How to make AI search notice your website in 2026
ChatGPT, Claude, and Perplexity cite the web differently than Google. Here's a field guide to AEO, GEO, and agent readiness.
Google ranked links. AI search cites sentences. That one shift changes what "ranking well" means, and most of the sites we scan are still optimized for the old game.
The four questions an LLM asks about your site
When ChatGPT, Claude, Perplexity, or Gemini needs to answer a question that involves your product, it runs through a mental checklist in milliseconds:
- Can I find it? Is the homepage reachable, does robots.txt allow me, does a sitemap or llms.txt exist?
- Can I read it? Is the content in the initial HTML, or locked behind JavaScript I can't wait for?
- Can I trust it? Is there structured data, an about page, author info, verifiable facts?
- Can I cite it? Are there quotable, self-contained sentences with enough context to stand alone?
Miss any one of those and the model picks somebody else.
Why blocking the wrong AI crawler is a quiet disaster
A lot of teams added one broad AI block in 2023 thinking they were "protecting" their content. In 2026, that setting is why their competitor keeps getting cited instead of them.
Training-time crawlers like GPTBot and Google-Extended are separate from search and retrieval user agents like OAI-SearchBot, ChatGPT-User, PerplexityBot, and ClaudeBot. OpenAI documents OAI-SearchBot as the crawler that can surface pages in ChatGPT search results. ChatGPT-User is user-initiated browsing. Block everything together and you remove useful AI visibility along with training access.
If you only make one policy change, separate training access from search and retrieval access. That's the traffic-sensitive distinction.
The llms.txt signal
llms.txt is to AI search what sitemap.xml is to Google: a structured hint that saves the crawler a dozen HTTP round-trips. It's optional, but publishing one (and its markdown-accessible pages) moves your site from "parseable with effort" to "trivially ingestible."
Our free scan checks whether you have one, whether it's well-formed, and whether your key pages expose Accept: text/markdown negotiation. The difference in extractability score is usually 15-20 points.
Agent readiness is the next frontier
MCP server cards, agent-skills manifests, WebMCP, and commerce protocols (x402, MPP, UCP, ACP) are how your site talks to autonomous agents, not just LLMs doing retrieval. It's early, but the sites publishing these endpoints today will be the ones agents transact with in 2027.
Run a free scan and see where your site stands across all eight categories.