Online shoppers increasingly ask an AI assistant to compare products, narrow options and sometimes help with the purchase. For a store owner, that changes what “optimised” means. A polished product page is no longer the only thing being judged. The structured facts behind it are too. This guide covers how AI shopping surfaces appear to choose products, where the evidence conflicts, and what a store can do now. Sarkar SEO treats ecommerce AI SEO as a data-accuracy discipline first and a copywriting task second.
The shift: from pages to product facts
A Search Engine Land piece frames the change bluntly: AI shopping uses structured product data to decide what shoppers see, which makes the product feed key to discovery. Evidence from a visibility-tracking platform points the same way. In its ChatGPT Shopping tracking, feed-integrated sources accounted for roughly 65% of the product recommendations observed by early September. The report itself notes that the figure comes with boundaries, and there is more than one route into these results. Treat it as one platform’s sample, not a universal rule.
The practical meaning is simple. If your catalogue data is incomplete, inconsistent or outdated, a beautifully written page may never be considered.
Where the sources disagree
Be careful with confident advice on which feed feeds which engine, because the claims conflict.
One agency says ChatGPT pulls from Bing’s index via Bing Merchant Center, making Bing presence non-negotiable. A different guide says research from Profound and Backlinko indicates ChatGPT pulls product card data from Google Merchant Center feeds, not Microsoft’s. Meanwhile, a vendor reports that OpenAI has opened a route where retailers can send their catalogue directly to OpenAI through Ads Manager, in the way they do with Google Merchant Center. your product feed is your llm resume +2
These sources are commercial and none is a primary OpenAI document. The safest approach is to keep a complete, accurate feed in every channel available to you, and to check each platform’s current official documentation before building a strategy around one pipeline.
The fact-match check
The following is my own practical method for turning the research into a routine. It is not an industry standard.
One product, one truth. For your top products, compare these four places:
- The feed submitted to shopping platforms.
- The product page a visitor sees.
- The structured data (Product schema) in the page code.
- Third-party listings, such as marketplaces and retailer sites carrying your goods.
Check price, availability, brand, model, size, colour, shipping and identifiers match in all four. One analysis notes that ChatGPT can surface whichever version of the product facts reaches its retrieval layer. If merchandising, search and development teams manage the same facts in separate systems, mismatches are likely.
Feed attributes that carry weight
You do not need a technical rebuild to start. Review these basics:
- Product identifiers. One feed specialist explains that a missing GTIN triggers warnings in Merchant Center, which in turn leads to poor indexing by AI agents.
- Titles that describe the product. Include brand, type, key attribute and variant, in plain language.
- Descriptions that answer purchase questions. Materials, dimensions, compatibility, care and use.
- Accurate stock and price. Stale data is the fastest way to lose trust.
- Quality images with clear views of the item.
A feed that merely passes platform checks is not necessarily competitive. One agency argues that a feed passing Merchant Center checks is not a feed that performs. It is a commercial claim, but the logic is sound: complete data gives a recommendation engine more to work with.
Product pages still count
Feeds are not the whole story. Searchers and AI systems also read pages, and classic ecommerce SEO still applies. One practitioner notes that transactional queries still go through organic results and product pages, with two more surfaces, feeds and AI answers, appearing alongside them.
For pages, focus on:
- Product schema that is complete and matches the visible page. One study reported that 57.5% of top-ranking ecommerce pages implement schema markup, while 42.5% have none. Google’s own guidance cautions against treating structured data as a shortcut to AI visibility, so use it for accuracy, not as a trick.
- Honest buying guidance. Say who a product suits and who should choose something else. Specific limits build credibility.
- Real reviews. Volume and recency help both shoppers and recommendation systems. Never fabricate or incentivise them against platform rules.
- Answers to common questions such as returns, sizing, warranty and delivery times.
Do not ignore off-site evidence
Perplexity and others draw on more than your store. One source says Perplexity pulls from merchant pages, review sites, editorial comparisons, Reddit discussions and structured data. Independent roundups, expert reviews and community discussions can shape what AI says about your products. Send review units to credible publishers, answer questions where customers ask them and keep your brand details consistent everywhere. useagentshop
The agentic commerce horizon
Shopping assistants are moving toward completing purchases. One source describes Google’s Universal Commerce Protocol, announced at NRF 2026, extending Merchant Center to the cart and checkout inside AI Mode. This is early and optional for most small stores. Keep it on your watch list and prioritise getting your catalogue accurate first.
A realistic starting plan
- Audit your top twenty products with the fact-match check.
- Fix identifiers, titles and missing attributes in the feed.
- Align schema with visible page content.
- Strengthen reviews and buying guidance on key pages.
- Test shopping prompts across several AI tools monthly and record which products and competitors appear.
Expect inconsistent results. AI answers vary between runs and shoppers, so look for trends, not single wins. Pair prompt tests with sales data, since visibility only matters if it produces orders.
