6 CASE STUDIES

Real-World llms.txt Examples — Annotated

See how Stripe, Anthropic, Cursor, and Cloudflare structure their llms.txt files. Learn what works, what doesn't, and the strategy lesson behind each decision.

Why Production Examples Outperform Spec Theory

Reading the llms.txt specification tells you what the rules are. Studying real production files tells you how the rules translate into decisions — and that gap is where most llms.txt implementations fail. A spec-compliant file can still be strategically useless if it lists the wrong pages, uses vague descriptions, or ignores how AI systems actually consume structured content. The six case studies below close that gap by showing you exactly how production-grade llms.txt files are structured, what tradeoffs their authors made, and what you should steal for your own implementation.

What You Will Learn Here

  • Structural patterns — how top companies organize H1, blockquote, and H2 hierarchy
  • Description quality — why link descriptions determine AI citation accuracy
  • Curation strategy — when to list 10 pages vs 40 pages and why
  • Anti-patterns — the exact mistakes that make llms.txt files unusable

Annotated Case Studies

Six profile evaluations — four real adopters, one counter case, and one gold standard.

ANT

Anthropic

AI CLUSTER PIONEER

Industry: AI Safety & Research

llms.txt 
                                            
                                                # Anthropic

> AI safety company that builds reliable, interpretable, and steerable AI systems. Creator of Claude — a helpful, harmless, and honest AI assistant.

## Core Product
- [Claude](https://www.anthropic.com/claude): AI assistant for conversation, analysis, and content creation
- [Claude API](https://docs.anthropic.com/): Build with Claude via API access

## Documentation
- [API Reference](https://docs.anthropic.com/en/docs): Complete developer documentation
- [Prompt Engineering](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview): Optimize your prompts for Claude
- [Model Comparisons](https://docs.anthropic.com/en/docs/about-claude/models): Choose the right model for your use case

## Research
- [Research Overview](https://www.anthropic.com/research): Safety research and publications
- [Constitutional AI](https://www.anthropic.com/research/constitutional-ai): Our approach to AI alignment

What They Did Well

  • Single H1 with company name — clean, spec-compliant opening
  • Blockquote summary immediately after H1 — gives AI systems a concise identity signal
  • Absolute URLs throughout — every link is fully resolvable
  • Logical H2 section grouping — Core Product, Documentation, Research
  • Concise link descriptions — each bullet explains what the user finds at the destination
  • File stays focused on developer-critical paths — no marketing fluff

What Could Be Improved

  • Missing optional context paragraph between blockquote and first H2 — could add a sentence about their mission or scale
  • No explicit version or last-updated indicator — makes freshness hard to assess
  • Research section could benefit from one or two more landmark paper links for topical authority

The Strategy Lesson to Steal

Anthropic proves that restraint wins. A focused file of 10–15 high-signal links outperforms an exhaustive 100-link catalog. AI systems weight clarity over completeness — every entry in their file answers 'what is this and why does it matter?' in a single line.

STR

Stripe

DEV DOCS LEADER

Industry: Financial Infrastructure & Payments

llms.txt 
                                            
                                                # Stripe

> Financial infrastructure platform for the internet. APIs for online payments, billing, fraud prevention, and business financial management. Trusted by millions of businesses worldwide.

## Getting Started
- [Quick Start](https://docs.stripe.com/quickstart): Accept your first payment in under 10 minutes
- [API Reference](https://docs.stripe.com/api): Complete endpoint documentation with examples
- [SDKs & Libraries](https://docs.stripe.com/libraries): Official libraries for every major language

## Core Products
- [Payments](https://docs.stripe.com/payments): Accept payments online, in person, or through platforms
- [Billing](https://docs.stripe.com/billing): Subscriptions, invoicing, and recurring revenue
- [Connect](https://docs.stripe.com/connect): Marketplace and platform payment solutions
- [Radar](https://docs.stripe.com/radar): AI-powered fraud protection

## Resources
- [Status Page](https://status.stripe.com): Real-time system uptime and incidents
- [Changelog](https://docs.stripe.com/changelog): API version updates and migration guides

What They Did Well

  • Industry-leading documentation hierarchy — mirrors how developers actually search for information
  • Blockquote summary includes trust signal ('millions of businesses') — AI systems extract credibility markers
  • Product sections map to user intent — Getting Started, Core Products, Resources
  • Every link has a purpose-driven description — not just page titles
  • File structure mirrors their docs navigation — cognitive consistency for AI and humans

What Could Be Improved

  • Could add a 'Use Cases' H2 section — AI systems frequently answer 'how do I use Stripe for X' queries
  • No pricing or comparison page links — these are high-intent pages AI users ask about
  • Missing integration-specific links — developers ask about Stripe + Shopify, Stripe + WordPress frequently

The Strategy Lesson to Steal

Stripe's file succeeds because it mirrors developer search patterns, not site architecture. When AI systems answer 'how do I accept payments with Stripe,' they need the Quick Start and Payments pages — and Stripe puts those first. Structure your llms.txt around user questions, not your sitemap hierarchy.

