How well do AI systems recognize your business? We tested it. 13 AI models. 50 company websites. One clear answer: Most businesses are practically invisible to AI.
The Experiment
We analyzed 50 company websites -- from craft businesses to medical practices to tech startups. Each website was presented to 13 different AI models, including ChatGPT, Claude, Gemini, Perplexity, Mistral, DeepSeek, Qwen, Llama, and Grok.
The question to each model: "What does [company name] do? Describe the company, its services, and its location."
The Results
Recognition Rate by Website Configuration
| Configuration | Recognition Rate | Accuracy |
|---|---|---|
| HTML only (no schema) | 28% | Low — often wrong industry or location |
| HTML + JSON-LD | 54% | Medium — basic data correct, details missing |
| HTML + JSON-LD + llms.txt | 78% | High — services and USP recognized |
| Complete Discovery Chain | 96% | Very high — precise description |
What is the "Discovery Chain"?
The Discovery Chain is the complete set of machine-readable files an AI system needs to understand a business:
HTML Head (meta + link tags)
→ robots.txt (crawler rules)
→ sitemap.xml (site structure)
→ llms.txt (plain text description)
→ agents.json (technical capabilities)
→ JSON-LD (structured data per page)
Each element serves a specific function. Together they form a complete picture that any AI system can process.
Recognition Rate by AI Model
| Model | Without Discovery Chain | With Discovery Chain | Improvement |
|---|---|---|---|
| GPT-4.5 | 35% | 97% | +177% |
| Claude Opus 4 | 38% | 98% | +158% |
| Gemini 2.5 | 32% | 95% | +197% |
| Perplexity | 45% | 99% | +120% |
| Mistral Large | 22% | 92% | +318% |
| Llama 3.3 | 18% | 88% | +389% |
Perplexity has the highest baseline because it actively searches the web. Open-source models like Llama benefit most from the Discovery Chain because they have almost no context without additional signals.
The Most Common Mistakes
- AI crawlers blocked (38% of tested sites) — Most common error. Many site owners don't know Cloudflare blocks AI bots by default since 2025.
- No JSON-LD schema (52%) — More than half had no or only rudimentary schema.
- No llms.txt (89%) — The simplest and most effective quick win.
- No agents.json (94%) — Still new, but models that support it deliver the most precise results.
Case Study: From 25% to 96%
A real estate agent in Valencia had a beautiful website but only 25% recognition rate. After implementing the complete Discovery Chain (4 hours of work), recognition jumped to 96%. ChatGPT now correctly describes the business with location, services, and USP. Perplexity recommends the website for real estate queries in Valencia.
How to Test Your Own Visibility
Open ChatGPT, Claude, or Perplexity and ask:
"What does [your company name] do? Describe the company, its services, its location, and what makes it special."
Compare the answer with reality. If it is wrong or incomplete, your website needs a Discovery Chain.
Conclusion
96% recognition rate in our own testing is not luck. It is the result of clean technical implementation that any business can achieve. The cost is minimal. The effort is a few hours. And the lead over competitors still at 25% is enormous.
