilerplate.
  
  

• Similar pages may compete for attention.
  
  

• An agent may waste time and processing capacity discovering the site’s structure.
  
  

An llms.txt file attempts to reduce that friction.

Instead of forcing an AI system to interpret the entire site, the business can provide a short list of canonical resources covering areas such as:

• services;
  
  

• products;
  
  

• expertise;
  
  

• research;
  
  

• case studies;
  
  

• policies;
  
  

• documentation;
  
  

• and frequently asked questions.
  
  

That's the thinking behind llms.txt files, to improve navigational convenience.

But do they lead to greater public AI visibility?

Does llms.txt improve Generative Engine Optimisation (GEO) performance?

Short answer: we don't know.

There is no reliable evidence that publishing an llms.txt file currently improves:

• AI crawling;
  
  

• retrieval;
  
  

• citations;
  
  

• rankings;
  
  

• entity recognition;
  
  

• or recommendation likelihood.
  
  

Google has stated that Google Search does not use llms.txt and that publishing one will neither help nor harm Search visibility.

OpenAI, Anthropic, Perplexity, Microsoft, Apple and Meta document crawler access, robots directives, structured data and other mechanisms, but not files.

But that does not mean we should forget about this option.

In fact, many of these organisations use llms.txt files on their own sites (OpenAI, Anthropic, Perplexity, Microsoft, Meta and Cloudflare).

Why publishing an llms.txt may still make sense

A basic llms.txt file can be made quite quickly. It is a plain-text Markdown file, requires no special software and usually has negligible hosting cost.

When it contains only public, canonical pages, the practical risk is also extremely low.

The cost-benefit calculation therefore looks different from many speculative GEO tactics:

• The current measurable benefit is uncertain.
  
  

• The implementation cost is very small.
  
  

• The downside is minimal.
  
  

• Future AI systems may adopt the convention more widely.
  
  

• Custom agents and private retrieval systems can already use it.
  
  

The potential upside does not need to be particularly large to justify the initial effort.

What should an llms.txt file contain?

An effective file should be selective. It's an executive summary, not a map, so it shouldn't reproduce the entire XML sitemap or list every blog post.

A typical file may include:

• the business or website name;
  
  

• a short description;
  
  

• the main service or product pages;
  
  

• important expertise or methodology pages;
  
  

• original research;
  
  

• strong case studies;
  
  

• key policy or support pages;
  
  

• and a small number of authoritative articles.
  
  

The file should direct AI systems towards the pages that best explain:

• who the business is;
  
  

• what it offers;
  
  

• who it serves;
  
  

• what it knows;
  
  

• and what evidence supports its claims.
  
  

  
  

A short sample llms.txt file

Here's an example of what an llms.txt might look like for a fictional nursing home.

# Pembroke House Dublin

> Pembroke House is a private residential care and retirement home in Dublin offering long-term residential care, assisted living, respite care and specialist nursing support in a discreet, comfortable setting.

Pembroke House serves older adults and families seeking high-quality private care in Dublin. The home provides individually planned support, private accommodation, nursing oversight, dining, activities and family liaison.

## Care Services

- [Long-Term Residential Care](https://www.pembrokehouse.ie/residential-care)
- [Assisted Living](https://www.pembrokehouse.ie/assisted-living)
- [Respite Care](https://www.pembrokehouse.ie/respite-care)
- [Nursing Care](https://www.pembrokehouse.ie/nursing-care)
- [Dementia Support](https://www.pembrokehouse.ie/dementia-care)

## Accommodation and Daily Life

- [Private Rooms and Suites](https://www.pembrokehouse.ie/accommodation)
- [Dining and Nutrition](https://www.pembrokehouse.ie/dining)
- [Activities and Social Programme](https://www.pembrokehouse.ie/activities)
- [Gardens and Facilities](https://www.pembrokehouse.ie/facilities)

## Choosing a Care Home

- [How to Choose a Private Retirement Home](https://www.pembrokehouse.ie/guides/choosing-a-retirement-home)
- [Questions to Ask During a Care Home Visit](https://www.pembrokehouse.ie/guides/care-home-visit-questions)
- [Residential Care Costs and What Is Included](https://www.pembrokehouse.ie/costs)
- [Admissions Process](https://www.pembrokehouse.ie/admissions)

## Standards and Evidence

- [Care Standards](https://www.pembrokehouse.ie/care-standards)
- [Clinical Governance](https://www.pembrokehouse.ie/clinical-governance)
- [Staff Qualifications and Training](https://www.pembrokehouse.ie/our-team)
- [HIQA Registration and Inspection Information](https://www.pembrokehouse.ie/hiqa)
- [Family Testimonials](https://www.pembrokehouse.ie/testimonials)

## About Pembroke House

- [About the Home](https://www.pembrokehouse.ie/about)
- [Our Care Philosophy](https://www.pembrokehouse.ie/care-philosophy)
- [Location in Dublin](https://www.pembrokehouse.ie/location)
- [Contact Pembroke House](https://www.pembrokehouse.ie/contact)

## Optional

- [Frequently Asked Questions](https://www.pembrokehouse.ie/faqs)
- [News and Advice](https://www.pembrokehouse.ie/articles)
- [Arrange a Private Visit](https://www.pembrokehouse.ie/book-a-visit)

The purpose is not to catalogue the entire website, but to identify the most useful canonical resources.

Who is most likely to benefit?

The strongest use cases are websites with substantial bodies of organised information, including:

• API documentation
  
  

• software documentation
  
  

• technical knowledge bases
  
  

• research libraries
  
  

• product-support centres
  
  

• educational resources
  
  

• and large specialist content repositories
  
  

These sites create a genuine navigation problem for AI agents and are why llms.txt files were suggested in the first place.

The case is weaker for a small brochure website with ten or fifteen well-linked pages. Such a site should already be easy to navigate through conventional crawling and internal linking.

Even so, a small business may reasonably publish a basic file because the cost is low and the format may gain wider adoption.

What should businesses do before creating one?

An llms.txt file should not displace more important GEO work.

Before creating one, a business should make sure that:

• important pages can be crawled (robots.txt)
  
  

• the business is clearly identifiable as a distinct entity
  
  

• service and product pages contain substantive information
  
  

• internal links reveal the relationship between topics (this helps your SEO as well)
  
  

• claims are supported with evidence
  
  

• structured data is accurate and AI digestible (answer capsules)
  
  

• business information is consistent across platforms
  
  

• and the website contains information gain rather than recycled category-level content.
  
  

Your llms.txt file is a nice-to-have, but the above take priority.

Are there any disadvantages?

The risks are small rather than literally nonexistent.

A poorly maintained file may:

• link to deleted or outdated pages;
  
  

• promote weak pages over authoritative ones;
  
  

• duplicate an already clear navigation system;
  
  

• or become inconsistent with the live website.
  
  

These problems are easy to avoid.

Keep the file short, use canonical URLs and update it whenever important pages are added, removed or substantially changed.

Do not use it to publish private information, crawling permissions, licensing terms or training restrictions. Those functions are not part of the original proposal.

Should your business publish an llms.txt file?

For most SMEs, the practical answer is yes, but only after the real work of Generative Engine Optimisation (GEO) has been done.

Closing takeaway

llms.txt is not a proven AI-ranking mechanism.

It is a developing convention that makes important website content easier for AI systems and agents to find and interpret.

No business should rely on it.

But for the small amount of work involved, many businesses have little reason not to use it.

_____________

A first step to improving your AI visibility is a professional audit. You can book yours HERE.

If you'd like to discuss working together on your AI visibility, click HERE to reach out.