Six AI shopping assistants compared: ChatGPT to Claude

ChatGPT, Gemini, Perplexity, Alexa for Shopping, Copilot and Claude each take your catalog a different way, finish the order in a different place and publish a different amount about how they rank. Treating them as one "AI channel" is the most expensive shortcut we see brands take.

Six AI shopping assistants now name products to shoppers who never opened your site. They differ on how a brand gets in (a feed, a Merchant Center account, an Amazon listing, an MCP server), on where the order completes, and on what the operator publishes about ranking. Only OpenAI publishes a ranking rule.

Last updated: 13 October 2026.

Why "one AI channel" is the wrong frame

The pattern we see in planning decks: "invest in AI", singular. The team wires one surface, almost always ChatGPT, and files the other five under fast follow. The fast follow rarely comes, and the surfaces left uncovered are the ones growing at the retailer level: Gemini-led, Perplexity-led and Claude-led retailers all rose in ReFiBuy's Q2 AI1000 while ChatGPT-led retailers fell from 844 to 722 (ReFiBuy, 18 August 2026).

Each surface differs on four things you can look up: the entry path (a feed, a Merchant Center account, an Amazon listing, or a tool the agent can call), where checkout happens, what the operator publishes about ranking, and whether a paid slot exists. One operator publishes a ranking rule; five publish none, so every optimisation method for them is inference.

Four of the six start from a product feed, and that structured catalog is the same one your on-site AI shopping assistant reads. The agents that buy on the shopper's behalf, Muse included, are in AI shopping agents: how they buy. This post is the per-surface view.

ChatGPT

Operator: OpenAI. How a brand gets in: a "secure, regularly refreshed feed (CSV or JSON)" under the Agentic Commerce Protocol (OpenAI Developers, 2026); Shopify stores through the Shopify Catalog toggle. Checkout: rendered "in the OpenAI UI", processed "on the merchant's systems" (OpenAI Developers, 2026). Published ranking signals: results "are not ads"; merchants are ranked on "availability, price, quality, and whether they are the maker or primary seller"; "Instant Checkout items are not preferred" (OpenAI Help Center, 2026).

What to do this quarter: submit a complete feed with identifiers, price, live inventory, media and fulfilment, and mark yourself as maker or primary seller where you are; the spec is in the ChatGPT product feed guide. Add a question-and-answer block to each product page. Surface page: ChatGPT for brands.

Gemini and Google AI Mode

Operator: Google. How a brand gets in: a Merchant Center feed carrying the checkout-eligibility attribute (Google, 2026). UCP, launched 11 January 2026, was "co-developed with industry leaders including Shopify, Etsy, Wayfair, Target and Walmart" and "endorsed by more than 20 others across the ecosystem" (Google, 11 January 2026). Checkout: in AI Mode and the Gemini app, with Google Pay "using payment methods and shipping info already saved in Google Wallet" and PayPal to follow; "Retailers remain the seller of record" (Google, 11 January 2026). Published ranking signals: none. Paid: a "Direct Offers" ads pilot in AI Mode (Google, 2026).

What to do this quarter: clean the Merchant Center feed, validate Product, Offer and Review schema on every product page, and add UCP attributes as your platform exposes them. Google publishes what it ingests, not how it ranks, so the feed is the lever. Surface page: Gemini for brands.

Perplexity

Operator: Perplexity AI. How a brand gets in: PayPal's "store sync" puts a merchant catalog into Perplexity results (PayPal Newsroom, November 2025). Checkout: Instant Buy for US users, launched with PayPal ahead of Black Friday 2025; the retailer stays "merchant of record" (PayPal Newsroom, November 2025). Published ranking signals: none. Momentum: Perplexity-led retailers rose from eight to 21 in Q2 2026 (ReFiBuy AI1000, 18 August 2026).

What to do this quarter: Perplexity shows its citations inline, so cited pages matter as much as the feed. Publish honest comparison pages on your own domain, earn specific third-party reviews, and sync the catalog through PayPal if US checkout inside the answer is worth it. Surface page: Perplexity for brands.

Alexa for Shopping, formerly Rufus

Operator: Amazon. Rufus was renamed Alexa for Shopping in 2026 (Retailgentic, 5 August 2026). How a brand gets in: sell on Amazon. Off-Amazon brands reach it only through Buy for Me, where Amazon completes the order on the brand's own site and "brands have the choice if they want to participate" (About Amazon, 2025); the decision is worked through in should your brand opt in. Checkout: Amazon's. Published ranking signals: none. Paid: sponsored prompts, which Amazon says "converted to a sale 48% more often and spent 21% more on average" (Amazon Q2 2026 call, via Retailgentic, 5 August 2026).

