Google On AI Overviews & AI Mode Being Isolated Systems (Or Not)

Nikola Todorovic, Director of Software Engineering at Google Search, spoke about a presentation he gave at Zurich in December at Google Search Central live (I was there). He dug into it on a recent Search Off the Record named How AI Is Changing Google Search and SEO. In there, he explained how AI and machine learning have been used in isolation within Google Search.

The reason these technologies have been used in isolation (and I covered this before) is because it is hard to debug and fix these types of systems otherwise. If you use them in specific cases, you have a better chance at understanding why something went wrong and how to adjust it to fix it going forward. So it was used for SafeSearch, image comprehension, specific signals, etc.

Nikola explained that artificial intelligence machine learning models function kind of like a black box and you don’t always understand what’s happening underneath. The earlier models (linear models, etc) are the easiest to understand and debug. So Google used these AI solutions in isolation, like early on with SafeSearch. So it is easier to debug and deal with in that isolated space.

Then new AI technologies came out, like transformers, with BERT, MUM, etc. These systems were built in isolation, as new signals and supported the whole ranking infrastructure as one layer on top of it.

AI Overviews: AI Overviews are a normal search with a few fan out queries, he explained. So it triggers a more complex and longer query and the AI Overview combines it all into a summary. With AI Overviews, the whole retrieval and whole ranking system is the old style and the AI Overview feature is on top of it and operates on its own AI.

AI Mode: AI Mode communicates with longer queries and longer conversations and the users like the conversational aspect. AI Mode does use search, it uses fan out queries and it does have linked results and citations. But it is on its own because its infrastructure is new, so it runs on search but it has its own bigger platform.

So these systems are kind of their own thing, but they use other systems as well.

Nikola then spoke about the ecosystem and said, he doesn’t know if there is a “magic wand” but he believes you need to provide value to your users and that will help. With AI, bringing value continues to be important in this AI world. Learn the AI tools, do not neglect the technology, and use it to provide value and improve your content.

Here is the podcast, 100% worth listening to while you work:

Glenn Gabe also summed up his thoughts on this:

Some interesting nuggets from the latest Search Off The Record Podcast about Google using AI in Search rankings. Here is Google’s Nikola Todorovic, Director of Software Engineering at Google Search, about how AI is used to impact rankings and why simpler linear systems are… pic.twitter.com/J4GQRanyEO

— Glenn Gabe (@glenngabe) May 2, 2026

Forum discussion at X.