Does the llms.txt map work?

Genesis

Large language models face a critical architectural limitation: their context windows are too small to process entire websites. Converting complex websites containing navigation, ads, JavaScript and CSS into LLM-friendly clean text is both difficult and imprecise.

Jeremy Howard, co-founder of Fast.ai and lecturer at the Universities of Queensland and Stanford, observed that most HTML on sites consisted of menus, tracking scripts, repetitive sections and ads - elements that eat up valuable tokens in the LLM context window without contributing substantive value. Inspired by the simplicity of robots.txt, Howard created a standard that allows site owners to provide LLM with structured, expert knowledge in a single, accessible location.

Technical specifications

The llms.txt standard defines a precise Markdown-based structure that combines human readability with programmatic parsing capabilities. The file must be in the root path of the  /llms.txt  sites and include the following sections in a specific order:

Required components:

• H1 header - name of the project or site (the only mandatory section)

Optional but recommended items:

• Blockquote - a concise summary of the project containing the key information necessary to understand the rest of the file
• Descriptive sections - zero or more markdown sections (paragraphs, lists) without headers, containing detailed information about the project and how to interpret the supplied files
• H2 sections with file lists - zero or more sections separated by H2 headers, containing lists of URLs with additional details
• Optional" section - section with special meaning. The URLs contained therein may be omitted when a shorter context is required

Link Specifications:

Each file list must contain the required hyperlink markdown name(url) , followed optionally by  :  and file notes.

The .md extension


The proposal also implies that sites containing information useful to LLM should provide a clean version of markdown under the same URL with added  .md  (or  index.html.md  for URLs without file names).

Example to implement

Below is a professional llms.txt template ready for customization and implementation:

> A concise description of your business, one or two sentences explaining the specialization, offering, or purpose of the project. This section helps LLMs understand the context of the remaining resources.
Key contextual information:
- First important note about the nature of the business or technology
- Second note specifying scope or limitations
- Third note clarifying the target audience
## Main Resources
- [Home Page](https://example.com): Introduction and latest announcements
- [API Documentation](https://example.com/api): Complete technical documentation with examples
- [Quick Start Guide](https://example.com/quickstart): Step by step onboarding for new users
- [Best Practices](https://example.com/best-practices): Proven patterns and recommendations
## FAQ and Support
- [Frequently Asked Questions](https://example.com/faq): Answers to the most common user questions
- [Troubleshooting](https://example.com/troubleshooting): Diagnostic guide for common issues
- [Contact](https://example.com/contact): Contact form and company details
## Developer Resources
- [API Reference](https://example.com/api-reference): Full endpoint documentation
- [Code Examples](https://example.com/code-examples): Practical implementations and case studies
- [Changelog](https://example.com/changelog): History of changes and updates
## Optional
- [Company History](https://example.com/history): Project evolution and milestones
- [Blog Archive](https://example.com/blog-archive): Older blog posts
- [Privacy Policy](https://example.com/privacy): Detailed information on data protection

Comprehensive guidelines can be found at: https://llmstxt.org

Important guidelines:

• File size should be limited to ~100 KB for optimal performance
• Encoding: UTF-8
• Format: pure Markdown without HTML
• All URLs should be absolute (absolute e.g. https://example.com/url), not relative
• Links must lead to active resources (avoid 404 errors)

Optional llms-full.txt

The standard also provides for an optional file  llms-full.txt , which contains full, developed documentation in a single file. While  llms.txt  acts as a table of contents,  llms-full.txt  provides the complete content of all linked documents, allowing AI systems to access the entire knowledge base in a single request.

