Skip to content

benchmark.md

Latest commit

 

History

History
53 lines (36 loc) · 2.83 KB

benchmark.md

File metadata and controls

53 lines (36 loc) · 2.83 KB
Raw

100-Store Agentic Readiness Benchmark

This benchmark is the promotion-friendly story for Agent Lighthouse: most stores are readable by browsers and search crawlers, but not yet ready for autonomous AI agents.

Headline Findings

Finding Result
Stores with OpenAPI, WebMCP, or MCP action surfaces 0%
Stores with llms.txt or llms-full.txt 4%
Average agent-readiness score 53.6 / 100
Fastest practical score lift llms.txt, AI catalog, accessible icon buttons

These figures come from the pre-v2 run. v2 changed the registry, the pass conditions, and the scoring formula, so the average score is not comparable to a v2 scan; re-run the script to refresh it.

What The Benchmark Checks

Agent Lighthouse evaluates 215 audits across 8 agent-journey categories:

  • Access & Crawl Control: robots.txt rules for GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Applebot-Extended, and related crawlers.
  • Content Extraction: clean main content, semantic structure, render cost, response time.
  • Machine Discovery: llms.txt, llms-full.txt, sitemaps, RSS, .well-known surfaces, AI-file delivery.
  • Structured Data: Schema.org Product, Offer, Review, Organization, Article, FAQPage, JSON-LD validity.
  • Answer Readiness: direct answerability, step lists, comparison tables, unique data, citations, authorship.
  • Agent Interfaces: OpenAPI, WebMCP, MCP discovery, agents.json, AI catalogs, search endpoints.
  • Agentic Commerce: product offers, availability, and transaction certainty.
  • Agent Operability & Safety: HTTPS, security headers, accessible controls, forms, broken agent endpoints.

Narrative For Promotion

Traditional ecommerce sites are optimized for human browsing and Google search. Agent Lighthouse shows the missing layer for AI-driven discovery and task completion. A site can have strong Product schema and still fail because agents cannot find llms.txt, cannot access crawler-safe content, or cannot discover any machine-readable action surface.

Reproduce

pnpm tsx scripts/benchmark-stores.ts

The script writes incremental output to:

reports/investigation/benchmark-stores-data.json

Post Copy

We benchmarked 100 ecommerce storefronts for AI-agent readiness.

The pattern: most sites have classic SEO basics, but almost none expose the files and action surfaces AI agents need: llms.txt, OpenAPI, WebMCP, MCP discovery, and clean machine-readable interaction paths.

Tool: https://github.com/ForkPoint/agent-lighthouse