Google Routes Complex AI Overviews Queries to Gemini 3 Pro

Google is now routing complex queries in AI Overviews to its most capable model, Gemini 3 Pro, while faster models continue handling simpler searches.

The intelligent routing system, which Google first deployed in AI Mode last year, now extends to the AI-generated summaries that appear directly below search queries for 2 billion monthly users. The upgrade is available in English globally—but only for paying subscribers to Google AI Pro ($19.99/month) or AI Ultra.

“Behind the scenes, Search will intelligently route your toughest Qs to our frontier model while continuing to use faster models for simpler tasks,” wrote Robby Stein, VP of Product for Google Search, in the announcement.

Subscribers Get First Access to Frontier Reasoning

The tiered rollout continues Google’s strategy of reserving its most powerful AI capabilities for paying customers. While Gemini 3 Flash now powers the default experience for all users in AI Mode, accessing Gemini 3 Pro’s enhanced reasoning requires a subscription.

Google AI Pro subscribers receive 100 Gemini 3 Pro prompts per day, while AI Ultra users get 500. The company recently increased these limits—AI Pro’s “Thinking” mode allowance jumped from 100 to 300 daily prompts—suggesting subscriber demand for the more capable model.

The distinction matters because Gemini 3 Pro brings a significant capability gap over Flash. Google positions it as “highly capable at solving complex problems across a vast array of topics like science and mathematics with a high degree of reliability,” with a 1-million-token context window that can process roughly 1,500 pages of text.

For complex queries requiring multi-step reasoning or nuanced understanding, Pro’s deeper processing could meaningfully improve answer quality. But the paywall means most of AI Overviews’ 2 billion monthly users won’t see these improvements—they’ll continue receiving responses from the faster, less capable models.

Google Search Product Manager Robby Stein (X)

Query Fan-Out Powers the System

The routing intelligence relies on what Stein has previously called “query fan-out”—a technique where the AI model uses Google Search as a tool to perform additional queries, gathering more relevant information before synthesizing a response.

This approach differs from simple retrieval-augmented generation. Rather than pulling from a fixed set of indexed documents, the system dynamically expands its search based on the complexity and nature of the question, theoretically producing more comprehensive and accurate answers.

Google has been aggressively expanding where Gemini appears across its products. The same week as the AI Overviews upgrade, the company brought Gemini AI to Gmail with summaries and smart replies, and earlier launched Personal Intelligence features that connect Gemini to users’ personal data across Google services.

The company reported that the Gemini app now has over 650 million monthly users, more than 70% of Google Cloud customers use Google AI, and 13 million developers have built with its generative models.

The Paywall Question

Google’s decision to gate the improved AI Overviews behind a subscription raises questions about the future of search. The company built its dominance on providing free access to the world’s information; now its best AI-powered search requires $20 monthly.

The competitive calculus is clear. OpenAI’s ChatGPT search, Perplexity, and other AI search engines are all vying for users who want AI-powered answers. Google can’t afford to fall behind on capability, but giving away its most expensive models for free would crater margins.

For now, the compromise is a two-tier system: everyone gets AI Overviews powered by fast, efficient models, while subscribers get access to frontier reasoning when questions get hard. Whether that division holds as AI search competition intensifies—or whether Google eventually pushes more capabilities behind the paywall—remains an open question.

The upgrade also arrives just days after Google pulled AI Overviews from certain medical queries following a misinformation report, a reminder that more powerful models don’t automatically mean more trustworthy answers. As Google routes complex queries to its most capable AI, the stakes for getting those answers right only increase.