AI shopping assistants: the 2026 map for brands

Six assistants now sell on a brand's behalf inside apps the shopper already uses; a seventh kind runs on the brand's own store. Two families, different rules, one shared catalog.

An AI shopping assistant takes a shopper's question in plain language, picks products from a catalog and moves the shopper toward a purchase. In 2026 it comes in two families: assistants inside platforms shoppers already use, and assistants a brand runs on its own store.

Last updated: 10 October 2026.

Two families, one name

The first family lives where the shopper already is. Ask ChatGPT, Gemini, Muse, Copilot, Alexa or Perplexity what to buy and it names a handful of products from data it holds or pages it reads. You do not set the ranking or see the conversation; on several surfaces the order settles without your storefront loading. Personal agents such as Muse go further and buy for the shopper; see AI shopping agents that buy on a shopper's behalf.

The second family runs on your store. It sees only your catalog, talks only to shoppers who already arrived, and turns a constraint-shaped question into a cart. You control all of it, including whether it finishes the sale in the chat. We separated it from chatbots and site search in the four things sold under one name; the per-surface differences cover the first family.

What they share is the catalog. Neither can answer "cabin trolley bag under 2000 with a TSA lock" unless size, price and lock are structured fields with live availability. That enrichment is paid for once and read by both.

The six platform assistants, compared

Three things stand out. In-assistant checkout is now the norm, and in every documented case the merchant stays merchant of record. Only OpenAI has said what it weighs. Four of the six start from a product feed, so one clean feed serves ChatGPT, Google, Copilot and Perplexity; Shopify stores reach ChatGPT through the Catalog toggle, Google through the Google sales channel, and Meta by default [Retailgentic, 25 August 2026; Shopify Help Center, read 26 September 2026].

What each family can and cannot do for you

Platform assistants reach shoppers who never searched for your brand, and increasingly take the order. What they cannot do is be configured: Meta's launch post describes what Muse can do and nothing about how it weighs one product against another [Meta Newsroom, 8 September 2026]; Google, Microsoft, Amazon and Perplexity are the same. Measurement is thinner too: for Shopify stores, orders on Meta surfaces "are attributed to Meta" and "Google Analytics and custom pixels won't fire in Meta's direct checkout" [Shopify Help Center, read 26 September 2026].

The on-site assistant holds the shopper through constraints and sizing, adds to cart and starts checkout in the chat, and records every question a search box never hears. It cannot reach the shopper who never arrived. We counted four jobs it does; conversion is one. In our deployments, connecting one takes two days on a clean Shopify catalog and two to three weeks on a custom stack; see what the deployment takes.

What the data says about engagement and conversion

Platform-side, the published numbers describe traffic quality. Adobe measured AI-referred traffic to retail sites up 138% year over year in May 2026, converting 54% better than non-AI traffic [Adobe, via Digital Commerce 360, 17 June 2026]. Amazon says more than 350 million customers used Alexa Shopping in 12 months and US users spend more than 40% more per order [Amazon Q2 2026 call, via Retailgentic, 5 August 2026].

On-site, vendors publish lifts and rarely the funnel; we audited every published conversion claim and found denominators rarely stated. Here is one with every step named. On Provogue's Shopify storefront over the 90 days to 24 September 2026, the assistant handled 4,045 shopper messages across 1,332 sessions that sent a message: 732 ran a search (55.0%), 278 viewed a product (20.9%), 97 added to cart (7.3%), 92 started checkout (6.9%) and 6 paid (0.5%). Model spend was ₹171.61 [Tru Commerce, Provogue case study, 2026].

7.3% is an add-to-cart rate of engaged sessions, not a conversion rate; the order rate is 0.5%, and with no holdout none of it is a lift. The biggest drop is between search and product view: merchandising, not the model. Error classes and instrumentation faults are in 90 days of an on-site assistant.

A buyer's checklist

Where Tru Commerce fits

Our AI shopping assistant is the on-site family. On Provogue it handled 4,045 shopper messages across 1,332 engaged sessions in 90 days, with 97 add-to-carts, 92 checkout starts and 6 settled orders on ₹171.61 of model spend; recorded volumes, not a lift. 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. If you want the same run on your store, book a demo.