citation guide · AEO
how to get your AI agent or MCP server cited by ChatGPT, Claude, and Perplexity.
every engine has a different front door. here’s where each one looks.
what “being cited” actually means
When someone asks ChatGPT “what AI agent or MCP server helps with X,” the response is a synthesized answer, not a list of search results. Your agent or server either appears in that answer or it doesn’t. There’s no page two.
Getting cited means the engine knows you exist (discovery), knows what you do (entity clarity), and trusts you enough to surface you (citation signals). Each AI engine has a different path to that trust, and MCP servers have one extra advantage worth calling out: Anthropic queries Smithery’s catalog directly, so an MCP server listed in Smithery is one integration away from Claude. What works for Perplexity isn’t identical to what works for ChatGPT, but the foundation overlaps enough that doing it once covers most of it.
engine by engine
ChatGPT (OpenAI)
crawlers: GPTBot, OAI-SearchBot, ChatGPT-User
- ,GPTBot crawls the web; OAI-SearchBot is used when web browsing is enabled at query time.
- ,Smithery (MCP-server catalog) and the OpenAI plugin / GPT store catalogs are heavily indexed, listing there is the highest-leverage single action.
- ,Agent.directory, GitHub, and other AI-agent + MCP-server indexes are in scope.
- ,Allow GPTBot in robots.txt. The wildcard * covers it, but being explicit signals intent.
Claude (Anthropic)
crawlers: ClaudeBot, anthropic-ai
- ,Claude's training corpus is the primary source for recommendations made without tool use.
- ,Smithery has a direct catalog relationship with Anthropic. MCP servers listed there have the shortest path to appearing in Claude responses, this is the single highest-leverage move for an MCP server.
- ,For AI agents (non-MCP): llms.txt + schema.org JSON-LD + at least one external citation is the baseline.
- ,Allow ClaudeBot and anthropic-ai in robots.txt.
Perplexity
crawlers: PerplexityBot
- ,Perplexity is live-web, it crawls sources at query time and cites them directly in responses.
- ,Being cited by Perplexity requires a page that directly answers the query. Your agent or MCP server's homepage, a directory listing, or a how-it-works post all qualify.
- ,Clear H1/H2 structure, explicit question-answer formatting, and fresh content help Perplexity rank your page over a competitor's.
- ,PerplexityBot must be allowed in robots.txt.
Gemini (Google)
crawlers: Google-Extended, Googlebot
- ,Gemini draws from Google's index, so traditional SEO signals (backlinks, domain authority, structured data) apply in addition to AEO-specific ones.
- ,Google-Extended is the crawler for AI training; allow it alongside Googlebot.
- ,schema.org JSON-LD matters more here than for the other engines.
the universal foundation
Regardless of which engine you care most about, these five things are required before any of the engine-specific work matters:
- 1.llms.txt at your root domainOne page, plain text. What your agent or MCP server does, who it's for, where to access it. The engines that support it check this first.
- 2.schema.org JSON-LD in your <head>SoftwareApplication or Organization type. Name, URL, description, and at least one knowsAbout or applicationCategory field.
- 3.robots.txt explicitly allowing AI crawlersGPTBot, ClaudeBot, PerplexityBot, Google-Extended, anthropic-ai, cohere-ai. Don't block what you're trying to get discovered by.
- 4.At least one directory listingSmithery for MCP servers (queried directly by Claude). Agent.directory for AI agents. GitHub for open-source. Pick the most relevant one for your category and get listed there first.
- 5.One external citationThe engine needs to see you from somewhere other than your own domain. A directory listing counts. A blog post counts. A GitHub README counts.
why timing matters
AI engine answer corpora are forming right now. The agents and MCP servers that appear in responses to “what AI agent does X” or “which MCP server for Y” today are almost all those that got into the indexes early, before the space got crowded. Citation presence compounds: once you’re cited somewhere, other sources cite you because the engine already surfaced you, which means the engine surfaces you more.
The window to claim early positioning is open. It won’t stay open forever.
· what envvoy handles
envvoy generates your llms.txt, schema.org JSON-LD, agent-card.json, and sitemap automatically. it configures explicit AI crawler permissions in robots.txt. it submits to Smithery (for MCP servers), Agent.directory, GitHub, Virtuals, and the others that matter for your category. and it monitors your citation presence weekly, alerting you when your position slips before competitors notice.