TL;DR:
llms.txt is a curated Markdown directory hosted at your domain root that tells AI agents which pages to prioritize, bypassing crawl noise.- It works alongside
robots.txt and sitemap.xml, not instead of them. Each file serves a different audience and purpose. - Prioritize buyer-intent pages: pricing, feature comparisons, integrations, security documentation, and getting started guides.
- Page quality still determines citation rate. The file improves discoverability; on-page structure and off-page consistency determine whether agents extract and cite your content.
- Keep it updated. A stale file with outdated pricing or broken links may generate inaccurate citations, which is worse than no entry at all.
Most advice on preparing your site for AI agents focuses on schema markup. The fastest path to guiding LLMs to your core product data, however, is a simple, structured text file that tells the agent exactly where to look.
This guide provides a complete, annotated llms.txt template built specifically for B2B SaaS platforms. You will learn how to structure your core product details, educational resources, and API documentation so that engines like ChatGPT and Claude can parse and cite your brand accurately during the buyer research phase.
This article is part of our pillar guide on agent-readiness for AI search for B2B SaaS.
Why llms.txt is critical for AI search visibility
AI agents do not browse your site the way a human does. They retrieve semantically relevant passages during inference, which is the active generation of an answer. The llms.txt file can guide those agents by indicating where to look rather than forcing them to parse your full HTML.
We work across three surface areas of organic search: web search, citations, and training data. The llms.txt file operates at the real-time retrieval layer, not the training layer, signaling which pages may be prioritized when an AI composes an answer to a buyer query.
How LLMs select and cite sources
Modern AI answer engines commonly use dense passage retrieval techniques as part of the retrieval-augmented generation (RAG) pipeline. Rather than matching keywords like a traditional search index, dense retrievers encode both the query and candidate passages as numerical representations of meaning and retrieve the passages with the closest semantic alignment. Dense retrievers have been shown to outperform traditional keyword matching on passage retrieval tasks (Karpukhin et al., 2020).
The practical implication: AI agents tend to favor content that answers one question clearly in a self-contained section, not content that buries the answer across three paragraphs. An llms.txt file linking directly to pages structured this way may make it easier for the agent to retrieve and cite your brand. Our CITABLE framework is built around exactly this passage-extraction model.
Capturing B2B demand via AI citations
When a buyer asks ChatGPT or Claude which incident response platform integrates with PagerDuty, the consideration phase happens entirely inside the chat interface. If your product is not cited in that answer, you are absent from a deal you never knew existed. An llms.txt file is one component of a broader citation strategy. It ensures the agent can efficiently locate the content that should appear in those answers.
The incident.io case study covers how building a structured citation strategy changed their AI search visibility.
Complete annotated llms.txt example for B2B SaaS
The official llms.txt specification defines a Markdown file hosted at the root path of your domain. According to the specification, it contains an H1 with your site name, optional summary information, supporting detail sections, and H2-delimited file lists linking to detailed Markdown pages. The format is intentionally simple: lightweight, token-efficient, and readable by both models and humans.
Below is a complete, copy-paste template built for a generic B2B SaaS product. Replace the bracketed placeholders (ICP = Ideal Customer Profile) with your actual content.
Recommended llms.txt template structure
# [Company Name]
> [Company Name] is a B2B SaaS platform that helps [ICP] [achieve outcome].
> Founded [year]. Headquartered in [city]. Serves [X]+ customers in [verticals].
## Product
- [Product Overview](/product): Core capabilities, use cases, and integrations.
- [Pricing](/pricing): Plan tiers, per-seat costs, and enterprise terms.
- [Integrations](/integrations): Supported tools including [CRM], [HRIS], [Slack].
- [Security & Compliance](/security): SOC 2 Type II, GDPR, SSO documentation.
## Education
- [Getting Started Guide](/docs/getting-started): Step-by-step onboarding.
- [Use Case Library](/use-cases): Industry-specific templates and examples.
