Google's Robby Stein on AI Mode reaching 200 countries and why great SEO still matters in the AI era

Summary

Google's AI Mode has expanded to more than 200 countries and 40 languages within roughly two months of launching in the US in May 2025, according to Robby Stein, who leads product on the Google Search team. The rollout was gated by both policy considerations and infrastructure requirements, including multilingual quality checks for a multi-turn conversational product and the need for local compute to handle inference at scale.

AI Mode sits within google.com as a generative, end-to-end chat experience that executes what Stein describes as 'query fan-outs,' running dozens of underlying Google searches to construct responses. It is distinct from AI Overviews, which surface selectively on standard results pages where AI adds clear value. For simple lookups, sports scores, or quick phone numbers, the classic results page remains the default path.

Agentic and Commerce Capabilities

Google has already launched agentic booking in the US through AI Mode, pulling real-time restaurant availability from platforms including Resy and OpenTable and completing reservations inside the search experience. Stein frames this as the beginning of a broader shift from information retrieval to task completion, with shopping cited as a near-term priority.

A new visual AI layer within AI Mode now surfaces inspirational imagery and shoppable product grids for open-ended queries such as 'design a bedroom' or 'landscape lighting.' Stein points to price-change alerts and assisted purchase completion as logical next steps, positioning the feature as Google's answer to top-of-funnel commerce discovery.

Search Volume and Growth Signals

Search queries in the category of highly specific, complex questions are growing at roughly 10% year-over-year in large markets including the US and India, which Stein describes as an enormous absolute number given Google's query base in the billions. Visual search via Google Lens is seeing even sharper acceleration, with multimodal queries, such as photographing an item to shop it, rising 70% year-over-year.

Stein characterises the AI era as expansionary rather than substitutional. Users are not migrating away from search; they are asking questions they could not practically have Googled before.

What SEO Still Means Under AI

Stein's position on organic search strategy cuts against the narrative that AI disrupts traditional SEO. Because AI Mode and competing AI products routinely retrieve live web content when parametric knowledge falls short, high-authority, topically trusted pages remain the primary input into AI-generated answers. Content that ranks well in Google's core algorithm has a high probability of being pulled into the context window and surfaced to users.

The practical advice is to study Google's publicly available human rater guidelines, build original authoritative content for specific query types, and pay attention to where AI use is growing fastest, specifically advice, nuanced troubleshooting, and mental health-adjacent topics, which Stein identifies as disproportionately expanding use cases.

Video and Audio as Emerging SEO Surface

Stein acknowledges that previously unindexed or difficult-to-parse content formats, long-form video and audio chief among them, stand to gain search visibility as AI improves at segmenting, transcribing, and contextualising that material. A video conversation that previously required a manually created transcript to reach keyword-based rankings could, in the AI era, surface organically in response to nuanced thematic queries.

Publisher Controls and TPU Allocation

Publishers currently have opt-out mechanisms covering crawling and rich snippet inclusion. Stein does not detail further granularity but frames AI search as a net discovery driver for content creators rather than a traffic threat.

On compute, Stein confirms Google runs a formal internal TPU allocation process across Search, Google DeepMind, and Cloud, describing compute as 'by far the most important' infrastructure constraint. He declines to detail the prioritisation methodology beyond confirming Search is treated as one of the highest-priority workloads given its scale of AI interaction.