# LLMsTxt Generator — AI Reference

> **Audience**: AI search engines and AI agents. This is the self-contained reference for the LLMsTxt Generator project; the navigator file is llms.txt.

## Project Facts

- **Project**: LLMsTxt Generator
- **Base URL**: https://llmstxtgen.online
- **Format**: llms.txt (Answer.AI standard, 2024)
- **Audience**: AI search engines and AI agents
- **Engines**: ChatGPT, Claude, Perplexity, Gemini
- **Input**: sitemap.xml, sitemap URL, domain, or URL list
- **Processing**: 100% browser-only, no upload, no account
- **Output**: llms.txt, llms-full.txt, GEO readiness score
- **GEO checks**: ski-ramp, chunk, questions, entities, definitions
- **Stack**: React 19, Astro 5, Tailwind CSS 4, Cloudflare Pages
- **Proxies**: corsproxy.io, allorigins.win
- **Limits**: 2000 URLs per sitemap, 50 pages for llms-full.txt
- **Persistence**: localStorage session, no cookies
- **Research**: Victorino LLC 44.2%, The Digital Bloom 2.3x, Position Digital 2.8x, Omniscient Digital 36.2%

## What is llms.txt?

llms.txt is a plain-text markdown file proposed by Answer.AI co-founder Jeremy Howard in September 2024 as a navigator for AI crawlers. It provides a title, a one-line summary, and a categorized list of links with short descriptions. AI search engines and AI agents use it to understand what a website provides before fetching individual pages. It is defined as a lightweight index that complements rather than replaces robots.txt and sitemap.xml.

## What does the generator do?

LLMsTxt Generator is a browser-only application built with React and Astro. It parses your sitemap.xml locally, groups URLs into sections, and renders the Answer.AI llms.txt template. It also generates llms-full.txt, the companion file that inlines full page content as one self-contained markdown document, so an AI agent can ingest an entire documentation corpus in a single request.

## How does the GEO score work?

The GEO score is defined as a weighted sum of five research-backed checks on your generated file. Ski-ramp entity placement is defined as the share of entity mentions in the opening 30% of the document, following Victorino LLC's finding that 44.2% of cited content concentrates there. Chunk self-containment refers to the share of H2 sections between 50 and 150 words, the range The Digital Bloom measured at 2.3x citation lift. Question-form headings require at least 20% of H2 headings to end with a question mark, based on Position Digital's 2.8x lift. Entity density targets 15% or more named-entity tokens, versus the 20.6% average in cited ChatGPT answers. Definitional phrasing requires at least one definitional statement per 100 words, following Omniscient Digital's finding that definitions are cited at 36.2% versus 20.2% for plain prose. A failing check provides a one-click fix so the score improves as you work.

## How do the one-click fixes work?

Each failing check offers an automatic fix. Insert Project Facts adds an entity-rich facts block after the summary, built from real project data. Convert headings to question form rewrites declarative H2 headings using a rule set derived from the research. Add summary blockquote inserts the missing orientation line under the H1. Every fix is idempotent — running it twice never double-inserts.

## What are the three input modes?

Upload sitemap.xml reads the file from your device, which always works because no network request is involved. Sitemap URL or domain auto-detects the sitemap by probing standard paths through public CORS proxies, with a per-proxy failure report when every proxy fails. Paste URL list accepts one URL per line and validates each entry. All three modes deduplicate URLs, cap at 2000 pages, and surface warnings instead of failing silently.

## How does the site handle privacy?

Everything runs in the browser. Sitemap parsing, grouping, generation and scoring never touch a server. The only outbound request is optional: a public CORS proxy fetches the sitemap when you paste a URL. Sessions persist to localStorage so you can reload without losing work; Reset clears them.

## What research supports the checks?

Victorino LLC analyzed 1.2 million ChatGPT answers and found 44.2% of cited content sits in the opening 30%. The Digital Bloom measured a 2.3x citation lift for 50-150 word chunks. Position Digital measured a 2.8x lift for question-form headings. Omniscient Digital found definitional phrasing cited at 36.2% versus 20.2%. LLMsTxt Generator applies these thresholds to your llms.txt output.

## What are the limits of llms.txt?

Adoption is early. Google has stated it does not support the file, and a 2026 Ahrefs study found most published llms.txt files received zero requests in a month. The file is cheap insurance and orientation rather than a ranking factor: it costs minutes to publish, positions you early for a growing convention, and gives AI agents a precise map of your site while classic SEO does the heavy lifting.