- [Comparison: [Product] vs [Competitor]](/compare/[competitor]): Feature-by-feature breakdown.
## API & Developer Docs
- [API Reference](/api): Full endpoint documentation with authentication.
- [Webhooks](/docs/webhooks): Event schema and retry logic.
- [SDKs](/docs/sdks): Official libraries for Python, Node, and Ruby.
## About
- [About Us](/about): Founding story, team, and mission.
- [Case Studies](/case-studies): Customer outcomes with attribution data.
Optional: [llms-full.txt](/llms-full.txt): Extended documentation for agentic use cases.
This template covers the core sections an AI agent needs to answer pricing, feature, integration, and comparison queries accurately, without padding the file with low-signal pages.
Key components of your llms.txt file
The specification at llmstxt.org requires one element absolutely: an H1 with the site or project name. Everything else is structured guidance. We recommend placing your highest-signal content in the blockquote summary directly below the H1. It should state who you serve, what you do, and one credibility signal, in two to three sentences.
H2 sections act as category headers for your file lists. Each list item is a Markdown link with a one-sentence description. Per the spec, descriptions should be usage-oriented: "reference page," "canonical definition," "quantitative data." Pages that include proof, such as statistics, methodology, or sourced claims, increase citability because agents can extract verifiable passages from them. You can also split the file using an optional llms-full.txt for extended developer documentation, keeping your primary file focused on buyer-facing content.
Comparing llms.txt to standard SEO files
The three files serve different audiences and different purposes. llms.txt is for curation. robots.txt is for access control. sitemap.xml is for discovery. They complement each other and none replaces the others.
File | Target audience | Primary purpose | Format |
|---|
llms.txt
| AI agents and LLMs | Curate high-signal content for semantic retrieval | Markdown |
robots.txt
| Search engine crawlers | Control which pages bots may request | Plain text directives |
sitemap.xml
| Search engine indexers | List all public URLs for indexing | XML |
As the llmstxt.org specification clarifies, robots.txt handles security and access, sitemap.xml aims for completeness across hundreds or thousands of URLs, and llms.txt focuses on curation. Many AI crawlers are regulated via robots.txt. The llms.txt file curates content for cases where those crawlers are already allowed in, or where users open your site directly inside a chatbot session. Our agentic browsing readiness guide covers how agents interact with robots.txt and sitemap.xml directives in practice.
Pricing and feature pages are the highest-priority entries in any B2B SaaS llms.txt file. For each entry, the destination page also needs to be structured for passage extraction: answer-first headings, plan names in plain text rather than buried in CSS-rendered tables, and concise descriptions of what each tier includes. For integrations, list tools by name. Agents match on entity names, not category labels. "Integrates with Salesforce, HubSpot, and Slack" is more retrievable than "supports leading CRM and communication platforms."
For developer agents browsing on behalf of a user, include a direct path to your API reference and authentication documentation. Developer-focused AI tools increasingly request and process Markdown-formatted content. Link to the canonical API reference, not a marketing overview of developer capabilities.
What content types to include in llms.txt
Not every page on your site belongs in this file. The context window an AI agent allocates to your llms.txt is finite, and low-value entries dilute the signal from high-value ones.
Prioritizing commercial buyer queries and filtering noise
Start with the pages that answer the queries buyers ask AI assistants during vendor evaluation: pricing, feature comparisons, security and compliance documentation, and integration lists. Our citation research suggests that pages with verifiable facts, sourced claims, and structured data tend to perform well in AI citation rates compared to general blog content.
Tag pages, category archives, author profiles, and date-based URL patterns often add noise without contributing retrievable answers. Each unnecessary entry uses context window budget that could go to a high-signal product page. If a URL does not directly answer a buyer's pricing, feature, security, or integration question, leave it out.
Structuring educational content for LLMs
Educational resources belong in llms.txt when they directly support a buyer's evaluation process. Getting started guides and use case libraries qualify. General thought leadership posts and author bio pages do not. The distinction is intent: include pages that help a buyer evaluate whether your product solves their problem. Pages that exist primarily for brand awareness or top-of-funnel traffic belong in your sitemap, not here.
