AI Overviews don’t show up consistently across Google Search because the system learns where they’re useful and pulls them back when people don’t engage.

Robby Stein, Vice President of Product at Google Search, described in a CNN interview how Google tests the summaries, measures interaction, and reduces their appearance for certain kinds of searches where they don’t help.

How Google Decides When To Show AI Overviews

Stein explained that AI Overviews appear based on learned usefulness rather than showing up by default.

“The system actually learns where they’re helpful and will only show them if users have engaged with that and find them useful,” Stein said. “For many questions, people just ask like a short question or they’re looking for very specific website, they won’t show up because they’re not actually helpful in many many cases.”

He gave a concrete example. When someone searches for an athlete’s name, they typically want photos, biographical details, and social media links. The system learned people didn’t engage with an AI Overview for those queries.

“The system will learn that if it tried to do an AI overview, no one really clicked on it or engaged with it or valued it,” Stein said. “We have lots of metrics we look at that and then it won’t show up.”

What “Under The Hood” Queries Mean For Visibility

Stein described the system as sometimes expanding a search beyond what you type. Google “in many cases actually issues additional Google queries under the hood to expand your search and then brings you the most relevant information for a given question,” he said.

That may help explain why pages sometimes show up in AI Overview citations even when they don’t match your exact query wording. The system pulls in content answering related sub-questions or providing context.

For image-focused queries, AI Overviews integrate with image results. For shopping queries, they connect to product information. The system adapts based on what serves the question.

Where AI Mode Fits In

Stein described AI Mode as the next step for complicated questions that need follow-up conversation. The design assumes you start in traditional Search, get an Overview if it helps, then go deeper into AI Mode when you need more.

“We really designed AI Mode to really help you go deeper with a pretty complicated question,” Stein said, citing examples like comparing cars or researching backup power options.

During AI Mode testing, Google saw “like a two to three … full increase in the query length” compared to typical Search queries. Users also started asking follow-up questions in a conversational pattern.

The longer AI Mode queries included more specificity. Stein’s example: instead of “things to do in Nashville,” users asked “restaurants to go to in Nashville if one friend has an allergy and we have dogs and we want to sit outside.”

Personalization Exists But Is Limited

Some personalization in AI Mode already exists. Users who regularly click video results might see videos ranked higher, for example.

“We are personalizing some of these experiences,” Stein said. “But right now that’s a smaller adjustment probably to the experience because we want to keep it as consistent as possible overall.”

Google’s focus is on maintaining consistency across users while allowing for individual preferences where it makes sense.

Why This Matters

In July 2024, research showed Google had dialed back AIO presence by 52%, from widespread appearance to showing for just 8% of queries. Stein’s description offers one possible explanation for that pattern.

If you’re tracking AIO presence week to week, the fluctuations may reflect user behavior patterns for different question types rather than algorithm changes.

The “under the hood” query expansion means content can appear in citations even without matching your exact phrasing. That matters when you’re explaining CTR drops internally or planning content for complex queries where Overviews are more likely to surface.

Looking Ahead

Google’s AI Overviews earn placement based on usefulness rather than appearing by default.

Personalization is limited today, but the direction is moving toward more tailored experiences that maintain overall consistency.

See the full interview with Stein below:

Featured Image: nwz/Shutterstock