For a list of llms.txt and llms-full.txt inspirations and maps, visit: https://llmstxt.site

Validation and testing

Check:

• Location and reachability (HTTP 200, no redirect loops)
• HTTP headers (Content-Type: text/plain or text/markdown; charset UTF-8)
• Content-Length and GZIP/Brotli compression
• Content freshness and canonical URLs
• Licensing and AI Relations attributes, or how AI models can use your content
• Correctness of Markdown syntax, structure (H1, H2, correctness of links), for example, on the page: https://markdownlivepreview.com

Manual accessibility testing:

# Check: HTTP/1.1 200 OK
# Content-Type: text/plain; charset=utf-8
# Content-Length: [size]
curl https://yourwebsite.com/llms.txt
# Verify content and formatting

Tests with LLM

• Provide URL to llms.txt to ChatGPT, Claude, Gemini models
• Ask questions about key information from your site
• Verify that AI correctly uses the indicated resources and links

Log monitoring and traffic analysis in GA4:

Watch for traffic from user-agents:

• GPTBot (OpenAI)
• Claude-Web (Anthropic)
• GoogleOther (Google AI)
• PerplexityBot (Perplexity)
• Other AI bots

The increase in visits to these bots after the implementation of llms.txt is an indicator of effectiveness. Research Insightland demonstrated 600% increase in GPTBot visits after llms.txt implementation.

10 most common mistakes

Error 1: Incorrect file location

The problem: File placed in a subdirectory instead of the root directory
Solution: Always place the file exactly under  https://yourwebsite.com/llms.txt , not in  /seo/llms.txt  neither  /ai/llms.txt .

Error 2: The required H1 header is missing

The problem: Starting a file without a level 1 header

Solution: The first line must include  # Project Name.

Error 3: Incorrect encoding

The problem: A file saved in an encoding other than UTF-8.

solution: Save the file with explicit UTF-8 encoding. In most editors: File → Save with Encoding → UTF-8.

Error 4: Size limit exceeded

The problem: File larger than 100 KB

Solution: Limit content to the most important resources. Use llms-full.txt for complete documentation.

Error 5: Incorrect Markdown link syntax

The problem:  Title (https://url)  instead of Title(https://url)  (space before parentheses)

Solution: Make sure there are no spaces between  ]  a  ( .

Error 6: No blockquote with description

The problem: Omitting the contextual description of the project.

solution: Add  > Short description  after the H1 header for better understanding by LLM.

Error 7: Dead links and 404 errors

The problem: Links leading to non-existent resources.

Solution: Regularly test all URLs with tools like broken link checker.

Mistake 8: An excess of irrelevant content

The problem: List of all subpages without prioritization.

solution: Select 5-15 most important resources. Quality > quantity.

Error 9: Not using the "Optional" section

The problem: All resources on an equal level of importance

Solution: Place secondary resources in the section  ## Optional , so that they can be overlooked with limited context.

Error 10: No post-deployment verification and no updates

The problem: Assumption that the file works without testing, no content updates.
Solution: Conduct validation and accessibility testing after each change. Add update information to the map -> Last update

Does it work?

According to the data BuiltWith october 2025, 844,473 sites has implemented the llms.txt standard. SE Ranking's analysis of 300,000 domains showed an adoption rate of 10.13%, with the majority of deployments coming from the developer tools sector, technical documentation platforms and technology companies, where AI coding assistants are critical to the business.

The llms.txt standard has been adopted by leading technology companies Anthropic (Claude documentation), Cloudflare, Stripe, Perplexity, Cursor, Solana, ElevenLabs, Hugging Face, Raycast, Yoast, DataForSEO, Zapier, Mintlify.

Case Study 1: Insightland

Results:

• Increase in GPTBot visits by 600% (from a few hundred to almost 2,000 visits)
• Perplexity-User appeared 7 times in 3 days
• TikTok, Moz, Amazon, Petalbot, Bytedance and Bing bots visited llms.txt file
• Increase in overall site traffic
• No negative impact on traditional SEO rankings

Case Study 2: WordLift

Results:

• Increase in organic traffic by ~25% after llms.txt implementation
• Better indexing by AI, richer visibility in knowledge panels and snippets

Case Study 3: Mintlify

Results:

• Reduction in time for LLM to process documentation by 40%
• Improving the accuracy of AI responses by 30%
• Thousands of technical documentation sites automatically received llms.txt files

Case Study 4: Cloudsential

Results:

• Significant increase in visibility in AI
• Cloudsential emerges as top source for ChatGPT SEO-related queries

Evidence of GEO's effectiveness

Generative Engine Optimization (GEO) study conducted by the research team showed that using content optimization strategies for generative engines can increase source visibility by up to 40% in responses generated by AI.