What to do this quarter: the inputs you control are the listing: title, bullets, attributes, reviews and the customer Q&A section. If you do not sell on Amazon, it sends no traffic off-platform without Buy for Me. Voice adds a twist, covered in Alexa+ for brands. Surface page: Amazon Rufus for brands.

Microsoft Copilot

Operator: Microsoft. How a brand gets in: a Microsoft Merchant Center account; Shopify stores were enrolled "following an opt-out window" (Microsoft Advertising, 8 January 2026). Checkout: Copilot Checkout, launched 8 January 2026, lets shoppers "complete purchases directly inside Microsoft Copilot" on PayPal, Shopify checkout or a Stripe Shared Payment Token; "You are the merchant of record" (Microsoft Advertising, 8 January 2026). Published ranking signals: none.

What to do this quarter: upload the complete feed to Microsoft Merchant Center as the same artefact you built for Google; Microsoft's own retailer guide tells merchants to implement UCP (Microsoft Advertising guide, via Retailgentic, 1 September 2026). If your range includes bulk packs or work-context products, make those attributes explicit. Surface page: Copilot for brands.

Claude

Operator: Anthropic. How a brand gets in: through tools. Claude reads catalogs exposed over the Model Context Protocol rather than a crawled index, and on 2 September 2026 Anthropic released a commerce blueprint with "reference implementations of a shopping agent and a merchant agent for retail, travel, telecom, and ticketing platforms" (Anthropic, 2 September 2026). Shopify, Visa and Mastercard are among the named partners. Checkout: the blueprint "leaves payment to you, whether that is your existing checkout or an agentic payments provider" (Anthropic, 2 September 2026). Published ranking signals: none. Paid: none; "Claude will remain ad-free" (Anthropic, 4 February 2026).

What to do this quarter: expose your catalog as an MCP server so the agent can read price, stock and attributes directly, and keep product content specific and honest. The build order is in Claude Commerce: how Claude shops and how brands get in. Surface page: Claude for brands.

A seventh arrival, Meta's Muse (launched 8 September 2026, Meta Newsroom), is an agent that buys rather than an assistant that recommends; see Meta AI for brands.

The one thing all six share: the order hides in your analytics

Scot Wingo of ReFiBuy coined the term Dark Agentic Commerce Traffic, DACT, for AI-driven sessions and orders that land in GA4 as Direct. The mechanism differs by surface: an in-assistant checkout fires none of your pixels; an agent that pays with a single-use card leaves a guest order with no marker; a shopper who hears a brand name and types it into Google lands in branded search.

For MyMuse, GA4 reported ₹11.5K/mo of AI revenue against ₹81.2K/mo actual over a 12-month attribution window, 7× hidden, and the Year 1 net surfaced was +₹29.34L (Tru Commerce case study). The method is in dark agentic commerce traffic and server-side AI attribution.

How to sequence coverage

Two defensible orders.

By category fit. Mainstream consumer goods: ChatGPT, then Gemini. Amazon-native: Alexa for Shopping, then ChatGPT. Research-heavy and technical categories: Claude and Perplexity alongside ChatGPT. Work-context products: Copilot earlier than its volume alone would justify.

By measured revenue. Run the attribution fix first, read which surface is already converting, and wire that one properly before the others.

Either way, do the catalog work once. Every one of the six reads size, price, availability and pack count as fields, not prose, and a clean, variant-level catalog is the same asset your agentic checkout depends on when the order completes inside the assistant.

Related: what Anthropic actually shipped for commerce

On 2 September 2026 Anthropic published an Apache 2.0 blueprint for a shopping agent a business embeds in its own app and a merchant agent its staff use; it ships the conversation and the cart, and leaves payments, ranking and authorization to the deployment (Anthropic, 2 September 2026). The explainer is Claude for Commerce explained; how its shopping agent selects products is in How a Claude shopping agent picks products.

Where Tru Commerce fits

We have run this on Gray Porter and a group of US DTC brands: we made the product pages and checkout readable and finishable by an agent, then placed a real Muse order on the store. On measurement, MyMuse is the reference: GA4 showed ₹11.5K/mo of AI revenue, the real figure was ₹81.2K/mo, 7× hidden, and the Year 1 net surfaced was +₹29.34L. To see what each of the six sees on your store, book a demo.