Silicon Valley GirlGoogle’s AI Search Expert: How to Get Ahead Before AI Changes Everything
CHAPTERS
AI recommendations are the new growth lever for businesses
Marina frames the stakes: AI-driven recommendations in ChatGPT/Google can make businesses “blow up overnight.” She introduces Robby Stein (VP of Product, Google Search) as the person who helps shape ranking and discovery in Google’s new AI experiences.
- •AI mentions can rapidly change demand (restaurants going viral from AI recommendations)
- •Visibility shifts from “ranking for people” to “being legible to AI systems”
- •Robby Stein’s role and why his perspective matters for search and business strategy
From keywords to natural language: how Google Search behavior is expanding
Robby explains that search is evolving from keyword queries to full natural-language questions, without replacing classic search use cases. Google’s AI increasingly uses web and world context to make answers easier and more complete.
- •Search still serves traditional intents (research, finding websites), but expands into richer queries
- •Natural-language, multi-sentence questions reduce the need for “keywordese”
- •AI leverages Google’s context about the web, products, and the world to improve results
Personalization: what Google can (and can’t yet) use about you
Marina presses on whether Search uses private Google data (Gmail, Drive, YouTube analytics). Robby describes opt-in personalization experiments and early Labs capabilities, while noting broader integration is still TBD.
- •Personalization is positioned as opt-in and experimental (announced at I/O)
- •Early personalization in Labs: shopping and local restaurant recommendations
- •Future possibility of connecting more Google services (e.g., Gmail) for more helpful search experiences
AI Mode feature tour: local recommendations that feel “all-in-one”
They walk through AI Mode returning lunch recommendations tailored to Marina’s location (Los Altos). Robby highlights how AI reasoning is combined with Google’s place context—hours, menus, reviews—so users can browse inside one experience.
- •AI Mode uses location context to surface relevant nearby options
- •Results integrate rich place data: hours, menu highlights, reviews
- •Product goal: combine AI reasoning + Google knowledge into a single browsing flow
Live demo: AI Mode searches, compares, and books restaurants for you
Robby demonstrates a task-style flow where Google fans out across sources like OpenTable/Resy, researches options, and returns bookable times. The point is time compression: what used to take 15 minutes becomes a few clicks.
- •AI Mode performs multi-source research in the background
- •Outputs actionable booking options with available times
- •Focus on reducing user effort for multi-step decisions (discovery → comparison → booking)
How AI Mode ranks results: query fan-out + Google’s knowledge systems
Robby explains the mechanics behind recommendations: a reasoning model breaks a prompt into many related searches (“query fan-out”), uses Search as a tool, and pulls from Google’s real-time knowledge bases (including local place data). The final recommendations reflect constraints like vibe, occasion, and review signals.
- •“Query fan-out” runs dozens of related sub-queries behind the scenes
- •AI uses both web results and Google knowledge bases/real-time systems
- •Local results can draw from hundreds of millions of places and their updated listings
- •Prompt nuance matters (date night, allergies, group size) because it drives sub-queries
Ads in the AI era: not disappearing, but evolving inside AI experiences
Marina asks whether AI recommendations are pay-to-play and if Google Ads will fade. Robby says AI Mode’s local recommendations don’t use ad info directly, but ads aren’t going away—search usage is expanding, and Google is experimenting with ad formats in AI experiences.
- •AI Mode recommendations are based on web + Google info systems, not ads data
- •Businesses can improve eligibility via complete/claimed local listings and menus
- •Google Ads likely persists as usage expands into new query types (multimodal, complex)
- •Early experiments: ads appearing within AI Mode / AI experiences, with product quality first
Live demo: agentic calling to offline businesses (pet grooming)
Robby shows a feature where Google collects requirements (pet type, size, service) and then places calls to local businesses on the user’s behalf. The system returns an email with quotes, availability, and which businesses couldn’t be reached, illustrating AI as an “offline concierge.”