Our real citation rate benchmarks post covers how agents weight educational versus commercial content differently across query types.
Designing llms.txt content for machine parsing
The format requirement is Markdown, not HTML. This matters because Markdown is more token-efficient than HTML, making the same information easier for a model to process without truncation.
The llmstxt.org standard recommends appending .md to your URLs to serve clean Markdown versions of pages directly to agents. For example, yourdomain.com/pricing.md returns a plain-text formatted version of your pricing page that an agent can parse without rendering JavaScript or stripping HTML tags. This approach is particularly valuable for pricing and feature pages where CSS-rendered tables are common.
If your platform does not natively support .md URL suffixes, tools like Firecrawl can crawl your site and return Markdown output, handling JavaScript rendering in the process. Some CMS platforms and SEO plugins support native llms.txt generation, handling updates automatically on content publish.
Each file list entry should include a short inline description rather than leaving the link title to do all the work. An agent encountering [Pricing](/pricing) has less context than one encountering [Pricing](/pricing): Plan tiers, per-seat costs, and enterprise terms. The description is a semantic signal that guides the agent toward the right resource. Tables and bulleted lists in your linked destination pages may also improve extraction accuracy compared to dense prose. Short, scannable paragraphs with clear H2 to H3 hierarchies can provide clean passage boundaries for retrieval systems.
Deploying and hosting your llms.txt file
Host the file at the root directory of your domain: yourdomain.com/llms.txt. The specification permits subdirectory paths, but root-hosted files are typically picked up more reliably. Use the same URL format (relative or absolute) consistently for all internal links within the file.
Configuring and verifying the file
Your server should return an HTTP 200 status code and serve the file with a text/plain MIME type. Most servers handle .txt files correctly by default, but if yours serves the file as HTML, add a MIME type rule in your .htaccess or server configuration. The file should also be accessible over HTTPS; an HTTP-only URL may break accessibility for AI tools that enforce secure connections.
Quick-start implementation
Approach | Best for | Maintenance level | Key tool |
|---|
Manual creation | Teams with stable content | Update on pricing or feature changes | Text editor, version control |
CMS plugin | WordPress sites | Can automate on content publish | SEO plugins with llms.txt support |
Automated crawler | Sites with dynamic content | Can generate and update regularly | Firecrawl |
Docs platform | Developer-first products | Often automatic | Mintlify (may generate llms.txt and llms-full.txt natively - verify with your platform version) |
After you host the file, visit yourdomain.com/llms.txt in a browser. You should see your Markdown content as plain text, not HTML. If the browser renders formatting, the MIME type is incorrect.
Verification checklist:
- File is accessible at
yourdomain.com/llms.txt with HTTP 200 - MIME type is
text/plain - File opens at root domain over HTTPS
- H1 is present and matches your brand name
- Optional summary information is present below the H1
- All internal links use consistent URL format
.md variants exist for your highest-priority product pages
Maintaining and troubleshooting your llms.txt file
A stale llms.txt file may reduce citation effectiveness. We recommend updating the file whenever you update pricing, launch a new feature, add a major integration, or change security llms.txt certifications. Consider refreshing a "Last updated" date regularly so AI tools see the file as actively maintained.
This matters most for pricing. AI agents field pricing queries at high frequency, and an outdated link to a page that no longer reflects current plan names is worse than no entry at all because it generates an inaccurate citation. Our AI visibility tracker maps where your brand appears across AI engines, helping you identify when the content being cited may no longer reflect your current product data. Our broader guide on AI visibility tools for SaaS covers how to interpret those signals.
For redirects, always link to the final canonical URL in your file, not the redirect origin. If you migrate your site or restructure your URL hierarchy, regenerate the llms.txt file before launch and test every link against the live site. A broken link in llms.txt is a dead end for an agent that was on track to cite you accurately.
