How AI reads Vercel

vercel.com Jun 22, 2026 6 min read Emerging

Short

50/ 100

AEO Level 3Emerging

The short answer

Vercel scores 50 out of 100 on AEO readiness. Strong structured data (80) and solid agent-discovery signals (75) are undercut by a complete absence of agent interfaces (0) — meaning AI assistants can read Vercel's content but cannot interact with its platform programmatically.

What AI sees

An AI crawler landing on Vercel's homepage today finds a well-structured page but almost no signals that a machine is welcome to do anything beyond read.

The homepage carries genuine schema markup — product and organization data that search and AI engines can parse with confidence, reflected in the 80 structured-data score. Vercel's technical documentation and feature descriptions are well-organized and indexable. What the agent does not find: a Markdown alternative when it sends Accept: text/markdown, RFC 8288 Link headers pointing to API or discovery resources, or any well-known endpoints advertising platform capabilities. For a company whose customers deploy AI-native applications daily, the gap between Vercel's content readability and its agent interactivity is striking.

Vercel homepage captured by the AEO scanner — rich visual layout with strong schema markup, but no Link response headers or Markdown endpoint for agent requests

Vercel robots.txt at time of scan — standard crawl rules present, but no user-agent declarations for GPTBot, ClaudeBot, PerplexityBot, or other AI crawlers

Where it loses points

Agent interfaces score zero — not a single MCP server card, API catalog, WebMCP tool definition, or OAuth resource metadata endpoint exists at any standard well-known path.

How to fix it

Three targeted additions would close the most impactful gaps and move Vercel from a passively readable site to an actively agent-capable platform.

Add AI crawler rules to robots.txt

Goal

Declare explicit per-agent rules so GPTBot, ClaudeBot, and PerplexityBot know which paths Vercel authorizes for training and citation.

Issue

The scan found no user-agent rules for any major AI crawler in robots.txt — those bots currently operate without explicit guidance.

Fix

Add named user-agent blocks for GPTBot, ClaudeBot, PerplexityBot, and similar crawlers. Use Allow directives for documentation, blog, and changelog paths you want cited; Disallow private dashboard and account paths.

Publish an MCP Server Card

Goal

Expose Vercel's platform capabilities to AI agents via a standard MCP server card at the well-known discovery path.

Issue

No file exists at /.well-known/mcp/server-card.json, so agent orchestrators have no structured way to discover what Vercel's platform can do.

Fix

Serve /.well-known/mcp/server-card.json with a serverInfo block, your transport endpoint, and a capabilities list. For a deployment platform, primitives like trigger-deployment, get-build-logs, and list-projects would give agents direct utility.

Publish an API Catalog (RFC 9727)

Goal

Let agents auto-discover Vercel's REST and edge-config APIs through a machine-readable linkset at the standard catalog endpoint.

Issue

No API catalog exists at /.well-known/api-catalog, leaving agents unable to enumerate available APIs without reading human documentation.

Fix

Serve /.well-known/api-catalog as application/linkset+json with entries for each major API surface — deployments, projects, edge config, DNS records — each with an href, rel, and type field per RFC 9727.

Common questions

Why does Vercel score only 50/100 despite being a developer-focused platform?

Vercel's 50/100 reflects a split profile: structured data (80) and agent discovery (75) are strong, but agent interfaces score zero. A platform built for developers shipping AI-native products has not yet exposed its own capabilities through machine-readable protocols like MCP server cards or API catalogs.

What does an agent_interfaces score of 0 mean for Vercel users?

It means AI assistants cannot programmatically interact with Vercel's platform through standard protocols. No MCP server card, no WebMCP tool definitions, no OAuth resource metadata — agents must fall back to human documentation rather than structured capability declarations.

How quickly could Vercel improve its AEO score?

Substantially, and fast. Adding AI crawler rules to robots.txt takes minutes. Publishing an MCP server card and an RFC 9727 API catalog requires only static JSON files at well-known paths. Together those three fixes could lift the score well above 70 without touching core infrastructure.