AI Shopping Agents Compared: Who Recommends What and Why

Agentic commerce is no longer theoretical. In 2026, at least five major platforms have live AI shopping agents that recommend, and in some cases purchase, products on behalf of consumers. For retailers, the challenge is how to appear in all of them at once.

Each agent draws on different data sources, applies different ranking logic, and supports different checkout mechanisms. But the requirements they place on merchants are strikingly similar. This article compares the major AI shopping agents side by side and sets out why an agent-agnostic approach to agentic commerce is the only strategy that scales.

The landscape: who is live and who is building

By mid-2026, the agentic commerce landscape has taken shape around five principal players: ChatGPT (Operator and Instant Checkout), Google (AI Overviews and the Shopping Graph), Perplexity (Perplexity Buy), Amazon (Rufus) and Apple (Intelligence shopping features). Each represents a different philosophy, but they all compete for the same consumer moment: the point at which someone decides to buy.

ChatGPT Shopping: Operator and Instant Checkout

OpenAI's approach to agentic commerce is the most ambitious. ChatGPT Shopping combines Operator (which browses merchant websites on the user's behalf) with Instant Checkout, powered by Stripe's Agentic Commerce Protocol (ACP), to complete purchases without ever leaving the chat.

ChatGPT draws on its own web index, merchant product feeds, and structured data from participating retailers. Products with ACP-compatible checkout receive a significant advantage because the agent can close the sale immediately and is therefore more likely to recommend them.

Merchant requirements include structured product data (title, description, price, availability, images, attributes), Stripe integration for Instant Checkout, and clean, crawlable product pages. Retailers without ACP can still appear, but they lose the transactability advantage that increasingly determines what agents surface first.

Google Shopping AI: Overviews and the Shopping Graph

Google has embedded AI shopping directly into Search through AI Overviews, the generative summaries that now appear above traditional results for commercial queries. Behind them sits the Shopping Graph, Google's structured database of products, prices, reviews and availability, ingesting data from Merchant Centre feeds, schema.org markup, Business Profiles and crawled web content.

For retailers, the requirements are familiar but newly critical: complete Merchant Centre feeds with accurate pricing and stock, schema.org markup on every product page, and genuine review signals. Products missing from the Shopping Graph are missing from AI Overviews, and as those summaries absorb more commercial search traffic, the gap becomes significant. Our guide on moving from search keywords to AI recommendations covers this transition.

Perplexity Buy

Perplexity has positioned itself as the answer engine, and Perplexity Buy extends that positioning into commerce. When a user asks a product question, Perplexity synthesises information from across the web, compares options, and presents a curated recommendation with a one-click purchase option powered by its own checkout integration.

What makes Perplexity distinctive in the agentic commerce space is its emphasis on editorial-style reasoning. The agent explains its choices, citing sources and weighing trade-offs like a buying guide. Third-party corroboration, appearances in independent reviews and comparison articles, carries significant weight.

Merchants benefit from well-structured product data, but they benefit equally from credible third-party mentions. Perplexity Buy treats the open web as an evidence base, and products with strong independent endorsements rise to the top.

Amazon Rufus

Amazon Rufus operates within a closed ecosystem, but its influence on the broader AI shopping landscape is substantial. Trained on Amazon's product catalogue, customer reviews, Q&A threads and category information, Rufus helps shoppers navigate Amazon's vast inventory through conversational queries.

Rufus matters to the wider market because it is normalising conversational shopping, consumers who learn to ask Rufus for advice will expect the same everywhere. It also demonstrates that structured product data matters even within a closed marketplace: listings with complete attributes and strong reviews consistently outperform thin ones.

For retailers selling on Amazon, optimising for Rufus means the same things that optimise for external agents: complete attributes, specific descriptions, and genuine trust signals. The principles of agentic commerce are universal, even when the platform is closed.

Apple Intelligence shopping

Apple's entry into AI-driven shopping is the most privacy-centric. Apple Intelligence features, integrated into Siri and the Wallet ecosystem, prioritise on-device processing and user consent. Rather than building a centralised product graph, Apple aggregates information at query time from Safari context, App Store commerce data, and select retail partnerships.

For merchants, this means Apple Pay integration, App Clips for quick purchases, and structured data that Apple's crawlers can parse. The audience is smaller but high-value: Apple users have consistently higher average order values and stronger purchase intent.

What they all have in common

Despite their differences, every major AI shopping agent requires the same three things from merchants:

• Structured data. Complete, consistent, machine-readable product information. Not marketing copy, but data. Attributes, specifications, pricing, availability, all in formats agents can parse and reason over.
• Trust signals. Verified seller status, genuine reviews, clear returns policies, accurate stock information. Agents are designed to protect consumers, and they filter out merchants who present risk.
• Transactable endpoints. A way for the agent to complete a purchase, whether through ACP, native checkout, Apple Pay, or marketplace integration. Products the agent can buy outperform products the agent can only link to. For how a purpose-built channel compares to selling on Amazon or eBay, see Vendoora vs traditional marketplaces.

These are not optional extras. They are the minimum conditions for visibility in the agentic commerce era. As we explain in our piece on how AI agents choose products, missing any one of these signals can remove a product from consideration entirely.

Why you cannot optimise for one agent at a time

The temptation is to pick the biggest agent (likely ChatGPT or Google) and optimise specifically for it. This is the same mistake businesses made in early SEO when they optimised for Google at the expense of everything else.

The agentic commerce landscape is fragmented and will remain so. Consumers will use whichever agent is native to their device or preferred assistant. A retailer visible to ChatGPT but invisible to Perplexity is losing sales they will never know about, because the agent simply recommended a competitor instead.

The practical solution is to focus on universal requirements (structured data, trust signals, transactability) rather than platform-specific optimisations. A product with well-structured data and a working agent-compatible checkout performs across all current agents and any future ones. Platform-specific tactics are a bonus, not a foundation.

How Vendoora works across all agents

This is the problem Vendoora was built to solve. Rather than maintaining separate feeds and integrations for each agentic commerce platform, retailers create a single structured listing that works everywhere.

When a retailer lists on Vendoora, we generate structured data in every format the major agents require: schema.org markup, Merchant Centre-compatible feeds, ACP-ready checkout paths, and machine-readable product records. We handle trust (verified seller status, review aggregation, returns policy standardisation) and provide the transactable endpoint through our integrated checkout.

The result: a product listed on Vendoora is visible and purchasable across ChatGPT, Google, Perplexity and every other agent that consumes structured commerce data. One listing, universal agentic commerce coverage. No platform fee unless you sell, 7.5% of net sales, dropping to 5% above 100,000 pounds per month.

The agentic commerce revolution is not coming. It is here. The only question for retailers is whether their products are structured, trusted and transactable enough to be in the conversation. Apply as a Vendoora vendor and let us handle the rest.

Read next