01 · Search & citation
How they find and quote you
Robots.txt, sitemaps, content negotiation, and the user-agent rules that decide whether an AI search result quotes you — or your competitor.
- ChatGPT browse
- Gemini
- Perplexity
- Claude search
sys · diagnostic engine v2 · onlineagentweb.co / by aigency
Free diagnostic · by AigencyIs your site compatible
Is your site compatible
with the agent web?
ChatGPT, Gemini, Perplexity, Claude — and the autonomous browser agents on top of them — already shop your site. We tell you which ones can reach you, which ones can’t, and exactly why.
try
Tested against
ChatGPTGeminiClaudePerplexityOperatorMarinerAtlasCometACPMCPChatGPTGeminiClaudePerplexityOperatorMarinerAtlasCometACPMCP
scan / agent-compatibilitylive
target
https://glossier.com_
compat
2/7
agent classes ↦
partial· 3 fail · 2 partial · 2 pass
- pass
ChatGPT browse
robots ✓ · sitemap ✓
- fail
Gemini
Googlebot user-agent disallowed
- pass
Perplexity
PerplexityBot allowed
- part
Operator / Atlas
PDP JS-heavy, slow to settle
- part
Mariner
Cart needs sign-in
- fail
Claude Code · Cursor
No MCP server card
- fail
ACP checkout
agent-commerce-protocol missing
22 probes · 5 categories▸ open report
/ what we test
Four classes of agent. One report.
The web isn’t about “AI” in the abstract — it’s about whether this agent can find you, read you, and transact with you. We score each class against the open protocols it actually uses.
02 · Autonomous browser
How they actually shop you
Real browser sessions running 25 brand-aware shopping tasks. PDP load, filter, compare, cart, checkout. Replay video per task with a 1–5 self-eval.
- Operator
- Mariner
- Atlas
- Comet
- browser-use
03 · CLI & API
Whether your site is even discoverable
MCP server card, agent skills, A2A discovery. If you don't expose these, CLI agents have no way to know your API exists — they just don't visit.
- Claude Code
- Cursor
- Codex CLI
- Manus
- Devin
04 · Agentic commerce
Whether an agent can checkout for a user
agent-commerce-protocol endpoints, structured product data, identity hand-off. The difference between an agent recommending you and an agent buying from you.
- ACP
- OpenAI Instant Checkout
- Stripe Agent Pay
/ the gap
You’re running a storefront for an audience you can’t observe.
01the gap
Agents already shop your site. You can't see what they see.
ChatGPT, Claude, and Perplexity are sending agents to your storefront every day. They click, read, recommend — or give up. You can't watch over their shoulder, and your analytics treat them like noise.
02the gap
Your standards score is invisible until it's a sales call.
Robots.txt, sitemaps, MCP cards, agent-commerce protocols. None of it surfaces in any tool you already pay for. By the time someone asks why agent traffic dropped, it's a quarter late.
03the gap
You can't fix what you can't watch happen.
The fastest way to know is to watch a real agent attempt 25 real shopping tasks on your site, with the recording, the transcript, and the score next to it. Until now, you haven't been able to.
/ how it works
URL to interactive report in under fifteen minutes.
Two layers, one private report. Standards probes against open web specs, then a real agent runs 25 brand-aware shopping tasks. Every result has a chat next to it.
- 1t+0:00
Paste a URL
Standards probe runs inline in 6 seconds. L0–L5 score, per-check evidence, copy-paste fix prompts. Free, no email required.
- 2t+0:30
Sign in to claim
One magic link. The scan becomes a private permalink only you can see. From here, you can run the behavioral scan.
- 3t+12:00
Read the agent's report
Behavioral score, per-task drilldown with replay video, self-eval reasoning. Chat with the report to dig in.
/ one last look
See how an agent experiences your site — before someone else does.
Standards score is free. Behavioral scan is gated behind a magic-link sign-in. Reports are private to your email — there’s no public leaderboard.
a product of aigency · merchant control plane for AI agents