For two decades, ecommerce search visibility has been dictated by traditional Search Engine Optimisation (SEO). You built backlinks, optimized keyword density, and tried to rank high on a page of ten blue links. But consumer behavior is shifting. Shoppers are increasingly bypassing Google Search entirely, opting to ask conversational AI agents like ChatGPT, Gemini, and Perplexity complex, multi-variable questions: "Find me a minimalist, waterproof running jacket under £150 that fits true to size and ships to London within 2 days."
Traditional SEO cannot capture this intent. To get your products recommended by AI, you must engage in Generative Engine Optimisation (GEO). GEO is categorically different from SEO. It does not rely on domain authority or keyword stuffing; it relies exclusively on semantic data structuring, entity relationships, and rigorous schema validation. If your product data is not mathematically structured for a Large Language Model (LLM) to parse, your brand simply will not exist in the AI-first web.
How AI assistants actually index and retrieve product content
When a user asks Perplexity or ChatGPT a shopping query, the engine does not "read" your beautifully designed homepage. It executes a retrieval augmented generation (RAG) workflow. First, it identifies the core entities in the query (e.g., "waterproof," "running jacket," "under £150"). Then, it queries search indexes (like Bing API for ChatGPT) and its internal vector database to find exact semantic matches.
The AI agent evaluates the retrieved data based on confidence and completeness. If your product page has a generic title and a paragraph of marketing fluff, the agent's confidence score drops because it cannot definitively extract the material or the exact price. However, if your page exposes a strict JSON-LD schema defining the exact properties, the agent parses this instantly, scores it with high confidence, and outputs your product as the definitive recommendation.
Schema.org requirements for ecommerce GEO
Schema.org structured data is the native language of AI engines. While basic SEO might get by with a simple Product schema, GEO requires an aggressive, comprehensive schema architecture.
To achieve high AI visibility, your PDPs must dynamically inject nested JSON-LD containing:
- Product: Complete with
brand,color,material, andaudience. - Offer: Strict
price,priceCurrency, and critically,availability(InStock vs OutOfStock). AI engines actively suppress out-of-stock items. - AggregateRating: A synthesis of your reviews, allowing the AI to answer "is this highly rated?"
- Review: Individual reviews nested within the schema, allowing the agent to extract specific sentiment regarding sizing or comfort.
- MerchantReturnPolicy: A newer schema type that allows AI to confidently answer "can I return this if it doesn't fit?"
Product feed quality for AI search
Beyond on-page schema, AI shopping engines (like Google's SGE and Shopping Graph) ingest data directly from merchant feeds (e.g., Google Merchant Center). For traditional Google Shopping, a basic feed with an ID, Title, Link, and Price was often enough. For AI search, a sparse feed guarantees invisibility.
AI engines cross-reference attributes. Your feed must have a fill rate approaching 100% for optional fields. If you sell apparel, fields like age_group, gender, pattern, material, and size_type must be explicitly defined. If the AI is looking for "men's waterproof jackets" and your feed lacks the gender attribute, the engine will discard your product rather than guess.
The content structure AI agents prefer
LLMs are trained to extract direct answers. If your PDP forces the agent to parse 500 words of atmospheric brand storytelling before mentioning the technical specifications, the extraction process fails.
The optimal content structure for GEO is highly declarative. Use clear, entity-based naming conventions. Implement Q&A formatting directly in the DOM (e.g., an FAQ accordion) because it perfectly mirrors the prompt/response structure of conversational AI. If a user prompts ChatGPT with "Are Jhuns boots true to size?", and your page contains the exact string <h3>Are these boots true to size?</h3><p>Yes, they fit true to size.</p>, the semantic match is mathematically perfect.
The 8-Point GEO Self-Audit
Run your primary PDP through this yes/no checklist to determine your baseline GEO readiness:
- JSON-LD Format: Is your schema implemented as JSON-LD in the
<head>rather than deprecated Microdata scattered in the HTML body? (Yes/No) - Zero Schema Errors: Does the Google Rich Results Test show zero errors and zero critical warnings for your Product schema? (Yes/No)
- Nested Reviews: Are individual text reviews exposed in the raw HTML DOM so bots can crawl them, rather than locked in a JavaScript client-side render? (Yes/No)
- Explicit Attributes: Are technical specifications (material, dimensions, weight) formatted as HTML
<table>or<ul>lists rather than embedded in a paragraph? (Yes/No) - Variant URLs: Does selecting a new color/size update the URL and dynamically alter the schema payload to reflect the new variant's ID and availability? (Yes/No)
- Return Policy Schema: Have you implemented the
MerchantReturnPolicyschema type? (Yes/No) - Feed Completeness: Does your Merchant Center feed include all highly relevant optional attributes for your category? (Yes/No)
- Q&A Structure: Does the page contain explicit question-and-answer pairs regarding sizing, shipping, or compatibility? (Yes/No)
If you answered "No" to more than two of these, your products are likely being excluded from LLM-driven shopping recommendations.
Generative Engine Optimisation requires data engineering, not copywriting. We restructure commerce architectures to ensure perfect semantic visibility across ChatGPT, Gemini, and the next generation of AI search. Book a free 30-minute consultation or explore our AI Search & GEO Engineering service to future-proof your visibility.