Key considerations for your llms.txt strategy
When to deploy an llms.txt file
For technical leads, the fastest path is a manual file in version control, deployed via CI/CD on any change to pricing or feature documentation. For marketing leaders focused on content curation, the highest-leverage starting point is listing your top five buyer-intent pages (pricing, comparisons, security, integrations, getting started) and checking each one against the CITABLE framework before adding it to the file.
Linking llms.txt to citation success
The file itself does not guarantee citations. It increases the probability that an AI agent retrieves the right page and extracts an accurate passage. The citation rate lift depends on the quality of destination pages, the consistency of claims across independent sources (Reddit, industry publications, comparison content), and the technical structure of the content itself.
Our citation analysis shows that information consistency across sources is a stronger predictor of citation rate than any single on-site optimization. The llms.txt file addresses discoverability. Off-page consistency and on-page extractability address the rest.
If you want to map where your brand currently appears (or does not appear) across ChatGPT, Claude, Perplexity, and Gemini, our Search Visibility Diagnostic covers a full AI visibility audit and 10 optimized articles, designed as a one-off engagement to validate results before committing to an ongoing retainer. Book a call and we'll tell you honestly whether we're a fit.
FAQs
Does an llms.txt file guarantee my site gets cited by ChatGPT or Claude?
No. The file may improve discoverability by giving AI agents a curated index of your most relevant content, but citation rate also depends on the structural quality of your destination pages, information consistency across independent sources, and schema markup. Our research on 2 million citations suggests these on-page and off-page factors have significant influence on whether a passage is selected.
Will exposing an llms.txt file make it easier for competitors to scrape my content?
The llms.txt file curates what is already publicly accessible. While the structured, machine-readable nature of llms.txt can make it easier for competitors to identify and scrape your key pages, these pages are already public. If you are concerned about proprietary data or making it easier for aggressive scrapers to target specific content, exclude those pages from the file entirely. For most sites, the commercial benefit of AI agents accurately citing your pricing and feature pages outweighs the incremental scraping risk.
How often should I update my llms.txt file?
Update it any time you change pricing, launch a significant feature, add or remove a major integration, or restructure your documentation. Consider refreshing a "Last updated" note regularly so AI tools see the file as actively maintained. Stale files may be given less priority by some agents when compared to competitors with fresher versions.
Where exactly does the llms.txt file need to be hosted?
Host it at the root of your domain: yourdomain.com/llms.txt. Your server should return an HTTP 200 status code and serve the file with a text/plain MIME type over HTTPS. While the llmstxt.org specification technically permits subdirectory hosting, root hosting is recommended for maximum discoverability.
What is the difference between llms.txt and llms-full.txt?
llms.txt is a concise, buyer-facing index that links to your most commercially relevant pages. llms-full.txt is an extended version that may include complete technical documentation, code examples, and detailed API references for developer agents. Most B2B SaaS companies start with llms.txt alone and add llms-full.txt only when developer documentation would otherwise inflate the primary file beyond a practical working length.
Key terms glossary
llms.txt: A Markdown file hosted at your domain root that curates high-signal pages for AI agents. Rather than forcing an agent to parse your full site, the file tells it which pages to prioritize when answering a buyer query.
llms-full.txt: An extended version of llms.txt for developer agents. It holds complete API references, code examples, and technical documentation that would inflate the primary file if included there.
Retrieval-Augmented Generation (RAG): A method used by AI answer engines that retrieves relevant passages from external sources before generating a response, grounding answers in current, factual content rather than relying solely on training data.
Dense passage retrieval: A retrieval technique that encodes queries and candidate passages as numerical vectors and matches them by semantic similarity rather than keyword overlap. This is the mechanism that rewards answer-first, self-contained page sections.
Citation rate: The percentage of AI-generated answers to a defined set of queries that include a reference to a specific brand, page, or claim. It is the primary metric for measuring AI search visibility.
