About This Episode

I sat down with Sean Barber on Search With Sean for a long conversation about AI agents, websites built for machines, and how to cut through the hype that surrounds all of it. We started with what an AI agent actually is: software that can stop, make a decision, and choose what to do next, which means it can do almost anything a human can do on your website. For that to work, it has to be able to read the page. Most AI systems don't render JavaScript, so if your pricing, availability, or sizing only appears after scripts run, those systems never see it. I suggested a quick test: disable JavaScript, load a product page from any big sports brand, and then compare what is left with what the schema claims.

On SEO versus GEO, I compared it to moving from the 60 metre dash to the 100 metre dash. It is a different race, but if you have been doing the fundamentals properly (a clean entity, a site that is easy to crawl and read), the training is mostly the same. The people who should worry are the ones who built on keyword stuffing and programmatic content that doesn't provide any value.

Sean works with cancer content, so we spent time on hallucination in your-money-your-life topics. We talked about the Munich Regional Court ruling from May 2026, which found that Google can be held liable for false statements in its AI Overviews. You cannot fully control what an AI says about you, but you can control how consistent you are. I used the analogy of a group drawing a face where each person gets one feature and nobody gets a reference photo. That is how most brands build their web presence. The fix is to write your canonical identity down in one document and give it version control, the same way we already version README files.

On measurement, I argued that time on page and "sticky" metrics were probably the wrong things to track for most sites all along. The best website, for humans and agents alike, is the one where the visitor can finish their task as quickly as possible. The bigger organisational problem is that nobody owns AI optimisation yet. It isn't dev, SEO, UX, or CRO, and that will be the challenge for most companies over the next year or two. Informational sites are in the hardest position, because the whole value of an LLM as a summary engine is that you don't need to visit the source.

For tracking agents, we are back to the ancient art of reading server logs. I recommended starting with where agents acting on a user's behalf land on your site, then sending the commercial agents people actually use (Claude, ChatGPT, Google's) to complete real tasks, instead of relying on a custom Playwright setup that nobody else will use. From the agentic form testing I've been doing, two failures stood out. When a form has no programmatic success message, the agent keeps resubmitting it because it can't tell it succeeded. And browser-native validation errors are invisible to agents because they are not in the accessibility tree, so the agent keeps failing without knowing why.

We closed with predictions. The open web is splitting into open and gated, with pay-to-crawl and pay-to-access becoming normal, and trust and authenticity becoming far more important. I also talked about why chatbots only ever recommend three or four brands in a category (I tested this with air fryers), and why that shuts new players out of the conversation. My advice for the AI news cycle was to keep the receipts and check what happened to each big announcement three, six, and nine months later. OpenAI is the clearest example. The Sora app closed in April 2026, Instant Checkout in ChatGPT was wound down after very few merchants signed up and conversion came in below retailers' own sites, and the Atlas browser was shut down less than a year after its launch keynote, announced in a help center article about "evolving" it. The biggest lesson from 200+ episodes of No Hacks is that the top experts are the ones who are happy to be proven wrong, and I shared the Lukas Vermeer interview that taught me that. My last word was that anyone who promises to rank you number one in ChatGPT is someone to run from. AI visibility is like dental health: flossing and brushing won't guarantee anything, but they raise your chances, and that is where you start.

Key Topics Discussed

• What an AI agent is and what it can do on a website
• Why most AI systems don't render JavaScript, and the disable-JavaScript test for product pages
• SEO vs GEO: a different race, but mostly the same training
• Hallucination in your-money-your-life topics and how much of it you can control
• Writing down your canonical brand identity and giving it version control
• Whether time on page and "sticky" metrics were ever the right things to measure
• Why no team (dev, SEO, UX, CRO) owns AI optimisation yet
• Why informational sites are hit hardest, and giving people a reason to visit
• Tracking AI agents with server logs and testing with the agents people actually use
• Agentic form testing: missing success messages and invisible validation errors
• The web splitting into open and gated (pay-to-crawl, trust and authenticity)
• Why LLMs only recommend three or four brands, and what that does to competition
• Keeping the receipts on AI announcements (Sora, Instant Checkout, Atlas)
• Lessons from 200+ episodes of No Hacks