E-commerce SEO & GEO - How Stores Win Visibility in Google and AI Shopping

E-commerce SEO no longer ends at product-page rankings: in 2026, products are also discovered in ChatGPT's shopping carousels, in Google AI Mode, and by AI agents doing the research on the buyer's behalf. The scale has stopped being a curiosity: retail spend flowing through AI platforms is projected at ~$20.9B in the US alone this year (an eMarketer forecast, nearly 4× the year before), Shopify reports an 11× increase in AI-attributed orders between January and autumn 2025, and Adobe measured AI-driven traffic to retail sites growing by thousands of percent. The single most important fact for a store owner, though, is this: ~83% of ChatGPT's shopping carousel data comes from Google Shopping - so a well-optimized product feed has become the ticket to both Google and AI at once. (Selling to companies rather than consumers? The B2B funnel plays by different rules - I laid it out in the post on SEO and GEO for SaaS and B2B. And for the transaction layer - the ACP/AP2/UCP protocols and the fate of "Buy it in ChatGPT" - see the agentic commerce post.)

AI-driven shopping is already ~$21B in annual spend (4× YoY), AI traffic to stores grew by thousands of percent, and 83% of ChatGPT's shopping carousel data comes from Google Shopping. The complete guide: Product schema, the Merchant Center feed, AI crawler access, the ACP/UCP protocols and a store rollout plan.

This guide assembles the whole system: how AI picks products, the three foundations of visibility (Product schema, the feed, crawler access), what the agentic-commerce protocols (ACP/UCP) and the collapse of Instant Checkout actually mean - and a step-by-step rollout plan.

The new purchase funnel - data, not forecasts

Shoppers aren't "trying out" AI purchases anymore - they're moving their research there: comparisons, variant selection, "what should I buy for X" questions. Traffic arriving from models often converts markedly better than classic organic (per Semrush, a visitor from AI search is worth ~4.4× more on average than one from traditional results), because the user arrives after the decision, not before it. Forecasts point to ~50% of shoppers using AI agents by 2030 - but today's scale is already enough to treat AI visibility as a sales channel, not an experiment.

How AI picks products - the Shopping Graph and RAG

Behind most AI shopping answers sits the same mechanics you know from the GEO guide: retrieval + synthesis. What's specific to e-commerce is the data sources:

• The Google Shopping Graph - a graph of billions of offers fed by Merchant Center feeds and page crawls. It powers not only Google (AI Mode, AI Overviews, the Shopping tab) - other agents draw on it indirectly too. If 83% of ChatGPT's carousel data comes from Google Shopping, your feed works across multiple engines at once.
• On-page structured data - the citation stats are unambiguous: pages with valid schema are cited ~3.1× more often in AI Overviews; 71% of pages cited by ChatGPT and 65% cited by AI Mode carry structured data.
• Content around the product - guides, comparisons and reviews from which models build "what to choose" recommendations.

The architectural takeaway: product visibility in AI is a three-layer stack - schema on the page, the feed in Merchant Center, and crawler accessibility. Every layer must agree with the others.

Foundation 1: Product schema (JSON-LD) - the language your product page speaks

The minimum field set for a 2026 product page: name, image (several photos), description, sku and gtin (the key identifier linking your offer to the product graph), brand, offers with price, currency and availability, shipping costs and times (shippingDetails), the returns policy, plus aggregateRating and review wherever you collect opinions.

Three implementation rules:

1. 1.Render server-side. Some AI crawlers don't execute JavaScript - client-injected schema can be invisible to them. The same goes for product-page content.
2. 2.Schema must say the same thing as the page and the feed. A price/availability mismatch between layers is the fastest way to lose the trust of shopping systems.
3. 3.Not just Product: FAQPage on product pages (size, compatibility questions), BreadcrumbList, Organization with sameAs. The full toolkit is in my Schema.org guide.

Foundation 2: the Merchant Center feed - the new product SEO

The feed has stopped being an "Ads add-on" - it's the primary source of your product data for the AI ecosystem. The 2026 standard:

• 95%+ attribute completeness - every empty column (size, color, material, gender, age_group) is a buyer's question your product won't answer in filters or in an agent's reply.
• Titles 30+ characters built for real queries (brand + type + attributes), descriptions 500+ characters with genuine specs, GTIN always populated, at least 3-4 images.
• Freshness and consistency: price and availability in the feed must match the page to the hour - automated updates, not manual exports.
• Product ratings connected to the feed if you collect them.

This isn't theory: the rollback of ChatGPT's Instant Checkout (below) showed that in-chat purchasing technology guarantees nothing on its own - what survived is product discovery, and discovery rests entirely on data quality. An agent that recommends a product with a stale price or no stock loses the buyer's trust on the first click. Agentic systems will be merciless to stores with messy data.

Foundation 3: AI crawler access and performance

Before any model can cite your store, it has to be able to read it:

• Audit robots.txt: don't block OAI-SearchBot (it powers ChatGPT search and shopping), PerplexityBot or Googlebot. Make the GPTBot (model training) decision consciously - but remember that blocking search bots means vanishing from answers.
• Rendering: the core product content (name, price, description, availability) must be available without JavaScript.
• Performance: slow product pages get crawled less and convert worse - the fundamentals are in my Core Web Vitals guide, and the architecture-indexing link in crawl budget.

