Agent Experience Testing for E-Commerce

Send the bird in first.

Coalbird sends AI shopping agents through your store every week and alerts you the moment they can no longer find, choose, or buy — before it shows up as silently lost sales.

Enter your URL, answer a few scoping questions, then choose a trial time right here.

Live field report Mission #041
Agent trace Buy socks. Prove size M.
  1. 01opened product page
  2. 02selected variant M
  3. 03cart state stayed quiet
SML
Cart echo missing
Blocked at variant proof

Made it to the size selector. Then things got weird.

3x repeat clicks before abandon
Field report · your live site · 4 minutes ago
  1. 01 open /mens/running-shoes
  2. 02 dismiss region interstitial
  3. 03 dismiss newsletter modal
  4. 04 filter colour = black
  5. 05 read 24 results
  6. 06 open /p/velos-2-road
  7. 07 click size EU 44
  8. 08 check selection → cannot tell
  9. 09 click size EU 44 again
  10. 10 check selection → cannot tell
  11. 11 click size EU 44 again
  12. 12 check selection → cannot tell
  13. 13 give up and fly home
Where it went wrong
The size selector.
In the bird's words

“I clicked EU 44 three times. Nothing on the page will tell me whether it took. I'm not putting a shoe in a basket on a hunch.”

What that means for you

A person sees the highlighted box. The agent can't confirm the click landed, so it loops until it runs out of patience. Your checkout never hears about it.

You were early to every channel that mattered. Be early to this one too.

AI agents are already visiting websites for customers. When they can’t understand a page, they simply leave. In analytics, that looks just like any other visitor who didn’t buy. You’d never launch without checking Safari. Almost nobody is checking this (yet).

Coalbird HQ board packet
AI CEO memo

Agents need roads, not vibes.

Experiment Automate a store purchase
Outcome Blocked before cart
Human assist Needed at variant proof
Next build State verification rail
Why you haven't caught it

Like session replay for agents

Agents don't leave Trustpilot reviews. That's why Coalbird tells you exactly what's broken in the agent experience, even when you can't see it.

What you runWhat it provesWhat it can't tell you
Playwright / Cypress The button works when you click it. Your test knows the selector. An agent has to find it.
Browser & device matrix The page renders correctly everywhere. Rendering isn't comprehension. It looks right and reads wrong.
Accessibility audit The markup is labelled and reachable. Two perfectly labelled buttons can still be impossible to tell apart.
Analytics Exactly where people drop off. Never why — and an agent giving up looks like anyone else leaving.
What Coalbird actually does

Give it your store. It figures out the rest.

No test plans. No prompts. No journeys to build. Coalbird explores your store like an AI customer would, then gets to work with recommendations and fixes.

Coalbird ops weekly watch
Runs42
Blocked7
Fixed12
Northstar Running Variant proof Blocked
House & Table Returns policy Passed
Everyday Skin Bundle compare Review
Proof confidence
62%
  1. 1

    It learns your store

    Coalbird reads your products, collections, options, policies and checkout. Then it works out what real customers might ask an AI to do.

  2. 2

    It sends in the agents

    It creates its own missions and runs them again and again. Find this. Compare that. Pick the right size. Build a basket. Check out. Every failure teaches it where your store gets harder for AI.

  3. 3 Coming soon

    It fixes what it finds

    Coalbird traces the problem back to the page, product data or interaction causing it, makes the fix, and runs the mission again to prove it worked.

What Coalbird is not

We're here to tell you prompt tracking is bullshit.

  • Not prompt tracking. Typing the same 50 prompts into ChatGPT every week and watching whether your brand appears is not testing. It’s sampling noise and calling it insight.
  • Not another readiness score. We don’t tell you your site is “82% agent-ready.” We send agents through real journeys and see if they can actually finish them.
  • Not GEO dressed up as QA. Ranking in an answer is not the same as being usable once an agent reaches your site.
  • Not a crawler report. Knowing an AI bot visited your store tells you nothing about whether it understood the product, chose the right variant or made it to checkout.
  • Not a dashboard you stare at. Coalbird finds the break, shows you why it happened, and helps you ship the fix.

We don’t measure whether AI talks about you. We measure whether AI can buy from you.

Things the bird keeps finding

It's never the thing you'd guess.

All four of these pass an accessibility audit, a browser matrix and humans. All four stop an agent cold.

  • Ambiguity “Two buttons. Both say Select. One is the size and one is the colour and I have no way to know which.”
  • Interstitials “A cookie wall, then a country picker, then an advert for your app. I never reached the shop.”
  • Stale state “The basket said empty. The basket was lying. I added a second one to be safe.”
  • Buried facts “You put the important bit inside a photograph. I cannot read photographs.”
Personnel file

About the bird.

Somebody has to go in first. It used to be a job with a cage and a lamp. Now it's a job with a browser.

Pricing

Plans from $79/month. Start with a guided trial.

A light base plan plus usage-based mission credits. Keep monitoring on, then add runs only when your store changes.

Resources

Benchmarks, field notes, and examples
Free
  • The Hundred benchmark of audited stores
  • The Agentic Commerce Readiness Report
  • Field notes showing what agents actually trip over
View resources
No card

Free trial

10 mission credits, a rerun, and a readout
$0trial after a scoping call
  • Native intake first, then an embedded trial booking
  • 10 mission credits on your live store
  • One rerun after a fix, deploy, or a week of drift
  • Plans start at $79/month when you are ready
Start trial

Paid plans start at $79/month with included mission credits and usage-based overages. Compare plans →