ObsurfableTwo AI systems now sit between a huge share of shoppers and their purchases: Amazon's Rufus (rebranded Alexa for Shopping in May 2026) and ChatGPT Shopping. Both replace keyword-matching with intent understanding, and both decide - on your behalf - which handful of products a shopper even sees. If you sell anything, understanding their decision logic is no longer optional.
The good news: their mechanics are increasingly documented. This article breaks down how each one chooses what to recommend, where they diverge, and what that means for getting your products surfaced.
How Amazon Rufus / Alexa for Shopping decides
Rufus is Amazon's generative AI shopping assistant, now embedded across the Amazon app and site and folded into the broader Alexa for Shopping system (unifying Rufus's product expertise with Alexa+'s personalization across Echo, the app, and the web). Under the hood it uses a real-time router across multiple models on Amazon Bedrock (including Anthropic's Claude, Amazon Nova, and a custom store-knowledge model) plus retrieval-augmented generation (RAG).
Rather than start from a typed keyword like Amazon's legacy search, Rufus interprets why you're buying. Its recommendations draw on roughly four inputs:
- The product catalog - your product detail pages (PDPs), titles, attributes, A+ content.
- Customer reviews - sentiment and themes across the reviews on a listing.
- Community Q&A - the questions and answers on product pages.
- Information from across the web - Rufus pulls context from reputable editorial sources (e.g. major publications) for product and trend questions.
Two more layers shape the final result:
- The COSMO knowledge graph. Amazon's knowledge graph maps relationships between products and human intentions - the contextual intelligence that lets Rufus turn "a good low-sugar snack for toddlers that doesn't melt" into the right products, rather than matching keywords.
- Personalization ("About You"). Rufus adjusts recommendations per shopper based on purchase history, browsing, and increasingly cross-ecosystem signals (Kindle, Prime Video, Audible). Two shoppers asking the same question get different suggestions.
The upshot: Rufus evaluates which PDPs directly answer the question being asked. The brands winning citations are the ones whose product pages, FAQs, A+ content, and review themes explicitly address the questions shoppers pose.
How ChatGPT Shopping decides
ChatGPT Shopping triggers when your prompt signals shopping intent ("best running shoes under $100"). It typically presents 3-5 products in a carousel with images, prices, and attributes. Product results are organic and unsponsored - ranked purely on relevance to the user (Instant Checkout availability is a minor merchant-ranking tiebreaker, not a paid boost).
ChatGPT retrieves product data from two sources, and the difference matters enormously:
- Public product detail pages, crawled via web search (primarily through
OAI-SearchBot). - Direct merchant product feeds, via the Agentic Commerce Protocol (ACP) - a direct data pipe from a merchant's catalog into ChatGPT (Shopify catalogs are already integrated automatically; Stripe and Salesforce are supported delivery paths).
When deciding what to surface, ChatGPT weighs:
- Semantic relevance to intent - it interprets meaning, not keyword overlap. This is the most fundamental signal.
- Structured metadata - price, description, availability, images. Schema markup (Product, Offer, variants) helps it parse PDPs accurately.
- Price and availability - out-of-stock or "contact for pricing" products are at a disadvantage; budget constraints in the query shift weighting heavily toward price.
- Reviews and authority.
- Context - Memory and custom instructions personalize results.
The single biggest finding from independent tracking: product feeds dominate citations. Merchants with a direct, comprehensive, frequently-updated feed overwhelmingly win top placement over those relying on PDP crawling alone. (ChatGPT Shopping also runs on a specialized shopping-tuned model variant, reportedly more accurate on complex multi-constraint queries than standard search.)
