Google Search Relations team members John Mueller and Martin Splitt discussed vibe coding websites on a recent episode of Search Off The Record.
Both have found that AI coding tools can produce functional sites quickly. But getting SEO right still requires some technical guidance, the same kind you’d give a human developer.
Telling AI to ‘Add More SEO’
Mueller compared the experience of writing about vibe to working with a developer who doesn’t specialize in search.
Mueller said on the podcast:
“You can always tell an AI program, now add SEO to it. But how that works is if you go to an engineer and add some SEO and it’s like, what do you mean. Sprinkle meta tags and add structured data.”
Imprecise instructions produce imprecise results, whether the constructor is human or AI. Mueller said he got better results by telling the program what he wanted in the first place. That included the domain name, canonical settings, sitemap files, and robots.txt.
Check that the pages use logical HTML and are well linked. He also set up a pre-publish check to ensure that URLs returned content and that JavaScript files were not blocked by robots.txt.
They didn’t build it
Mueller has been building test websites to see how Googlebot handles requests. Submitted them to Firebase hosting using Hugo as a static site generator, with GitHub for version control.
He recently replaced VS Code with Copilot to use command line tools. He named Claude Code and Gemini CLI as what he currently uses.
Splitt experimented with Google AI Studio to build a client-side tool in JavaScript. You described the output as readable, which looks like a typical Next.js application. But he hit a snag when the AI kept using a library he didn’t want.
Splitt says:
“I begged it for half an hour. I tried to get it to not do what it wanted to do, and to make it want to do what I wanted to do. And that was weird.”
Technical Knowledge Question
Both agreed to the ambivalence of the writing vibe’s promise that you don’t need to know how to code.
Mueller noted that technical understanding helps at all levels. Knowing what type of site generator you are looking for and how to schedule testing before publishing will produce better results. Without that background, the AI will do the thinking. It may choose a static site generator, a JavaScript-heavy setup, or a full CMS with a database backend.
Mueller says:
“These are all logical assumptions where when you talk to the engineers they will make these assumptions. But if you tell an AI program like I’m looking for a website, it will pick one.”
For personal projects and low-risk static sites, the stakes are low enough to be tested. But for anything that involves user data or a productivity service, Mueller added that you want someone who understands what they’re doing.
Sites with Vibe code and search visibility
The sites created by Mueller produced logical HTML that would not stand out as vibe-coded.
“Simply speaking, no one can see that you’re the same, this is a website with a vibe code,” he added, adding that the standard frameworks for writing vibe codes can leave visible patterns.
He also pointed out the danger associated with the content. If the site looks polished, it’s tempting for the AI to write the content too. Mueller acknowledged that the tool could do that but said that’s not where he sees much value.
Splitt agreed. AI-authored content raises the question of why a person would visit a site instead of talking to the AI directly.
Mueller has flagged similar spaces on vibe-coded sites before. He reviewed the vibe-coded Bento Grid Generator on Reddit. He identified problems with clarity, outdated meta tags, and content stored in JavaScript files that search engines could not access.
Looking Forward
The podcast did not include official guidance or policy positions on sites with the vibe code. Mueller and Splitt were sharing what they tried and what they experienced.
For people testing these tools, the message is that AI can handle parts of code generation well, especially for low-risk projects. It does not make SEO decisions on its own. Those still need someone who knows what to ask for.
Featured Image: YouTube.com/GoogleSearchCentral, May 2026.
