AI agents are a new kind of user
Agents browse, compare, and buy on behalf of humans. We test whether yours can complete real tasks and show exactly where they fail.
This is an additional usability layer for the agentic web. It does not replace QA and does not guarantee every AI agent can use your product.
Check what an agent can see and use on your website
Hover over the example page. The lens shows what an AI agent sees at the same spot: readable structure, missing buttons, and broken forms.
What you get
Real tasks, real agents
We define user goals that matter for your product. Then autonomous agents attempt them.
Multiple attempts & personas
Repeated runs separate random agent mistakes from consistent product blockers.
Screen recordings & screenshots
Watch exactly where agents hesitate, loop, misclick or give up.
Failure traces & fix roadmap
Step-by-step traces plus a prioritized list of what to fix first.
Sample report
This is what your Audit Report looks like
Score, task matrix, session video, and a fix roadmap — so you can see what agents get stuck on.
automationexercise.com
Product discovery works —
cart completion stays inconsistent
Needs work
What breaks
Agents reliably find products, but the cart confirmation modal and checkout CTA are flaky: some runs complete, others stall on unlabeled controls.
Where it hurts
Mixed agent-driven conversion: discovery is solid, but purchase completion is unreliable. Fetch agents also struggle to quote price from the initial HTML.
First move
Give cart and checkout CTAs clear text labels, then server-render product price in the first HTML response.
Session video
Add product to cart · Attempt 2 · fail
Every Audit Task includes a recording of the agent run — watch exactly where it stalled.
Five-pillar score
Consistency matrix
| Task | Attempt 1 | Attempt 2 | Attempt 3 |
|---|---|---|---|
| Search product and read product detail | pass | pass | partial |
| Add product to cart | pass | fail | partial |
| Complete checkout | partial | fail | partial |
Tasks tested
- Search product and read product detailpass
Open Products, search for "Men Tshirt", open a matching product detail page, and report name, category, price, availability, condition, and brand.
- Add product to cartpartial
From a product detail page, add the item to the cart and confirm the cart shows the product.
- Complete checkoutpartial
From the cart, proceed through checkout until an order confirmation or payment step is reached.
Fix roadmap
- 1Give the cart confirm CTA a clear text label and ensure it is a real button, not an icon-only control.
- 2Server-render product price (and stock) in the first HTML response so fetch agents can quote products.
- 3Add JSON-LD Product + Offer (name, price, currency, availability) on product detail pages.
Try it out
Try it for free
Your first Audit Task is free — one task, three attempts, per domain. See how agents actually use your site.
Start hereCommon questions
What is the Agent Test and how is it different from the scan?
The free scan answers 'Can agents read your site?'. The Agent Test is an Agent Usability Audit. It answers 'Can agents actually use it?'. Autonomous agents attempt real tasks (sign up, book a demo, complete checkout) with multiple attempts, screen recordings, failure traces and a prioritized fix roadmap. Quick Test starts at 99 €; Full Agent Audit is 299 €.
Will a better score guarantee that ChatGPT recommends my site?
No, and be wary of anyone who promises that. The score measures whether your site is easy for agents to discover, read and use. That improves your chances of being understood and cited correctly, but no one can guarantee rankings in AI systems.
How often should I re-scan my site?
Re-scan after every fix you ship, and roughly once a month otherwise. Agent ecosystems evolve quickly, and reports are kept at a permanent URL so you can track your progress over time.