The most effective GEO methods:

• Adding quotes - increase in visibility by more than 40%
• Adding statistics - increase in visibility by more than 30%
• Optimizing content fluidity - significant increase
• Citing sources - significant improvement
• Technical terminology - moderate improvement

The study conducted a systematic evaluation on a benchmark GEO-BENCH consisting of 10,000 diverse queries from multiple domains.

Integration with GEO/AEO ecosystem

The llms.txt standard is a fundamental component of a broader Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) strategy. Here are the highlights of the full, holistic approach:

1. SEO is no longer a "must have", bet on GEO/AEO

Traditional SEO remains crucial, but by itself does not guarantee visibility in LLM models. A growing body of analysis shows that even brands with well-developed SEO do not always appear in the models' responses. At the same time, there are examples of brands with virtually no SEO efforts, yet they are cited by LLM, indicating that visibility in AI depends on factors other than standard search engine authority.

2. Structural data (Schema.org)

Schema markup implementation for FAQ, Article, Product and other types of content increases the likelihood of citation. Pages with complete structured data are significantly more likely to be cited by AI

3. AI-friendly content architecture

• Front-loading: Key information at the beginning of the content
• Hierarchical structure: Clear H1-H6 headings
• Letters and bullet points: Increase extractability through AI - that is, the ability of a language model to extract, recall or reproduce data
• Short paragraphs: <25 words per sentence, <100 words per paragraph

See how AI model bots see your site, instead of https://semly.ai enter your address: https://r.jina.ai/https://semly.ai

4. Authority and content

• External citations: Mentions on authoritative third parties
• Domain authority: Overall industry visibility
• Freshness of content: Pages updated in the last 12 months are 2x more likely to get citations

5. Brand Visibility Score metrics

Design: (Responses mentioning your brand ÷ Total number of responses) × 100

Supporting metrics:

• Citation Rate: % of LLM responses mentioning or linking to your brand
• Sentiment Score: (Positive + 0.5 × Neutral mentions) ÷ All mentions
• Share of Voice: % of total citations compared to competitors

AI visibility monitoring tools

The market for AI visibility monitoring tools is growing rapidly, and companies are looking for ways to understand how ChatGPT, Gemini, Perplexity or other models present their brand or products. The following overview compares Semly, Profound and Searchable. Unlike its competitors, Semly not only measures visibility in AI, but is the only tool in this comparison actively creates correct product data under LLM and data aggregators, which realistically increases the chance of brands appearing in AI recommendations.

Both Profound, as well as Searchable are advanced analytical tools, but they focus on monitoring brand visibility and reputation. Semly works differently: it combines visibility monitoring with the function of creating data under LLM, so it influences what models can see and use. At the same time, the cost of entry for Semly is many times lower than for foreign enterprise platforms. As a result, Semly acts as the first real GEO tool designed for e-commerce and brands that not only reports, but actively increases the chance of sales in new AI channels.

The future of the standard

The llms.txt standard, while experimental, is evolving toward wider adoption. Google has included llms.txt in its Agents to Agents (A2A) protocol, signaling at least experimental interest. Mintlify in November 2024 enabled the automatic generation of llms.txt for every documentation site they host, instantly adding thousands of technical documentations to the ecosystem.

Jeremy Howard in March 2025, said the vision extends beyond the current reality - an AI-first web standard where language models no longer waste tokens on redundant HTML, but can focus on relevant knowledge.

Summary

The llms.txt file represents a fundamental change in the way web content is made available to AI systems. The standard, although experimental, has achieved mass adoption (over 844,000 sites) and is producing measurable results - a 20% to 40% increase in visibility in AI responses, a 600% increase in AI bot visits and a 30% improvement in response accuracy.

Key findings:

Implementation is simple, but requires precision: Markdown structure, UTF-8 encoding, location in the root directory and correct section hierarchy are key to effectiveness.

Validation is mandatory: Use ChatGPT, for example, to validate the map before publishing.

