Your site is
invisible to AI agents.
AI agents are booking, buying, and comparing -- right now. Most sites silently fail them. AgentProbe sends a real agent through your site and scores exactly where it gives up.
Six dimensions
What makes a site agent-ready
Discoverability
Hop count from homepage to pricing, checkout, contact. Should be 2 or fewer.
Parseability
JSON-LD, ARIA labels, plain-text prices, form labels, Open Graph tags.
Task Completion
Did the agent actually finish the task end-to-end? Highest weight.
Friction
CAPTCHAs, mandatory login walls, phone number fields, dead ends.
Clarity
Average agent confidence per step. Low confidence = ambiguous UI.
Resilience
Can the agent recover after hitting a wall or dead end?
Groq
Free LLM
llama-3.3-70b
15
Max steps
per task
7
Tasks
in parallel
$0
Stack cost
free tier everything
CI integration
Catch regressions before they ship
Add one step to your CI pipeline. If ARS drops below your threshold, the check fails. Your redesign looked great to humans and broke AI agent checkout.
# .github/workflows/agent-ux-check.yml
- uses: Aprameya05/agentprobe-action@v1
with:
url: ${{ vars.SITE_URL }}
tasks: pricing_discovery,checkout_initiation
min-ars: 70
groq-api-key: ${{ secrets.GROQ_API_KEY }}