How do I test whether AI agents can use my website?

You cannot infer this from analytics or from reading your own code. You have to run agents against the live site and watch what they do.

1. Pick the journeys that matter

Not pages — journeys. "Find the price of X and add it to the basket." "Sign in and download the latest invoice." "Book the first available slot next week." Five is plenty to start.

2. Define success precisely

A journey passes when a specific, checkable end state exists: an order confirmation number, a file downloaded, a booking visible in the account. "The agent said it worked" is not a result — self-reported success is one of the most common false positives.

3. Run more than one agent, more than once

Different agents fail differently, and the same agent fails intermittently. A single run tells you almost nothing; ten runs give you a rate.

4. Record the exact failing step

The useful output is not "checkout is broken" but "the agent could not identify the delivery-option radio group because the labels are images." Capture the trace: what it saw, what it clicked, what came back.

5. Fix, then keep the test running

Treat these as regression tests. Agents change on their own schedule, so a passing journey is a statement about today. PlateTale runs this loop for you and alerts you when a journey stops completing.