Avoid the 10 most common mistakes: Incorrect placement, missing H1, bad coding, exceeding size limit, incorrect link syntax, missing blockquote, dead links, excess content, ignoring Optional section and lack of testing.

Integrate with GEO/AEO: llms.txt is part of a broader strategy that includes SEO, structured data, AI-friendly content architecture and brand authority building.

Monitor effectiveness: Use tools like Semly.ai to track Brand Visibility Score, Citation Rate and Share of Voice.

Update regularly: Freshness of content is critical - pages updated in the last 12 months are 2x more likely to be cited.

In an era where AI is evolving into the dominant information discovery interface, controlling how language models interpret and present your brand becomes a strategic imperative. The llms.txt standard, supported by empirical evidence and growing adoption, represents a fundamental step toward an AI-first web.

FAQ - Frequently Asked Questions

Is llms.txt an official standard?
No, llms.txt is a proposed standard created by Jeremy Howard. No major LLM provider has officially confirmed that it reads these files, but empirical evidence (increase in AI bot visits, case studies) suggests that the standard is being used in practice.

Does the implementation of llms.txt guarantee citation by AI?
No, llms.txt does not guarantee citations. However, it does increase the likelihood and relevance of citations by making it easier for AI to access key content. Studies show a 20-40% increase in visibility after implementation.

Does llms.txt replace robots.txt or sitemap.xml?
No. Each of these files has a different purpose:

• robots.txt - controlling access of indexing bots
• sitemap.xml - list of all indexable pages for search engines
• llms.txt - a curated map of key resources for AI

How often should I update llms.txt?
At least quarterly or after any significant change in site structure, addition of key content or rebranding. Content that has not been updated for more than 12 months is 2x less likely to be cited by AI.

Can I have multiple llms.txt files for different sections of the site?
Yes, the specification allows files in subpaths, such as.  https://docs.example.com/llms.txt  for the documentation section. Always keep the main file in the domain root directory.

What is the optimal size of llms.txt file?
The recommended limit is ~100 KB. Larger files may overload LLM context windows. For extensive documentation, use llms-full.txt as a supplement.

Does llms.txt affect traditional SEO?
Studies have shown no negative impact on SEO rankings. The file is neutral to traditional search engines and can indirectly support SEO by improving brand visibility in AI, which generates traffic to the site.

How to measure the effectiveness of llms.txt?
Monitor:

• Logs and bot traffic in GA4 (increase in AI bot visits)
• Tools like Semly.ai will show you your brand's visibility in AI
• Brand Visibility Score and Share of Voice
• Traffic from AI search engines in Google Analytics

Should small businesses implement llms.txt?
Yes, if you care about visibility in the AI ecosystem. Implementation is simple (1-4 hours), inexpensive, and can yield significant benefits with minimal risk.

What if I don't have the resources to create .md versions for all sites?
Focus on the most important 5-10 resources. Quality and prioritization are more important than completeness. You can link directly to HTML, although Markdown is preferred.

Glossary

LLM (Large Language Model) - a large AI language model capable of understanding and generating text from massive training data sets

Markdown - a lightweight markup language for formatting text, characterized by simplicity and readability

Context Window - limit of tokens (units of text) that LLM can process in a single query

GEO (Generative Engine Optimization) - the process of optimizing content to increase the chances of appearing in AI-generated responses

AEO (Answer Engine Optimization) - gEO synonyms; optimization for AI response engines

Parsing - the process of analyzing the structure of data by a computer program

User-Agent - bot or browser ID in HTTP headers

Schema.org - common structured data dictionary for websites

Brand Visibility Score - a metric that measures the frequency of brand mentions in AI responses

Sources

llmstxt.org - official specification of the standard

answer.AI (Jeremy Howard) - proposal and justification of the standard

llmstxt.site - index of websites that have already implemented llms.txt or full-llms.txt map

r.jina.ai/https://semly.ai - check how AI bots see your site

Research: GEO - academic research (40% increase in visibility)

Aggarwal P. et al, "GEO: Generative Engine Optimization," KDD '24, 2024 - novel research and framework for optimizing content visibility under AI generative systems.

Share:

Copy link