agentweb
Sign in
How it works

We check every agent class that matters today.

Not a generic SEO score. agentweb tests against the specific agents your customers are already using, and tells you which can reach you, which can’t, and exactly why.

The agents we evaluate

Four classes. Different requirements. One report.

/ Search & Citation

  • ChatGPT
  • Gemini / Google AI
  • Perplexity
  • Claude (search)

The agents that read your site to cite or recommend you in answers. We check robots.txt allowlist for each crawler UA, sitemap availability, and what the agent actually sees on first paint.

/ Autonomous Browser

  • Mariner
  • Comet (Perplexity)
  • Operator (OpenAI)
  • Computer Use (Anthropic)

Real-browser-driving agents — they look like users, not bots. They struggle when your site is JS-heavy, has aggressive bot detection, or doesn't expose semantic structure. We diagnose the things that trip them up.

/ CLI & API

  • Claude Code
  • Cursor
  • Cline
  • Custom integrations

Agents that integrate with your services directly — they want MCP server cards, agent skills, OAuth metadata, an API catalog. If none of those exist, no agent can wire up to you without bespoke work.

/ Agentic Commerce

  • Stripe ACP
  • Coinbase x402
  • MPP
  • UCP
  • AP2

The protocols that let an agent complete a checkout end-to-end. Without one, an agent can find your products but has to hand off to a human at purchase — losing most of the agent buyer.

Pipeline

Probe → synthesize → run → chat.

1

Probe — six seconds, free

We hit ~22 open-standard endpoints (robots.txt, sitemap, MCP server card, A2A agent card, agent skills, content negotiation, x402, MPP, UCP, ACP, AP2, OAuth discovery, and more). No browser automation, no JS execution. Pure HTTP, parallel, fast.

2

Synthesize — per-agent verdicts

We map probe results onto each agent's actual requirements. ChatGPT cares about ChatGPT-User and GPTBot in robots.txt. Mariner cares about whether your DOM is legible without JS hydration. CLI agents care about MCP discovery files. The matrix shows pass / partial / fail with reasons.

3

Run — 25 brand-aware tasks

Behind a magic-link sign-in, a real AI agent visits your site and works through 25 brand-specific shopping tasks. Each one gets a transcript, products served, replay video, and a 1-5 self-evaluation score from a separate LLM judge.

4

Chat — ask the report

The report has tools. Ask why a task scored low, what evidence backs the standards score, how your behavioral score compares to peers. The chat agent calls the right tool and pulls answers from your scan data — not training data.

Privacy

Reports are private. There is no public leaderboard.

Every report is owned by the email that ran the scan. You can share a report by email — recipients get a private link, no sign-in required to view it, and the share can be revoked any time. We don’t publish names. We don’t rank brands. What you do with your report is your call.

Start with your URL.

Standards probe is free and instant. No email needed for the probe — only to claim, share, or run the behavioral scan.

Scan a URLWorks with →