When a potential customer asks ChatGPT "who's the best electrician in Dallas," ChatGPT does not search Google. It does not pull up Yelp. It draws on a combination of its training data, real-time web access (where available), and structured signals your website either does or does not send.
Most businesses have no idea this is happening. They've optimized for Google and assumed AI would follow. It doesn't. The ranking signals are different. The data sources are different. And the businesses winning AI recommendations are often not the ones winning search.
The Three Layers AI Uses to Make Recommendations
AI recommendation engines pull from three sources: training data (what the model learned before its cutoff), real-time retrieval (live web access, where the AI browses current pages), and structured signals (machine-readable data like llms.txt and JSON-LD that explicitly tell AI systems who you are and what you do).
Training data is hard to influence retroactively. Real-time retrieval depends on whether an AI system has web access enabled (not all do). Structured signals are what you can actually control right now.
What Makes a Business Recommendable to AI
AI agents evaluate several factors when deciding whether to name a business in a response. None of them are your Google ranking.
- —llms.txt — a plain-text file at yourdomain.com/llms.txt that tells AI systems what your business does, who you serve, and what questions you answer. Think of it as a resume for AI crawlers.
- —JSON-LD structured data — machine-readable markup that identifies your business entity, services, location, and authority signals. AI systems use this to verify that you are who you say you are.
- —Robots.txt configuration — whether you've explicitly allowed AI crawlers like GPTBot, ClaudeBot, and PerplexityBot to access your site. Many businesses accidentally block them.
- —Content authority — whether your site directly answers questions that your target customers are asking AI systems. Topic specificity matters more than domain authority.
- —Citation consistency — whether your business name, address, and description are consistent across your site, your Google Business Profile, and other authoritative sources AI systems reference.
Why Google SEO Doesn't Transfer
Here is the counterintuitive part: businesses with strong Google rankings often assume they are also visible to AI. They are frequently wrong.
Google's algorithm weights backlinks, domain authority, and click-through rates. AI systems weight structured data, explicit entity definitions, and content that directly answers conversational queries. A business that has invested heavily in link-building may have almost no structured AI signals. A newer competitor with good llms.txt and JSON-LD markup can outrank them in AI recommendations despite having a fraction of the domain authority.
How Perplexity, ChatGPT, and Claude Differ
Not all AI systems work the same way. Perplexity is primarily a real-time retrieval engine — it browses the web when answering queries, which means your current page content matters as much as structured signals. ChatGPT's behavior depends on whether the user has web access enabled; without it, the model relies entirely on training data and any files or context the user provides. Claude, by default, does not browse the web and relies on training data plus any MCP tools connected to the conversation.
The practical implication: optimizing for one AI platform is not enough. Each system has different access patterns, different training data cutoffs, and different weighting of signals. The businesses that appear consistently across all of them have invested in foundational structured signals that translate across platforms.
The MCP Factor
Model Context Protocol (MCP) is an emerging standard that allows AI systems to directly query your business data in real time. A business with an MCP server can expose its services, pricing, availability, and FAQs to any AI agent that supports the protocol. This is the next frontier of AI visibility — and it's still early enough that having it creates a significant competitive advantage.
Frequently Asked Questions
How do I know if AI systems are recommending my business?
Check your server access logs for AI crawler user agents: GPTBot, ClaudeBot, PerplexityBot, and Google-Extended. Also look for anomalous direct traffic in your analytics — AI-referred sessions often show no referrer header and get classified as direct. The Zygur AI visibility score shows you which AI crawlers can and cannot access your site.
Does having a Google Business Profile help with AI recommendations?
Partially. Google Business Profile data is used by Google's own AI systems (AI Overviews, Gemini). Other AI platforms like ChatGPT and Perplexity may reference it as an external authority signal, but do not rely on it primarily. It is a supporting signal, not a primary one.
How often do AI systems update their recommendations?
Models with training data cutoffs update when they are retrained, which can be months or years apart. Real-time retrieval systems like Perplexity update immediately as they crawl new content. Structured signals like JSON-LD are picked up whenever an AI crawler next visits your site.
Can I appear in AI recommendations without a lot of content?
Yes. Structured signals matter more than content volume. A single well-configured page with clear JSON-LD entity markup, a populated llms.txt, and an accessible robots.txt can outperform a site with hundreds of blog posts but no structured AI signals.
What is an AI visibility score?
An AI visibility score measures how well your site communicates with AI systems across the signals they use to evaluate and recommend businesses: crawler access, llms.txt, structured data, meta tags, and MCP server presence. Zygur's free tool scores any URL from 0 to 100 in 60 seconds and shows exactly which signals are missing.
Check your AI visibility score at zygur.com — free, instant, no account required. See exactly which signals AI systems are finding (and which ones are missing) for your business.