RELAY
Audit my site
Agent-readiness · free audit

You optimized for Google.
AI never got the memo.

See whether AI shopping agents can find, read, and buy from your store - or whether your own edge is silently blocking them.

Free · no login · ~30 seconds · probes as 6 AI agents

Agent access across models
Which AI agents can reach & buy from your store, over time
sample
020406080100Mar 15Mar 30Apr 15Apr 30May 15May 30
  • Hover a point · click a row to toggle
We probe every site the same wayRENDER-GAP CHECKFALSE-DECLINE PROBESCHEMA VALIDATIONROBOTS.TXT AUDITCHECKOUT REACHABILITYWAF / EDGE FINGERPRINTAGENT-UA SPOOF TEST·We probe every site the same wayRENDER-GAP CHECKFALSE-DECLINE PROBESCHEMA VALIDATIONROBOTS.TXT AUDITCHECKOUT REACHABILITYWAF / EDGE FINGERPRINTAGENT-UA SPOOF TEST·
● relay.app/a/sample-run

Drop a URL.
Get five evidence-graded checks.

01
Retrieval bots not blocked
robots.txt allows OAI-SearchBot, PerplexityBot, ChatGPT-User
● PASS
02
Edge serves 200 to AI UAs
Fetched as GPTBot / ChatGPT-User from outside - compared to a browser
● FAIL
03
Render-gap on key facts
Raw HTML vs. headless-rendered diff on price, name, availability
● FAIL
04
Structured offer data
schema.org Product + Offer with price, currency, availability
● WARN
05
Reachable checkout form
Native <form action> with labeled, non-hidden fields
● PASS

Sample output shown for illustration. Every check is tagged by evidence quality - proven, emerging, or inferential - never presented as more certain than it is.

THE MONEY FINDING

Legacy fraud and bot-defense systems were tuned on human behavior. A legitimate shopping agent looks “too clean” and gets false-declined - blocked like a bad bot, with no chargeback to even flag the lost sale.

WHAT YOU GET

A ranked list of what’s actually blocking agents - starting with the checks that gate everything else - and, if you want it, an ongoing watch that re-probes after every deploy so a WAF rule change doesn’t silently re-break you.

FAQs

No. It's entirely read-only - we fetch your site as a browser and as several real AI agent user-agents, and report what we see. Nothing is installed, and no code is added unless you later opt into Agent-Readiness Watch.

Four categories: discoverability (robots.txt, sitemaps), understandability (render-gap, structured data), actionability (can an agent reach and complete checkout), and identification (does your edge false-decline legitimate agents). The last one is our sharpest, most differentiating signal.

As of 2026, no major AI crawler - GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot - executes JavaScript. If your prices, availability, or reviews only exist after the page renders client-side, those agents see a blank shell. It's the single strongest check in the audit because it gates everything downstream.

We fetch key pages using the real user-agent strings of retrieval and shopping agents (GPTBot, ChatGPT-User, PerplexityBot, Claude-User) and compare the response to a plain browser fetch. A 403, a bot challenge, or a silent redirect on an agent that robots.txt says is allowed counts as a false decline.

We score them as hygiene, never as required. Real-world consumption of llms.txt by AI vendors is estimated at roughly 0.1% of requests, and Google states you don't need special AI files to be cited. If you already have them, we validate them; we won't tell you to add them and call it done.

The paid layer above the free one-time audit: scheduled re-probes and drift alerts, so a WAF rule change or a redesign that moves pricing behind client-side render doesn't silently re-break agent access for months before anyone notices.

Free audit → Agent-Readiness Watch

Find out before an agent - or a customer - does.

Start with a free, one-time audit. If it finds something worth watching, move up to Agent-Readiness Watch for scheduled re-probes and drift alerts. If you want the fixes served for you, ask us about the pilot.

Free · no login · ~30 seconds.