Google Merchant Center for AI Search: Get Into AI Shopping Answers

When someone asks Google's AI Mode or Gemini "what's the best insulated water bottle under $50", the products in that answer come from Merchant Center feeds. Same for shopping questions inside AI Overviews. The feed you set up once for Shopping ads is now the data source for a whole new surface, and the difference between appearing in AI answers and being invisible is mostly feed quality.

That is genuinely good news for small retailers: feeds are fixable in an afternoon, very few stores are optimising theirs for AI discovery yet, and Google has just started handing out the measurement tools. This guide covers the three layers: the new AI performance insights, the feed fields AI actually reads, and the agentic-commerce groundwork worth laying now. It extends the AI SEO playbook to ecommerce's most under-optimised surface.

Google's New AI Performance Insights in Merchant Center

Google is rolling out AI performance insights inside Merchant Center, its documentation covering the US, Canada, Australia, India and New Zealand. It reports how your products perform on conversational AI surfaces: AI Mode, AI Overviews in Search, and the Gemini app.

Three reads matter most. SHARE OF VOICE benchmarks your brand's visibility on AI surfaces against competitors, the shopping equivalent of citation share. FUNNEL STAGES show where products appear across discovery, evaluation and purchase questions. And PRODUCT TERM AND ATTRIBUTE INSIGHTS show the search terms shoppers use on AI surfaces and the attribute gaps holding your products out of those answers. Treat it like the Bing AI Performance report we cover in the AI visibility measurement guide: the free scoreboard you check monthly, and the direct to-do list for your feed.

The Feed Fields AI Actually Reads

AI shopping answers are assembled from structured product data, so the fields that decide Shopping performance decide AI performance, with a conversational twist.

• Titles still carry the match. Front-load the product type and the attributes shoppers say out loud: brand, category, colour, size, key spec, inside the 150-character limit with the first ~70 doing the work. A title that reads like the answer to "black women's running shoes size 10" gets recommended for it.
• product_type is the sleeper field. Your own taxonomy, written in customer language and taken deep (several levels, not one). AI surfaces use it to understand what the product IS in the terms buyers use.
• Attribute completeness is now visibility. Colour, size, gender, material, dimensions, compatibility: whatever drives purchase decisions in your vertical needs to be filled, because conversational queries filter on constraints ("fits a 15 inch laptop", "safe for sensitive skin") and products missing the attribute miss the answer.
• Descriptions go conversational. Benefit-led sentences that answer the questions buyers actually ask beat spec sheets, because AI lifts language that matches conversational phrasing. Keep them factual and policy-clean.
• Trust data seals the recommendation. High-quality images, accurate shipping and returns information, and ratings give AI the confidence signals it needs to recommend you over an equivalent product with gaps.

The free Merchant Center Optimizer skill runs this entire pass on your feed: disapproval triage, query-matching title rewrites, attribute completeness, and the never-touch rules that protect your product history.

Agentic Commerce Is Not Coming, It Is Here

Google's Universal Commerce Protocol (UCP) launched in January 2026 and, per Search Engine Journal's technical coverage, is already deployed on Etsy and Wayfair. UCP lets AI agents handle the full commerce lifecycle: product discovery, carts, identity, checkout, and order management. In plain terms: an AI assistant finding, comparing, and buying the product on the shopper's behalf.

The groundwork is unglamorous and worth doing now:

• Complete your Merchant Center policies: returns, shipping, and customer support details filled in properly; these gate eligibility for agent-driven transactions.
• Make your Product schema comprehensive and identical to the feed and the visible page: name, SKU, GTIN, brand, offers, aggregateRating, shipping details. Agents validate across all three before acting.
• Add the conversational layer to product pages: real Q&A, compatibility and substitution information, FAQPage markup. Agents answer buyer questions from whatever you have published.
• Check the page itself is agent-readable: our free Agentic Product Page Auditor scores whether an AI shopping agent can actually read and buy from your product page.

The pattern from every earlier platform shift holds: the boring data-completeness work done early compounds, and retrofitting it after agents are mainstream costs ten times more.

The 30-Minute Monthly Routine

This does not need a retainer. Once the feed is fixed, the maintenance loop is small:

• Check the AI performance insights (share of voice, attribute gaps) alongside your regular AI visibility scoreboard.
• Feed the product-term insights back into titles and highlights: shoppers on AI surfaces phrase things differently, and the report tells you how.
• Glance at diagnostics for disapprovals on top sellers (daily, ten seconds) and re-run the Merchant Center Optimizer monthly on your best sellers.