Start with a real task.
Choose something visitors actually need to complete: book an appointment, find a service, update details or submit a request. Follow it from the first page to the final confirmation.
One carefully tested task tells you more than an “AI-ready” badge.
Make the structure understandable
Use meaningful headings, labels, links, form controls and page landmarks. Give controls names that describe their purpose. Keep validation messages close to the field and explain what needs to change.
Make actions explicit
A structured action needs a clear name, documented inputs, predictable outputs and validation. Do not expose internal operations just because they are technically available.
Preserve the rules
Authentication, authorisation, consent, rate limits and business rules must apply to agent-initiated actions as they do to human actions. High-impact steps need deliberate confirmation.
Design for recovery
Users need to know what happened, what failed and how to continue. Handle timeouts, duplicate submissions, partial completion and expired sessions deliberately.
Accessibility benefits everyone
A well-structured, accessible website is easier for assistive technology to interpret and often easier to make agent-readable. Test with real assistive technologies and users; do not assume accessibility from markup alone.
Do not rebuild by default
Sometimes semantic fixes, better task flows and a few carefully scoped actions are enough. Sometimes deeper integration is necessary. Find out before selling a rebuild.
Quick tips
- Test keyboard-only navigation.
- Check labels and error messages with a screen reader.
- Test mobile session expiry and recovery.
- Verify that every consequential action has an approval step.