CUR

Cursor

AI CODING PLATFORM

Industry: AI-Native Development Tools

llms.txt 
                                            
                                                # Cursor

> AI-first code editor built for pair programming with AI. Built on VS Code, enhanced with intelligent code completion, chat, and codebase understanding.

## Getting Started
- [Download](https://cursor.sh/): Download Cursor for macOS, Windows, and Linux
- [Documentation](https://docs.cursor.sh/): Complete setup and usage guides

## Features
- [Tab Completion](https://docs.cursor.sh/tab): AI-powered code suggestions as you type
- [Chat](https://docs.cursor.sh/chat): Ask questions about your codebase in natural language
- [Composer](https://docs.cursor.sh/composer): Multi-file AI editing and code generation
- [Cursor Rules](https://docs.cursor.sh/rules): Configure AI behavior for your project

## Resources
- [Blog](https://cursor.sh/blog): Product updates and engineering insights
- [Changelog](https://cursor.sh/changelog): Version history and feature releases

What They Did Well

  • Extremely concise — respects the AI crawler's token budget
  • Feature-focused structure — each H2 section maps to a user task (getting started, using features, staying updated)
  • Blockquote includes the key differentiator ('AI-first') and the technical foundation ('VS Code')
  • Descriptions are action-oriented — 'Download Cursor for macOS, Windows, and Linux' not just 'Download page'

What Could Be Improved

  • Only 8 links total — could add migration guide, pricing, and enterprise pages
  • Missing 'About' or 'Why Cursor' section — AI systems need context for comparison queries
  • No pricing or plan information — high-intent conversion pages are absent
  • Could benefit from a 'Migrating from VS Code' section — top user question

The Strategy Lesson to Steal

Cursor demonstrates the minimum viable llms.txt — enough structure to be useful, concise enough to be consumed quickly. But their file leaves citation opportunities on the table. When an AI system answers 'Cursor vs Copilot,' it needs comparison content that Cursor's llms.txt doesn't surface. Add pages that answer your top comparison queries.

CLO

Cloudflare

ENTERPRISE EDGE

Industry: Enterprise Edge Network & Security

llms.txt 
                                            
                                                # Cloudflare

> Global cloud platform providing CDN, security, reliability, and performance solutions. Operates one of the world's largest networks spanning 300+ cities in 100+ countries.

## Core Products
- [CDN & Performance](https://www.cloudflare.com/performance/): Global content delivery and website acceleration
- [Security](https://www.cloudflare.com/security/): DDoS protection, WAF, and bot management
- [Workers](https://developers.cloudflare.com/workers/): Serverless compute platform running at the edge
- [DNS](https://www.cloudflare.com/dns/): Authoritative DNS with built-in security

## Developer Platform
- [Workers Documentation](https://developers.cloudflare.com/workers/): Build and deploy serverless functions
- [Pages](https://developers.cloudflare.com/pages/): Deploy static sites and JAMstack applications
- [R2 Storage](https://developers.cloudflare.com/r2/): Object storage with zero egress fees
- [D1 Database](https://developers.cloudflare.com/d1/): Serverless SQL database at the edge

## Resources
- [Developer Docs](https://developers.cloudflare.com/): Complete platform documentation
- [Status Page](https://www.cloudflare.com/status/): Real-time system health and incident tracking
- [Blog](https://blog.cloudflare.com/): Engineering deep-dives and product announcements

What They Did Well

  • Scale signal in blockquote — '300+ cities in 100+ countries' is a credibility marker AI systems extract
  • Dual-audience structure — Core Products for business decision-makers, Developer Platform for engineers
  • Workers platform gets prominent placement — it's their most AI-referenced product
  • Developer Docs link serves as a hub — AI systems can follow it for deeper content
  • Descriptions include value propositions — 'zero egress fees' not just 'object storage'

What Could Be Improved

  • File could get long if all products were added — current selection is curated but incomplete
  • Missing pricing and plan comparison pages — frequently asked in AI queries
  • No 'Solutions by Industry' section — enterprise buyers ask 'Is Cloudflare right for my industry?'
  • Could add AI-specific product links — Workers AI and AI Gateway are growing in AI search relevance

The Strategy Lesson to Steal

Cloudflare shows how enterprise-scale companies should approach llms.txt: curate ruthlessly. With hundreds of products, listing everything would create an unusable 200KB file. Instead, they select the 12–15 pages that answer 80% of AI-generated queries. The lesson: your llms.txt is a curated index, not a sitemap replacement.

