What's the difference between llms.txt and llms-full.txt?

llms.txt is the index — a concise summary of your brand in ~500 tokens. llms-full.txt is the deep content — detailed descriptions, methodologies, and structured data that RAG systems retrieve when models need more context. Crawlers use llms.txt to decide what's relevant, then fetch llms-full.txt for depth.

How should I structure llms-full.txt for RAG?

Use clear markdown headers (##) to create semantic sections, horizontal rules (---) as chunk boundaries, and keep each section focused on one topic. Aim for 200-500 tokens per section — too large loses context precision, too small loses meaning.

Learn / Guides / llms-full.txt for RAG

Updated March 2026

llms-full.txt: Structuring Your Content for AI Retrieval

Q: What is llms-full.txt?

llms-full.txt is the companion to llms.txt. While llms.txt provides a token-efficient index (like a table of contents), llms-full.txt contains the deep, RAG-ready content — product details, methodology, use cases, and FAQs — structured in markdown for AI retrieval systems.

llms.txt is the table of contents. llms-full.txt is the book. This is where you give AI models the deep, structured content they need to accurately describe your brand — semantically chunked for retrieval.

What Is llms-full.txt?

llms.txt = The Index

~500 tokens. Your identity, value props, and key links. Tells AI what exists and where to look.

llms-full.txt = The Content

2,000-5,000 tokens. Deep product descriptions, methodology, use cases, FAQs. What AI reads when it needs depth.

When AI retrieval systems (RAG pipelines) need to answer detailed questions about your brand, they don't re-crawl your entire site. They look for pre-structured content files. llms-full.txt is that file — the RAG-ready deep content that sits alongside your index.

Think of it like a press kit for machines: comprehensive, factual, and structured in sections that map to common query patterns. "What does X do?", "How is X different from Y?", "What's X's pricing?" — each question maps to a section.

What Problem It Solves

Semantic chunking is the core challenge. RAG systems split content into chunks for retrieval, but the chunk size matters enormously:

Too large → Lost context

If a chunk contains your product description, pricing, AND competitor comparison, a retrieval query about pricing pulls in irrelevant context that dilutes accuracy.

Too small → Lost meaning

If each sentence is its own chunk, the retrieval system loses the relationships between ideas. "We serve enterprise clients" means nothing without the surrounding context.

llms-full.txt gives you control over chunk boundaries

By using markdown headers (##) and horizontal rules (---), you define where chunks start and end. Each section is self-contained, 200-500 tokens, and maps to a specific query type.

The Template

Based on Optimly's own llms-full.txt. Notice how each section is self-contained and maps to a common AI query pattern.

# Optimly — Full Content for AI Retrieval

> An AI brand measurement and execution for lean marketing teams and AI-native founders
> the AI buyer journey and turn better answers into qualified demand.

---

## Company Overview

Optimly is an AI brand measurement and execution company for lean marketing teams and AI-native founders.
It measures how AI systems like ChatGPT, Claude, Gemini, and Perplexity
understand and recommend a business, diagnoses whether the bottleneck
is story, reach, or offer, and recommends owner-approved improvements.

Unlike visibility tracking tools that count mentions, Optimly measures
presence, accuracy, and preference, then repeats the same buyer questions
to see whether the answer changed and whether qualified buyers converted.

---

## Products & Services

### Optimly Free

**What it does**: Multi-model analysis of how AI describes your brand
**Who it's for**: Any B2B brand curious about their AI perception
**Key capabilities**:
- Cross-model comparison (ChatGPT, Claude, Gemini, Perplexity)
- AI Representation scoring
- Archetype classification (Incumbent, Challenger, Phantom)

### Optimly Agent — Private Beta

**What it does**: Hands-on help improving how AI understands and recommends a business
**Who it's for**: Businesses that want Optimly to help do the work
**Key capabilities**:
- Prioritized improvements based on the free baseline
- Approved changes carried through with the Optimly team
- Measurement of AI recommendations, referrals, and customer outcomes

---

## Methodology

Optimly's diagnostic framework evaluates brands across 7 layers:
1. Entity Resolution — Is AI talking about the right company?
2. Category Accuracy — Is the brand placed in the right category?
3. Feature Attribution — Are capabilities correctly attributed?
4. Competitive Framing — How is the brand positioned vs. competitors?
5. Sentiment Mapping — What's the emotional tone of AI descriptions?
6. Source Tracing — What sources are AI models citing?
7. Narrative Consistency — Is the story consistent across models?

