Why AI agents get stuck on websites: six common walls
Where agents stop, and why it is usually not the agent's fault.
Am I Ready for Agents? editors · 4 September 2026 · Reviewed September 2026
Agents usually fail for ordinary reasons: they are blocked, the page is empty until scripts run, the facts are in an image, a form will not accept input, a login has no standard flow, or a challenge appears at every step. Most of these also make a site harder for people.
Which walls come up most often?
- Blocked at the door: robots.txt or bot protection refuses the agent, sometimes by accident.
- Empty pages: the text appears only after JavaScript runs, and the agent does not run it.
- Facts in pictures: prices, sizes or opening hours appear only in images, PDFs or pop-ups.
- Forms that fight back: custom date pickers, unlabeled fields and errors that do not say what went wrong.
- Logins with no standard flow: the agent cannot ask the person for permission in a way it recognises.
- Challenges at every step: CAPTCHAs meant for bots stop the ones acting for customers too.
Is this the same as poor accessibility?
Often, yes. Clear labels, real text and predictable forms help screen reader users and agents for the same reasons. ora.ai's methodology puts it from the agent's side: blocked crawlers, JavaScript-only content and weak structure lead agents to describe a business wrongly. is-agentic.com's essential checks include server-rendered content, document structure, error handling and usable controls.
That overlap helps when you make the case internally: many fixes that help agents also help people who use assistive technology.
Which wall should you check first?
The one closest to your main task. For a store it is usually checkout; for a booking business, the date and availability step; for a publisher, whether the right crawlers are allowed in. The self-check will point to it.
Then work outward from it one step at a time, because an agent that clears one wall meets the next. If the answer is not obvious, the 30-minute review is a quick way to find it with your team.
How do you find out which wall you have?
Three ways, from cheapest to most thorough. Answer the self-check, whose questions map to these walls. Walk through your main task yourself as a guest and note every step where a person would have to step in. Then run a scanner or a tool that sends an agent at the page, as described on tools that check; ora.ai, for example, says its Journey tool shows what an agent found, where it stalled and whether it finished.