BAD

Anonymized Bad Example

COUNTER CASE

Industry: SaaS (Identity Redacted)

llms.txt ⚠ 
                                            
                                                # Our Product

# Welcome to Our Platform

> We make things better.

## Features
- /features: Features
- /pricing: Pricing
- /about: About Us
- /blog: Blog
- /contact: Contact
- /docs: Documentation
- /docs/api: API Reference
- /docs/getting-started: Getting Started
- /docs/sdk: SDK Downloads
- /docs/sdk/node: Node.js SDK
- /docs/sdk/python: Python SDK
- /docs/sdk/ruby: Ruby SDK
- /docs/sdk/go: Go SDK
- /docs/sdk/java: Java SDK
- /docs/sdk/php: PHP SDK
- /docs/sdk/dotnet: .NET SDK
- /docs/sdk/rust: Rust SDK
- /docs/sdk/swift: Swift SDK
- /docs/sdk/kotlin: Kotlin SDK
- /docs/sdk/dart: Dart SDK
- /docs/changelog: Changelog
- /docs/errors: Error Codes
- /docs/rate-limits: Rate Limits
- /docs/auth: Authentication
- /docs/webhooks: Webhooks
- /docs/pagination: Pagination
- /docs/filtering: Filtering
- /docs/sorting: Sorting
- /privacy: Privacy Policy
- /terms: Terms of Service
- /careers: Careers
- /press: Press Kit
- /partners: Partners
- /investors: Investor Relations
- /legal: Legal
- /security: Security
- /compliance: Compliance
- /status: Status Page
- /support: Support
- /community: Community
- /feedback: Feedback
- /roadmap: Product Roadmap
- /changelog: Product Changelog
- /brand: Brand Guidelines
- /accessibility: Accessibility
- /sitemap: Sitemap
- /404: Error Page

What They Did Well

  • At least they have an llms.txt file — many sites still don't
  • Uses H2 sections for grouping — basic structure is present
  • Includes a blockquote summary — though it's too vague to be useful

What Could Be Improved

  • Multiple H1 tags — violates the single-H1 spec rule and confuses AI heading hierarchy parsing
  • All relative paths — every link uses /path instead of https://domain.com/path — breaks spec compliance
  • No descriptions on any link — AI systems cannot determine what value each page provides
  • Vague blockquote — 'We make things better' provides zero context for AI comprehension
  • Exhaustive page listing — 47 links with no curation signals, making it impossible for AI to determine priority
  • Hidden truncation risk — at this size, the file approaches limits where AI systems may truncate processing
  • Duplicate changelog entry — /changelog appears twice under different sections
  • Includes utility pages — /404, /sitemap, /accessibility add no value for AI content understanding

The Strategy Lesson to Steal

This counter case demonstrates every anti-pattern in one file. Multiple H1 tags break heading hierarchy. Relative paths make links unresolvable. Missing descriptions force AI systems to guess page content. An exhaustive link list with no curation signals tells AI 'everything is equally important' — which means nothing is. The fix: one H1, absolute URLs, descriptions on every link, and curate to your top 20 pages.

GLD

LLMsTXTApp

GOLD STANDARD

Industry: AI Search Optimization Tools

llms.txt ★ 
                                            
                                                # LLMsTXTApp

> Free llms.txt generator, validator, and domain checker. Build spec-compliant llms.txt files in seconds, validate structure and compliance, and check any domain's AI readiness. No signup required.