Where the two engines differ
| Amazon Rufus / Alexa for Shopping | ChatGPT Shopping | |
|---|---|---|
| Universe | Amazon's catalog (+ web context) | The open web + merchant feeds |
| Core data | PDPs, reviews, community Q&A, web | PDPs (crawled) + ACP product feeds |
| Intelligence layer | COSMO knowledge graph (intent → product) | Semantic relevance + shopping-tuned model |
| Personalization | Deep ("About You," cross-ecosystem) | Memory / custom instructions |
| Best lever to win | PDP/A+/review/Q&A depth that answers the question | Integrating a rich, structured product feed |
| Paid placement? | Amazon ads exist in parallel | Organic, unsponsored results |
The strategic difference: on Amazon, you optimize the content on your listings; on ChatGPT, the highest-leverage move is often getting your structured feed in front of it (via Shopify/ACP), then optimizing PDPs on top.
What they share (and what to do about it)
Despite different plumbing, both engines reward the same fundamentals - because both are trying to match a shopper's intent to the product that best answers it:
- Answer real questions on your product pages. Both engines evaluate whether your PDP directly addresses the shopper's actual question. Add FAQs and specifics that map to how people ask ("does it melt," "is it good for sensitive skin," "will it fit a 10-person team").
- Use structured, complete, accurate metadata. Price, availability, attributes, variants, images. Rufus reads your catalog; ChatGPT reads your feed and schema. Both punish gaps and ambiguity.
- Cultivate reviews and Q&A. Rufus reads review sentiment and community Q&A directly; ChatGPT weighs reviews and authority. Genuine review depth is a recommendation signal, not just social proof.
- Keep everything current. Stale prices, out-of-stock items, and "contact for pricing" hurt you in both systems.
- Be specific, not promotional. Both interpret meaning. Concrete, honest attributes beat marketing adjectives.
- For ChatGPT specifically: integrate a product feed. It's the single biggest lever for top placement.
For the broader discipline of getting recommended by AI, see our guide to how LLMs decide what to recommend and our answer engine optimization guide.
How Obsurfable helps
The hard part of AI shopping visibility is that you can't see it from your own analytics. A shopper asks Rufus or ChatGPT, sees three or four products, and buys one - and if you weren't in that carousel, you never even register the lost sale.
Obsurfable makes that visible. You define the shopping Prompts your buyers use - "best [product] for [use case]," "[category] under $X" - and run retrieval to see which products get recommended, whether yours appears, and which competitors are winning the carousel. Insights turn that into specific PDP, feed, and content recommendations. Instead of guessing whether your listing "answers the question," you can watch whether the engines actually surface you.
FAQ: Rufus and ChatGPT Shopping
Does Amazon Rufus use keywords like old Amazon search?
No. Rufus interprets shopper intent using RAG and the COSMO knowledge graph, evaluating which product pages directly answer the question rather than matching keywords. Legacy keyword search still runs in parallel.
What data does ChatGPT Shopping use to recommend products?
Two sources: public product detail pages it crawls, and direct merchant product feeds via the Agentic Commerce Protocol. Rich, structured feeds overwhelmingly win top placement.
Are ChatGPT Shopping results paid placements?
No. Results are organic and unsponsored, ranked on relevance. Merchants pay a fee only on completed Instant Checkout purchases, which does not boost ranking.
What's the single best thing I can do to get recommended?
For ChatGPT: integrate a complete, accurate product feed. For Rufus: make your PDPs, A+ content, reviews, and Q&A explicitly answer the questions shoppers actually ask. Both reward accurate, current, specific data.
Do reviews matter for AI shopping recommendations?
Yes. Rufus reads review sentiment and community Q&A directly, and ChatGPT weighs reviews and authority. Genuine review depth is a real recommendation signal.
The bottom line
Rufus and ChatGPT Shopping both replaced keyword matching with intent understanding, but through different machinery: Rufus reasons over Amazon's catalog, reviews, Q&A, and the COSMO graph with deep personalization, while ChatGPT ranks organically from crawled PDPs and - decisively - direct product feeds. The shared truth is that both surface the product that best answers the shopper's real question. Make your product data complete, current, specific, and genuinely responsive to how people ask - and measure whether the engines actually put you in the carousel.