How LLMs discover and recommend products
Large language models don't crawl the web like Google does, but they are increasingly connected to real-time web access (ChatGPT with browsing, Gemini with Google Search integration, Perplexity as a web-powered answer engine). When a user asks "what's the best French press for home use under $50?", these systems retrieve and synthesize information from multiple sources to give a recommendation.
The sources LLMs rely on for product recommendations include: review sites and publications (Wirecutter, Consumer Reports, specialized blogs), your own product pages and content, structured data (schema.org markup) that makes product attributes machine-readable, UCP catalog data for real-time inventory and pricing, and your brand's authority signals across the web.
GEO vs SEO: what's different
Traditional SEO optimizes for: keyword density and placement, backlink authority, technical crawlability, and click-through rate signals. GEO optimizes for: factual accuracy and verifiability, comprehensive attribute coverage, third-party validation (mentions, reviews, citations), and machine-readable structure (schema.org, UCP). SEO and GEO overlap significantly in technical foundations, good SEO is a prerequisite for good GEO, but GEO adds a layer of content and data optimization that goes beyond keyword targeting.
Strategy 1: Write for natural language queries, not keyword queries
SEO-optimized copy is written to match short keyword queries: "best French press", "stainless steel French press". GEO-optimized copy anticipates the full natural language questions users ask AI assistants: "What's a good French press for someone who wants to make 4 cups at a time and doesn't want to deal with complicated cleaning?" Write product descriptions and content that directly answer these longer, conversational queries.
Practically: identify the top 10 natural-language questions users ask about your product category (use "People Also Ask" boxes, Reddit threads, customer service questions as sources). Create content that answers each question factually and completely. Use these as FAQ sections on product pages and dedicated content pages.
Strategy 2: Build factual authority with citable claims
LLMs favor content they can cite with confidence. Vague marketing claims ("the best quality on the market") are ignored. Specific, verifiable facts are cited: dimensions, materials, test results, certifications, customer return rates, manufacturing origin. For each product category you sell, identify the facts that matter most to buyers and make them prominently available, accurately stated.
For B2B products: include compliance standards met (ISO, CE, UL), technical specifications in standardized formats, and compatibility information. These are exactly the structured facts AI agents synthesize when making enterprise purchase recommendations.
Strategy 3: Earn mentions on authoritative sources
LLMs weight information from authoritative, frequently-cited sources more heavily than from less-referenced sources. Product mentions in Wirecutter, The Verge, Consumer Reports, or relevant specialty publications create "citation anchors" that LLMs use when making recommendations.
PR and content marketing for GEO purposes should focus on: securing product reviews in well-trafficked publications, getting your product featured in comparison articles ("best [category] of 2026"), contributing expert content to industry sites that discuss your product category, and building a Wikipedia presence for your brand where it's factually justified.
Strategy 4: Implement complete schema.org markup
Schema.org structured data is the most direct way to make product information machine-readable for AI systems. For each product page, implement schema.org/Product with: name, description, brand, sku, gtin, offers (price, availability, currency, URL), aggregateRating, review, and category-specific properties (for electronics: color, weight, dimensions; for food: ingredients, nutrition).
Also implement schema.org/FAQPage on product pages to make your Q&A content directly parseable by LLMs, and schema.org/HowTo for any instructional content related to your products.
Strategy 5: Deploy a llms.txt file
The llms.txt file (hosted at yourdomain.com/llms.txt) is an emerging standard that explicitly tells AI crawlers what content is available and how to access it. For e-commerce, a well-structured llms.txt file should list your UCP endpoint URL, your sitemap URL, your key product categories, and your brand information. AI agents that respect llms.txt will use this as a starting point for understanding your catalog.
Strategy 6: UCP as the real-time GEO layer
For AI agents that make real-time purchase queries (rather than relying on pre-indexed knowledge), UCP is the GEO mechanism. An agent that has learned your brand through content discovery will query your UCP endpoint to get current prices, availability, and product details at the moment of purchase. The quality of your UCP data directly determines your conversion rate in agentic commerce.
GEO and UCP are complementary: GEO gets your brand into AI agent "awareness" and preference, UCP enables AI agents to transact with you at the moment of purchase intent.
Measuring GEO success
Traditional SEO metrics (keyword rankings, organic traffic) partially capture GEO performance, but not completely. Additional metrics to track: brand mention volume across AI-generated content (use tools that monitor AI answer appearances), share of agentic commerce transactions attributed to AI assistant referral, and the gap between your SEO visibility and your GEO visibility for key product queries. Test regularly by asking ChatGPT, Gemini, and Perplexity about your product category and noting whether your brand appears in recommendations.