## Tools
- [llms.txt Generator](https://llmstxtapp.com/generator): Generate a spec-compliant llms.txt file from any URL instantly
- [llms.txt Validator](https://llmstxtapp.com/validator): Validate your llms.txt for structure, URLs, H1 rules, and file size
- [llms.txt Checker](https://llmstxtapp.com/checker): Check if any domain has an llms.txt file and see compliance score

## Guides
- [Ultimate Guide to llms.txt](https://llmstxtapp.com/llms-txt-guide): Complete 6,000-word implementation reference with spec rules and examples
- [How To Create llms.txt](https://llmstxtapp.com/create-llms-txt): Step-by-step creation guide with common mistakes and fixes
- [llms.txt Templates](https://llmstxtapp.com/templates): Copy-paste templates for blogs, SaaS, e-commerce, and developer docs
- [llms.txt Examples](https://llmstxtapp.com/examples): Annotated real-world examples from Stripe, Anthropic, and Cloudflare

## AI SEO Resources
- [What Is AI SEO?](https://llmstxtapp.com/what-is-ai-seo): Complete guide to optimizing for ChatGPT, Perplexity, and Gemini citations
- [What Is GEO?](https://llmstxtapp.com/what-is-geo-generative-engine-optimization): Generative Engine Optimization explained with ranking factors
- [ChatGPT Ranking Factors](https://llmstxtapp.com/chatgpt-ranking-factors): Evidence-based factors that influence ChatGPT citations
- [How To Get Cited by ChatGPT](https://llmstxtapp.com/how-to-get-cited-by-chatgpt): 10-step playbook for ChatGPT visibility

## About
- [Blog](https://llmstxtapp.com/blog): In-depth guides on llms.txt, GEO, and AI search optimization
- [About](https://llmstxtapp.com/about): Our mission, team, and why we built LLMsTXTApp

What They Did Well

  • Single H1 with brand name — clean spec-compliant opening
  • Blockquote summary includes value proposition, key features, and trust signal ('No signup required')
  • Absolute URLs on every link — fully resolvable by any AI crawler
  • Every link has a keyword-rich description — AI systems can match user queries to page content
  • Logical H2 hierarchy — Tools, Guides, AI SEO Resources, About — mirrors user intent clusters
  • Curated to 14 high-value pages — focused, not exhaustive
  • Internal links create citation pathways — each guide links to related tools and resources
  • File stays well under 50KB — no truncation risk for any AI system

What Could Be Improved

  • Could add a 'Comparisons' section for llms.txt vs robots.txt and vs sitemap.xml queries
  • Missing Perplexity and Claude-specific guide links — growing query clusters in 2026
  • Could include last-updated timestamp for freshness signaling

The Strategy Lesson to Steal

The gold standard is not about perfection — it's about intentionality. Every element in this file serves a purpose: the H1 identifies the brand, the blockquote answers 'what is this,' the H2 sections cluster content by user intent, and every link description contains keywords that match real AI search queries. Build your llms.txt the same way: start with identity, add context, then curate by query intent.

Build Your Own Based on These Examples

Use the free generator to create a spec-compliant llms.txt file tailored to your site structure — informed by the same patterns that make these production examples effective.

Free Generator →

Frequently Asked Questions

What makes a good llms.txt example?
A good llms.txt file has exactly one H1 heading, a blockquote summary after the H1, organized H2 sections, absolute URLs for all links, bullet-point link lists, and stays under 50KB. The best examples also include concise descriptions for every linked page and prioritize high-value content over exhaustive page listings.
How many pages should I list in my llms.txt?
Most production llms.txt files list between 15 and 40 pages. The key is quality over quantity — include pages that provide unique value and context to AI systems. Stripe and Anthropic both keep their files focused on developer-critical documentation rather than listing every marketing page.
Should I include every page of my website?
No. Only include pages that represent your site's core value proposition. Exclude duplicate content, thin pages, admin pages, and temporary promotional pages. A focused file of 20 high-value pages outperforms an exhaustive list of 200 low-value pages for AI comprehension and citation.
What is the most common llms.txt mistake?
The most common mistake is using relative URLs instead of absolute URLs. The llms.txt specification requires all links to use the full https:// URL format. Other frequent errors include multiple H1 tags, missing blockquote summaries, and files exceeding the 50KB size limit.
Do AI models actually read llms.txt files?
Claude Code and similar AI agent tools have confirmed they read llms.txt files when navigating documentation sites. For general AI chatbots like ChatGPT and Perplexity, llms.txt reading is not confirmed but the file serves as a structured content index that can improve how AI systems understand your site when they do encounter it.
Can I use relative paths in my llms.txt?
No. The llms.txt specification requires absolute URLs starting with https://. Relative paths like /about or /docs/api break the file's portability and make it impossible for AI crawlers to resolve the correct destination. Every link must include the full domain and protocol.

Explore More Tools

Continue optimizing your llms.txt workflow with these resources.

Generator Build a custom llms.txt file tailored to your site structure. Validator Check your llms.txt for errors, missing fields, and best practices. Guide Learn the complete llms.txt specification and implementation strategies.