SEO basics like canonicals, sitemaps, robots.txt, etc. are not automatically set up just because an AI tool builds your website. That’s the lesson Google Search Relations team members John Mueller and Martin Splitt learned in a recent episode of Search Off The Record.
Both tested AI coding tools for personal projects and found the same gaps. Although these tools quickly created functional sites, proper SEO still required specific technical direction.
Tell the AI to “Add SEO”
Mueller likened the experience of coding Vibe to working with a developer who doesn’t specialize in search.
Mueller said on the podcast:
“At any time, you can tell the AI system, “Add SEO.” But how it works is if you go to a developer and add SEO, what do you mean? Sprinkle in some meta tags and add structured data. ”
Ambiguous instructions produce ambiguous results, whether created by humans or AI. Moller said he got better results by telling the system what he wanted from the beginning. This includes your domain name, canonical setup, sitemap file, and robots.txt.
He checked that the pages used proper HTML and were properly linked. We also set up pre-publishing checks to ensure that the URL returns content and that the JavaScript file is not blocked by robots.txt.
what they built
Mueller has been building a test website to see how Googlebot handles requests. He deployed them to Firebase hosting using Hugo as a static site generator and GitHub for version control.
He recently switched from VS Code using Copilot to command line tools. He named the ones he currently uses Claude Code and Gemini CLI.
Splitt tried Google AI Studio to build client-side tools using JavaScript. He described the output as easy to read and looks like a standard Next.js application. However, I ran into a loop where the AI kept using libraries I didn’t want.
Split said:
“I requested it for about 30 minutes. I tried to stop it from doing what I wanted it to do, and I tried to get it to do what I wanted it to do. And it was weird.”
Technical knowledge questions
Both acknowledged that there is tension in Vibecoding’s promise that you don’t need to know how to code.
Mueller said technical understanding is helpful every step of the way. Understanding the type of site generator you need and how to configure pre-publishing checks will give you better results. Without that background, the AI will make guesses. You might choose a static site generator, a JavaScript-heavy setup, or a complete CMS with a database backend.
Muller said:
“These are all reasonable assumptions, and if you talk to developers, they’ll make these assumptions, too. But you just tell an AI system, ‘I want a website,’ and it picks one.” ”
For personal projects or low-risk static sites, the risk is low enough to experiment. But when it comes to anything involving user data or production services, Mueller added, you need people who know what they’re doing.
related: Google’s John Mueller flags SEO issues with Vibe-encoded websites
Vibe coded site and search visibility
The site Mueller built produced reasonable HTML that was unobtrusive as atmosphere coding.
“The truth is, no one can actually tell that this is a vibecoded website,” he said, adding that common vibecoding frameworks can leave recognizable patterns.
He also pointed to risks related to content. If your site looks polished, you’ll want AI to write your content as well. Muller acknowledged that the tool can do that, but said that’s not where it’s most valuable.
Split agreed. Seeing content written by AI begs the question of why someone would visit a site without talking directly to the AI.
Muller has previously noted similar gaps in mood-coded sites. He reviewed the vibecoded Bento Grid Generator on Reddit. He identified issues with crawlability, outdated meta tags, and content stored in JavaScript files that were inaccessible to search engines.
See also: How to use AI to quickly identify migration issues
Looking to the future
The podcast contained no formal guidance or policy positions regarding mood-coded sites. Muller and Split shared what they had tried and encountered.
The message to those testing these tools is that AI can handle some of the code generation just fine, especially for low-risk projects. It does not make SEO decisions on its own. These still require someone who knows what to look for.
See also: Why Google uses Vibe coding for its search algorithm
Featured image: YouTube.com/GoogleSearchCentral, May 2026.