- •User specifies constraints; AI initiates multiple calls to local businesses
- •Targets businesses without easy web booking—common for small local operators
- •No call recordings; results returned via email after several minutes
- •Output includes pricing/availability and unreachable providers
How to get recommended by AI: PR, trusted mentions, and “AI-readable” credibility
Marina asks for actionable guidance to improve AI recommendations for her businesses. Robby emphasizes that AI models evaluate signals similarly to people: reputable mentions, listicles, and widely found articles help establish reliability, alongside classic “helpful content” practices.
- •AI surfaces businesses mentioned in credible third-party articles and “top lists”
- •PR becomes a discovery input not just for humans, but for AI retrieval
- •Traditional best practices still matter: clear, helpful content that answers real questions
- •AI uses Search tool-calls, so ranking and discoverability influence AI inclusion
Reviews and reputation manipulation: why it’s messy and hard to reduce to one lever
They discuss purchased reviews and how that might impact recommendations. Robby avoids overpromising specific heuristics, reinforcing that systems look for broadly reliable, helpful information—similar to how top-ranked pages tend to be trustworthy for a query.
- •Bought reviews complicate trust; no single “hack” determines selection
- •Helpful, reliable information across the ecosystem remains the durable strategy
- •Think in terms of: what would rank top for this query if you Googled it?
Smarter search strategy: tools to find demand (Trends, Ads estimates, Search Console)
Marina asks how small businesses can understand what people are looking for in order to be recommended. Robby points to underused resources like Google Trends, ad traffic estimators, and Search Console, especially as queries become longer, more specific, and multimodal.
- •Google Trends provides real-time visibility into rising interests and keywords
- •Ads tooling can reveal traffic estimates and demand patterns
- •Search Console + other Google tools help map what users actually seek
- •Growing importance of long, specific queries and multimodal inputs (images/voice)
- •Potential future: broader visibility into new AI-style search behavior
AI shopping and multimodal discovery: Google’s shopping graph + photo search
Robby positions Google’s advantage in commerce: a massive shopping graph with live inventory/price updates, now connected to AI reasoning. They demo taking a photo of a product to find similar items, and discuss why purchases still typically hand off to merchant sites for safety and error reduction.
- •Google’s shopping graph: tens of billions of products with frequent updates
- •AI Mode can answer product questions with price/availability context
- •Photo-based shopping: prompts improve results (e.g., “similar ingredients”)
- •Purchasing remains cautious; “one-click buy” risks costly mistakes, but UX will keep simplifying
Google vs ChatGPT: what Google believes is its durable edge
Marina asks about the shift from “Google it” to “GPT it.” Robby argues Google’s strength is high-quality information retrieval grounded in Google’s knowledge systems (Finance, local, shopping) plus the open web, and notes fast growth in visual search via Lens.
- •Core differentiation: Google knowledge + tool-calls into specialized systems (e.g., Finance)
- •Search is optimized for information quality and verification (places exist, sources to click)
- •Lens/visual search growth is a major trend; multimodal inputs are accelerating
- •Google frames AI as extending search, not replacing the web exploration model
AI for inspiration and home design: using web imagery + live visual Q&A
Robby describes a newer focus: making AI better at inspiration (decor, design), an area where chat-style AI historically struggles. They show experiences that pull inspirational web images and enable follow-up questions, including a live camera mode that identifies objects in real time.
- •Inspiration is repositioned as a key “gap” AI can address in search
- •Results are based on web images (not necessarily generated), enabling exploration
- •Live camera mode enables real-time questions about what you’re seeing
- •System can guide discovery but still won’t complete purchases autonomously
Building products that stand out in the AI era: find gaps, talk to users, test for stickiness
The conversation shifts to product strategy in 2025: building is easier, so differentiation comes from insight, not code. Robby advises being a “student of gaps,” running deep user research, and identifying the moments that make people adopt—or abandon—your product.
- •As building becomes commoditized, “interesting/useful ideas” matter more than technical difficulty
- •Method: observe users, interview them, watch real behavior over time
- •Key questions: when did you love it and want it forever? when did you decide to stop?
- •Small samples can be enough if you go deep (≈ a dozen intensive interviews)
- •Mainstream consumer products often require daily value (DAU/stickiness) vs niche utilities with different models
