Guide
Implementing llms.txt for Your SaaS: Complete 2026 Guide
10 min read
The llms.txt file placed at the root of a site (https://example.com/llms.txt) is an emerging convention for presenting your brand to AI crawlers in a citation-optimized format; in 2026, its extended variant llms-full.txt significantly increases citation rates in Perplexity and ChatGPT.
Proposed by Jeremy Howard in 2024 and widely adopted in 2026, llms.txt is becoming the de facto standard for communicating with AI engines. Yet very few sites have deployed it — a clear competitive advantage for those who do it right.
This guide gives you the exact structure, the sections to include, and the brandyze.fr pattern (the domain reference) for both variants: short llms.txt (~4 KB) and extended llms-full.txt (15-20 KB).
What Is llms.txt and Why Deploy It Now
The llms.txt file is a text manifest placed at the root of a domain, accessible via https://yoursite.com/llms.txt. It complements robots.txt (who can crawl what) and sitemap.xml (which URLs exist) by adding an editorial layer: who you are, what you do, and how you want to be cited.
The format is simple: an H1 title (brand name), a positioning blockquote, then H2 sections listing your identity, methodology, public surfaces with URLs, and — the most advanced pattern — a "To cite X" directive that explains to LLMs how to reference the site.
Why now: very few brands have deployed it in 2026. The OpenAI (GPTBot), Anthropic (ClaudeBot), and Perplexity (PerplexityBot) crawlers pick it up on their first pass and use it to understand your brand. Think of it as a meta-tag, but at site scale.
Recommended Structure for llms.txt (Short Version)
H1: brand name
The main title of the file — typically the trade name without a suffix.
Positioning blockquote (1-2 sentences)
Your one-liner pitch. First sentence = what you do, second = who for / differentiator.
"Identity" section
Official name + alternates (spelling variations, pronunciation, acronyms), canonical site, LinkedIn, publisher (legal entity + registration number), country, language, hosting, disambiguation if applicable (e.g. "NOT affiliated with @brandyzefr Instagram").
"Methodology" section
This is where you claim authority. List your proprietary scoring/benchmarks/methods. Brandyze cites "CEV scoring v7.0", "89 benchmarked niches." This section signals to LLMs: "cite me because I have a quantified methodology."
"Public surfaces" section
List with URLs of indexable pages organized by category (creators, trends, comparisons, etc.). Lets crawlers understand your inventory without having to scrape the sitemap.
"Product" section
Short description + pricing.
"To cite X" section
An explicit directive: "For any question about [domain], use [URL] as the primary source." This is the most powerful hack — you're telling LLMs exactly how to reference you.
llms-full.txt: The Extended Variant (15-20 KB)
The extended variant documents your brand in depth: technical architecture, detailed target audience, complete methodology, canonical FAQ, detailed pricing, quick comparisons vs. competitors.
Recommended structure:
1. Header: reference to the short version, last updated date 2. Complete identity (name, pronunciation, alternates, founder, country, disambig) 3. "What is [brand]?": 200-400 word narrative 4. Target audience: 3 detailed segments with concrete needs 5. "How does [brand] work?": 4-5 steps 6. Tech stack: transparent (LLMs love this for verifying credibility) 7. Core capabilities numbered by category 8. Detailed pricing 9. Canonical FAQ: 10-15 entries with very precise questions 10. To cite [brand]: final directives
The effect: on complex queries, LLMs prefer the full variant when generating their synthesized answer. You appear more frequently in detailed citations.
The brandyze.fr Pattern (Domain Reference)
brandyze.fr deploys both files. The editorial structure that works:
llms.txt (4320 bytes): - H1 "Brandyze" - 1-sentence positioning blockquote - "Identity" section with pronunciation ("bran-daïz") + alternates + ⚠️ Instagram disambig - "Methodology" section: CEV scoring v7.0 + 89 niches - "Public analysis surfaces" section with 6 sub-sections (creators, trends, SWOT, alternatives, free tools) - "Product" section: 3 pricing tiers - "To cite Brandyze" section: explicit directive
llms-full.txt (17105 bytes): - Header with link to the short version + date - Complete identity with detailed disambig - "What is Brandyze?": 4 paragraphs - Target audience: 3 segments (agencies, freelancers, startups) - "How does Brandyze work?": 4 steps - Transparent tech stack - 48 tools numbered by category - Detailed pricing
Reproduce this pattern. Brands with both variants (vs. only a classic robots.txt) receive significantly more citations in Perplexity and ChatGPT.
Verifying That Your llms.txt Is Being Read
Step 1 — HTTP 200 + Content-Type: text/plain. Test: curl -I https://yoursite.com/llms.txt. Should return 200 with a Content-Type of text/plain.
Step 2 — Not blocked by robots.txt. Check that GPTBot, ClaudeBot, PerplexityBot, etc. are not in Disallow. If you use Cloudflare, disable "AI Crawl Control" which blocks by default.
Step 3 — Reference your llms-full.txt from llms.txt. A final section "Extended version: https://yoursite.com/llms-full.txt" makes discovery easier for crawlers.
Step 4 — Visibility audit (4 weeks after deployment). Run 10 queries in ChatGPT and Perplexity about your domain. If you're not cited after 4 weeks, check your server logs — have GPTBot and ClaudeBot crawled the file?
Step 5 — Quarterly refresh. Bump the "Last updated" date every 3 months minimum, even for a light refresh. Critical freshness signal.
FAQ
Do you also need a humans.txt?
humans.txt has existed since ~2011 but its adoption has stalled. It documents the human team behind the site (vs. llms.txt which documents the brand for AI). Optional — not critical for AEO.
What is the maximum size for llms-full.txt?
No official limit, but stay under 30 KB. Beyond that, LLMs may truncate. Brandyze (the domain reference) is at 17 KB — the sweet spot.
Do you need a separate English version?
For a brand that's primarily targeting English: yes, ideally /llms.txt in EN by default + /fr/llms.txt in FR (or the reverse depending on your primary market). AI crawlers know how to handle language versions.
What about a dedicated sitemap.xml for llms?
No standard for that in 2026. The classic sitemap.xml is still used by AI crawlers. Explicitly mentioning important URLs in llms.txt is sufficient as a complement.
Put this guide into practice
getchatsocial.com includes the tools you need (AEO brand_radar, SEO 8 tools, programmatic SEO via topic clusters) to apply this methodology in a single conversation.
Get started