How AI Shopping Assistants Pick Products and How to Show Up

A growing number of shoppers no longer start with a search box and 10 blue links. They ask an assistant: “What’s a good waterproof hiking boot under $150 for wide feet?” The assistant answers with a short list of products, reasons and prices. If your product isn’t on that list, you never got a chance to compete.

That raises 2 questions every online seller is asking: how do these AI shopping agents decide what to recommend, and what can a small store do to show up? The short answer is that assistants rely on structured, trustworthy product data. The stores that are easiest to understand get recommended. This guide explains what the assistants use and how to prepare your catalog.

What AI shopping agents are

AI shopping agents are assistants that help people research, compare and sometimes buy products in a conversation. The main ones US shoppers use today include:

• ChatGPT shopping: OpenAI shows visual product results, side-by-side comparisons and up-to-date product details in chat. In March 2026, OpenAI shifted its focus to product discovery and said merchants can use their own checkout experiences, after its first in-chat Instant Checkout version didn’t offer the flexibility it wanted.
• Google AI Mode and Gemini: Google pairs Gemini with its Shopping Graph, which Google says holds more than 50 billion product listings, with more than 2 billion refreshed every hour. In January 2026, Google introduced the Universal Commerce Protocol (UCP), and select merchants can now offer checkout in AI Mode and Gemini.
• Amazon Rufus: Amazon’s shopping assistant, which Amazon says is trained on its product catalog, customer reviews, community Q&A and information from across the web.

The broader idea is “agentic commerce”: AI agents that don’t just answer questions but can take actions such as building a cart or completing checkout. For a primer on how agents work, see our AI agents guide.

How assistants decide what to recommend

No assistant publishes its full ranking formula, and you should be skeptical of anyone who claims to know it. But the inputs are well documented, and they fall into 4 groups.

The pattern is simple. An assistant can only recommend what it can understand, and it can only trust what it can verify. A product with a vague title, no size information and a stale price is hard to recommend, even if it’s a great product.

Key takeaway: AI assistants recommend products they can understand and verify. Complete, accurate, structured product data is the closest thing to “AI SEO” that actually works.

Product feeds: the foundation

A product feed is a structured file or API connection that lists every product with its key attributes. It’s the most direct way to tell an assistant what you sell.

Google Merchant Center. Feeds submitted to Merchant Center power Google Shopping and feed into AI Mode. Google’s checkout in AI Mode and Gemini only shows a Buy button on listings with the required native commerce attribute, and it’s currently available to select merchants for products eligible in the US, Canada or Australia. Google recommends providing both structured data and a Merchant Center feed, because some experiences combine them.

OpenAI product feeds. OpenAI’s Agentic Commerce Protocol documentation lets merchants share a structured product feed by API or file upload so ChatGPT can index and display products with current price and availability. The discovery feed requires 9 fields for every item:

1. item_id, a stable ID per item or variant
2. title, up to 150 characters
3. description, up to 5,000 characters
4. url for the product page
5. brand
6. seller_name
7. image_url
8. availability, such as in stock or out of stock
9. price, such as “79.99 USD”

Optional fields include GTIN, condition, category, dimensions, weight, sale price, shipping price, return policy, review count and star rating. OpenAI’s guidance is to update price and availability when they change. Many ecommerce platforms, including Shopify, already pass catalog data to ChatGPT, so check your platform’s settings before building anything custom.

Marketplaces. On Amazon, Rufus draws on your listing content, so a complete title, bullet points, attributes and images matter as much as any keyword. Our guide to Amazon seller fees and the FBA vs FBM comparison cover the rest of the Amazon side.

Structured data on your product pages

Structured data is code, usually JSON-LD using the schema.org vocabulary, that labels the facts on a page so machines can read them. Google’s documentation describes 2 main types of Product markup:

• Product snippets, for pages where people can’t buy directly, such as review or comparison pages.
• Merchant listings, for pages where customers can buy. These support richer details such as apparel sizing, shipping costs and return policy.

At a minimum, mark up product name, image, description, brand, GTIN or SKU, and an offer with price, currency and availability. Add shipping details and a return policy where you can. Google also supports product variant markup so it knows which items are sizes or colors of the same parent.

Check that the structured data matches what’s visible on the page. A price in your markup that differs from the page or feed is a red flag for any system that tries to verify data.

Reviews and the questions shoppers ask

Assistants answer questions like “Is it good for wide feet?” or “Does it run small?” They often find those answers in reviews and Q&A. That makes genuine, detailed reviews one of your strongest signals.

Ways to earn better review content:

• Ask at the right time. Send a review request after the customer has had time to use the product, not the day it ships.
• Prompt for specifics. Ask buyers about fit, use case and durability, not just a star rating.
• Answer questions publicly. Add a Q&A or FAQ section on product pages, written in the words customers use.
• Respond to negative reviews with fixes, not arguments.

Stay on the right side of the law. The FTC’s rule on consumer reviews and testimonials, announced in August 2024, bans fake reviews, including AI-generated reviews attributed to people who don’t exist, buying reviews conditioned on positive or negative sentiment, undisclosed insider reviews and review suppression. The FTC can seek civil penalties against knowing violators. This is not legal advice.

Content that answers shopping questions

Beyond feeds and markup, assistants also read regular web content, especially for research-heavy purchases. Pages that perform well tend to:

• Answer one specific question clearly, such as “How to choose a hiking boot for wide feet.”
• Use clear specs and comparisons, in tables rather than vague marketing copy.
• State facts consistently across your site, feed, marketplaces and social profiles.
• Show real expertise, such as sizing guides, care instructions and honest “who this isn’t for” notes.

If you use AI to draft content, check every claim. Our guide to what AI hallucinations are explains why generated text needs review.

A readiness checklist for AI shopping agents

Use this checklist to audit your store:

Where to start

1. Pick your 20 best-selling products and audit them with the checklist above.
2. Fix titles and attributes first, because they’re used everywhere.
3. Validate your structured data with Google’s Rich Results Test and fix errors.
4. Clean up your Google Merchant Center feed and resolve every disapproval.
5. Check your platform’s ChatGPT integration, or review OpenAI’s product feed spec if you run a custom store.
6. Ask an assistant about your category and note which competitors appear and why.
7. Set up a review request flow that asks specific questions.

For the bigger picture on how AI is reshaping retail operations, see AI in retail and ecommerce and our retail tech trends for 2026.

Conclusion

AI shopping agents are becoming a front door to ecommerce, but they aren’t magic. They pick products they can understand and trust. Clean feeds, accurate structured data, genuine reviews and helpful content won’t guarantee a spot in every answer, but they’re the levers you control, and they help your traditional search and marketplace performance at the same time.

Sources

• OpenAI: Commerce developer docs — Agentic Commerce Protocol, product feeds, agentic checkout and delegated payment
• OpenAI: Product feed spec — required and optional feed fields and update guidance
• Google Merchant Center Help: Universal Commerce Protocol checkout — UCP definition, checkout eligibility and native commerce attribute
• Google Search Central: Product structured data — product snippets, merchant listings, variants and feed recommendation
• Amazon: Meet Rufus — data Rufus is trained on
• FTC: Final rule banning fake reviews and testimonials — banned review practices and penalties