Agentic commerce protocols: ACP, UCP and the Instant Checkout lesson

2025/26 brought a race for the "purchase via agent" standard:

In short: ACP (OpenAI and Stripe) and UCP (Google, Shopify and a coalition of retailers and payment companies) are open standards through which an agent reads an offer and places an order. In March 2026 OpenAI scaled back Instant Checkout and shifted to product discovery, while Google is building in-conversation purchases in AI Mode and Gemini. The full timeline, a comparison of ACP, AP2 and UCP and a "join now or wait" matrix are in the post on agentic commerce.

The practical takeaway for a mid-sized store: you don't need to implement any protocol today. You need to be agent-ready: clean data (schema + feed), current inventory and a frictionless checkout. Protocols will mature; stores with messy data won't benefit from any of them.

Content above the product page - where recommendations are decided

A product page answers "where to buy X". But models mostly answer the questions before that: "what to buy", "what to choose for…", "X or Y". That's decided by content:

• Buying guides ("how to choose an espresso machine under $500") - they target exactly the questions people ask agents.
• Comparisons and rankings - your own, plus presence in other people's roundups; the mechanics are in my digital PR guide.
• Topic clusters around categories - the category as the pillar, guides as spokes; the architecture from the topical authority guide maps onto e-commerce one to one.
• UGC and substantive reviews - models read opinions, not just star counts.

How to measure a store's AI visibility

• AI referral traffic in analytics (chatgpt.com, perplexity.ai, gemini) - a separate segment; watch conversion, not just sessions.
• AI Share of Voice on shopping questions: a fixed set of 15-20 questions ("which X for Y do you recommend?") asked every 2-4 weeks across 3 models - methodology in my Share of Voice guide.
• Presence in shopping carousels (ChatGPT/Google) for key categories.
• Merchant Center: feed quality reports, disapproved offers, attribute coverage.

The step-by-step rollout plan

This plan covers the product data and AI layer. I described the order of all SEO work in a store, with a 3, 6 and 12-month roadmap, in the post on ecommerce SEO strategy.

1. 1.Audit the three layers: schema on product pages (Rich Results Test), feed health in Merchant Center (disapprovals, completeness), robots.txt for AI bots.
2. 2.Fill in GTINs and attributes to 95%+ completeness - start with bestsellers and highest-margin categories.
3. 3.Rewrite feed titles and descriptions for real queries (brand + type + key attributes; 500+ character descriptions with specs).
4. 4.Implement/fix Product schema with SSR - the full field set, consistent with feed and page.
5. 5.Automate price/inventory sync feed ↔ store (hourly, not weekly).
6. 6.Unblock AI crawlers in robots.txt and verify product pages render without JS.
7. 7.Connect product reviews to the feed and schema (aggregateRating/review).
8. 8.Build the content layer: 2-3 buying guides for key categories plus presence in external roundups.
9. 9.Set up measurement: an AI traffic segment in analytics plus a monthly Share of Voice test on shopping questions.
10. 10.Watch the protocols (UCP/ACP) without panic - decide on adoption when your platform (Shopify, PrestaShop, Magento) ships native support.

The most common mistakes

• A feed built "for Ads", not for discovery. Empty attributes and "Product 123" titles exclude your offer from agent answers.
• Data mismatch between layers. One price on the page, another in the feed, a third in schema - shopping systems detect it and downgrade trust.
• Client-side-only schema. Some AI crawlers won't execute JS - your structured data doesn't exist for them.
• Blocking AI bots "just in case". A blocked OAI-SearchBot = non-existence in ChatGPT shopping; decide bot by bot, consciously.
• Product pages only, zero content. Without guides and comparisons the store doesn't exist in "what should I choose" questions - where the decision is made.
• Waiting for AI commerce to "mature". Associations and citations take months to build - stores that start after their competitors will be chasing.

Summary

E-commerce has entered an era where sales are decided by two parallel discovery channels: classic Google and generative answers. The good news is both feed on the same data - a complete feed, valid schema and crawler accessibility work everywhere at once, and the 83% dependence of ChatGPT's carousels on Google Shopping makes Merchant Center the highest-yield SEO investment in a store. The Instant Checkout collapse was a reminder that in the agent era data quality is strategy, and content above the product page - guides, comparisons, roundup presence - decides whose product makes the recommendation.

Strategically: treat the store as an entity with a content cluster (topical authority) and a trust layer (E-E-A-T) - and product data as an API consumed by all of your customers' agents.

If you also sell on Allegro or Amazon, visibility works differently there - I cover it in the post on product visibility on marketplaces in the AI era.

---

I build store visibility in Google and AI answers - from feed and schema audits to content strategy and measurement - as part of my e-commerce and AI optimization (GEO) services. Get in touch - I'll start with a three-layer audit of your product data and a test of whether ChatGPT recommends your products or your competitors'.

If you want me to audit your store and implement these changes (Product data, Merchant Center, filters, categories), see ecommerce SEO.

Related articles

• BaseLinker AI automation for e-commerce
• Schema.org structured data - the complete guide
• Google AI Overviews & AI Mode - optimization
• GEO - Generative Engine Optimization: guide & strategy
• Digital PR and brand mentions
• AI order and fulfillment automation