Why ChatGPT Agents Fail on Most Websites (But Not These)
The video tests ChatGPT's agent on booking tasks across multiple websites, finding stark performance differences.
While the agent took 8 minutes to book a restaurant table and struggled elsewhere, it completed a demo booking with SE Ranking in just 30 seconds and downloaded a gated PDF from HubSpot in 3 minutes.
The creator identifies four key factors that make websites agent-friendly: clear calls-to-action, the absence of pop-ups, logical information flow without redirects, and clean HTML with structured data.
The takeaway is that websites optimized for human users may inadvertently block AI agents, signaling that businesses need to adapt their UX now rather than wait for disruption to force change.
I asked JBT agent to book a table for four and it took 8 minutes. That's not so great. But when I asked it to book a demo with SE ranking, it took 30 seconds. So I started wondering, is it agents that are the problem or are most businesses just not ready for semi-autonomous tools like chatbt agent? The answer to this question came to me in my next test where I asked it to download a gated PDF of HubSpot's 2025 state of marketing report. It took 3 minutes. Not quite as quick as SE Rankings 30 seconds, but I couldn't expect it to do things as fast because of the number of steps involved. So, here's what I discovered. SE Ranking and HubSpot both work perfectly. Agents have no issues navigating these websites. So, what exactly are these websites doing differently? I condensed it down to four common denominators. Number one, both websites had a clear CTA. For HubSpot, that was a large orange button that said, "Get the trends." For SEO ranking, it was a large white button that said request a demo. Number two, both
websites had no pop-ups. As agents really struggle with pop-ups, more so than humans ever will. So, if you want to create an agent-friendly experience, make sure that you just get rid of pop-ups altogether. Number three, both websites had a logical flow. That means no weird redirects, no mystery steps, and no guessing on the agents part. And number four, and maybe even the most important, is that both websites had a clean HTML. You see, agents rely on structured data and semantic markup. If your website is a mess of divs and JavaScript, then agents just aren't going to be able to read them. So, here's the takeaway. Just because your website works for people doesn't mean it works for agents, too. And maybe inconsistent booking times don't signal a flaw in the agent. Maybe it's a warning sign that businesses just aren't ready for them. So, here's what we need to do. We need to start adapting now. That means not waiting for disruption to happen before we make changes. What we need to do is start fixing our funnels, which means making our UX machine friendly. And it also means testing our visibility the way an agent would see
it, not just the way a user would see it. So here's the takeaway after the takeaway. The AI user has arrived. You don't want to be invisible to them. So, it's time to start adapting to them