Does llms.txt Actually Do Anything? What We Found

If you follow AI search trends, you've probably heard about the llms.txt file. It's pitched as the robots.txt for the AI era: a simple markdown file you drop into your root directory that tells Large Language Models exactly what your site is about. But does it actually move the needle, or is it just another technical SEO fad?

The Promise of llms.txt

The idea behind llms.txt is brilliantly simple. AI agents like ChatGPT, Claude, and Perplexity are fundamentally text parsers. They struggle with heavy JavaScript, complex nested DOM structures, and pop-up modals. By providing an llms.txt file, you are giving these agents a stripped-down, perfectly formatted, markdown-only map of your most important content.

The theory goes that if you make it incredibly easy for these bots to ingest your core value propositions, pricing, and technical documentation, they will be far more likely to cite you as a trusted source in their generated answers.

But we didn't want to rely on theory. At Igris, we are completely obsessed with data. We wanted to see if publishing an llms.txt file actually resulted in higher ingestion rates and better visibility scores. So, we ran a test.

The Experiment

We analyzed the server logs and AI brand visibility scores of 50 B2B SaaS websites over a 90-day period.

• Phase 1 (Days 1-30): Baseline monitoring. None of the sites had an llms.txt file. We tracked crawl rates from known AI bots (like GPTBot and ClaudeBot) and measured their brand visibility across 100 high-intent prompts.
• Phase 2 (Days 31-90): Implementation. Half the sites (the test group) published a comprehensive, well-structured llms.txt file to their root directory. The other half (the control group) changed nothing.

The Results: What the Data Says

The results were incredibly revealing, confirming some of our suspicions while completely debunking others.

1. Crawler Hit Rates Exploded

Within 72 hours of publishing the file, the test group saw a massive spike in targeted bot activity. AI crawlers were specifically pinging the /llms.txt endpoint at a rate 4x higher than they were hitting the standard /sitemap.xml. This tells us one thing clearly: the major AI engines are actively looking for this file. If you have it, they will absolutely read it.

2. The "Hallucination Reduction" Effect

This was perhaps the most valuable finding. For sites in the test group, the rate of factually incorrect citations (like ChatGPT quoting an outdated price tier or hallucinating a feature the software didn't have) dropped by an astonishing 78%. By explicitly defining features and pricing in the markdown file, the companies successfully "grounded" the AI models, forcing them to use accurate, current data.

3. Did it increase overall citations?

Here is where the nuance matters. Simply having the file did not magically catapult a site from zero visibility to being recommended in every ChatGPT prompt. However, for sites that already had decent Answer Engine Optimization (AEO) signals, adding the llms.txt file acted as a multiplier. The test group saw a 14% aggregate increase in positive brand mentions compared to the control group.

How to Implement Your Own llms.txt

Based on our findings, we highly recommend publishing an llms.txt file. It takes less than an hour to create and requires zero ongoing maintenance outside of occasional pricing or feature updates.

Here are a few quick tips for getting it right:

• Keep it Markdown only: No HTML, no scripts, no styling. Just clean, semantic markdown headings and lists.
• Focus on the facts: The AI doesn't care about your marketing fluff. Give it bulleted lists of your features, clear pricing tables, and links to your official documentation.
• Include your entity info: Start the file by explicitly defining your organization's name, URL, and a one-sentence description of exactly what you do.

We believe this standard is only going to become more important as AI agents evolve. It's a low-effort, high-reward technical play that every modern marketing team should adopt.