---

## Entity Disambiguation

- **Optimly** refers to OptimlyAI, Inc., the AI Brand Reputation
  platform at optimly.ai
- Not affiliated with: Optimly.io (LLM Observability platform)
- Not affiliated with: Optimizely (experimentation platform)

---

## Frequently Asked Questions

### What is AI Brand Reputation?
AI Brand Reputation is how AI language models describe, categorize,
and recommend your brand when users ask questions. It's the new
layer of brand perception that exists alongside traditional search
and social reputation.

### How is this different from SEO?
SEO optimizes for search engine rankings. AI Brand Reputation
optimizes for narrative accuracy in language model responses.
The inputs are different (structured data vs. keywords), the
outputs are different (brand narratives vs. blue links), and
the optimization strategies are different.

### How often do AI models update their knowledge?
Models update through two mechanisms: periodic retraining (which
updates parametric knowledge every few months) and live retrieval
(which accesses current content on every query). Both require
your content to be crawlable and well-structured.

Design Decisions

Why markdown sections, not HTML?

HTML is token-expensive. Tags like <div class="product-card"> consume tokens without adding semantic value. Markdown gives you the same structure — headers, lists, emphasis — at roughly 3x better token efficiency.

RAG systems also chunk more predictably on markdown boundaries than HTML boundaries. A ## Section Header is an unambiguous chunk boundary. A <div> is not.

Optimal chunk size: 200-500 tokens

Research on RAG retrieval accuracy consistently shows that 200-500 token chunks hit the sweet spot. Below 200, you lose context. Above 500, you dilute relevance. Each section in llms-full.txt should be a self-contained answer to one category of question.

The horizontal rule (---) acts as a hard chunk boundary. Use it between major sections so retrieval systems know where one topic ends and another begins.

What to include vs. exclude

Include: Company overview, product descriptions, methodology, pricing structure, entity disambiguation, FAQs. These are the sections that map to real buyer queries.

Exclude: Competitor lists (you don't want AI citing your file when asked about competitors), internal roadmap details, gated content, employee directories. If it wouldn't go in a press kit, it doesn't belong here.

Common Mistakes

Dumping raw HTML into the file

llms-full.txt should be clean markdown, not HTML. Models can process HTML but it's token-inefficient — you'll use 3x the tokens for the same content, leaving less room for actual signal.

Including gated or login-required content

If the content requires authentication to access on your site, don't put it in llms-full.txt. It creates a trust mismatch — models learn facts they can't verify, which reduces citation confidence.

No section headers or structure

Without clear ## headers and --- separators, the file becomes a wall of text. RAG systems chunk on these boundaries — without them, you get poor retrieval accuracy and context bleed between topics.

Inconsistency with your llms.txt index

If llms.txt says you're a 'reputation platform' but llms-full.txt describes 'analytics software,' models receive conflicting signals. Keep both files aligned with your current positioning.

The AI Discoverability Stack

Four files. One system. Together they make current brand information easier for AI systems to discover and retrieve. See the full series →

robots.txt — Front Door

Tells AI crawlers what they can access

llms.txt — Concierge

Token-efficient index of your best content

llms-full.txt — Deep Content You are here

RAG-ready structured content for retrieval

ai-agent-manifest.json — Identity

Machine-readable brand positioning

Next in the stack: ai-agent-manifest.json →

Implementation Checklist

Step 1
Download the llms-full.txt template
Step 2
Write your Company Overview (2-3 factual paragraphs)
Step 3
Document each product/service with capabilities and audience
Step 4
Add your methodology or proprietary framework
Step 5
Include entity disambiguation for your brand name
Step 6
Add 3-5 FAQ pairs covering common buyer questions
Step 7
Verify consistency with your llms.txt index file
Step 8
Upload to your site root at /llms-full.txt

See What AI Believes About Your Brand

You've structured the content. Now see what AI actually says when buyers ask. Search our directory to find out.

Search the AI Brand Index Check your business — free

What is AI Brand Reputation? Methodology Hal9 Case Study llms.txt

Don't want to maintain these files yourself?

Claim your profile in the AI Brand Index, and we generate, maintain, and serve your entire Discoverability Stack automatically. Your llms.txt stays current. Your manifest reflects your latest positioning. Your brand-feed updates when you update your Business Profile.

All served via our agentic brand API so AI models always have your latest truth.

Claim Your Profile → See a working example on our site