A shopper agent is a conversational artificial intelligence (AI) agent that a retailer or brand deploys on its own website, app or messaging channels to take a customer from a vague request to a completed order. It answers questions about products, compares options, builds the basket, applies offers, takes payment and, after the sale, handles returns or delivery queries. It exists because retailers fear losing the customer relationship to outside assistants such as ChatGPT or Google's Gemini, and a shopper agent is their attempt to offer the same experience under their own name, with their own data and their own checkout.

The phrase has a second, looser use. Analysts and vendors sometimes say "AI shopping agent" or "shopper agent" for any agent that shops, including assistants a consumer brings from outside. Salesforce, whose product carries the name, draws the line clearly: shopper agents belong to brands, while personal AI agents belong to consumers. That split is now visible in survey data, and it is the one this article follows.

How a shopper agent works

The mechanics fall into four layers.

Conversation. A large language model (LLM) interprets the request ("a waterproof jacket for a toddler, under $60, delivered before Friday"), asks follow-up questions and keeps the session's context.

Grounding. The agent answers from the retailer's own systems rather than from the open web: the product catalogue, live inventory, pricing and promotions, customer reviews, order history and loyalty status. This grounding is what separates a retailer's agent from a general chatbot, and it is why vendors sell shopper agents as part of commerce platforms rather than as standalone software. Salesforce's version, according to its July 6, 2026 release, "carries the customer conversation from discovery through checkout and into service, speaking in your voice on your storefront."

Action. The agent calls tools exposed by the commerce platform: add to cart, apply a coupon, check stock at a nearby store, start a return. Many vendors expose these tools through the Model Context Protocol (MCP), the open standard Anthropic published in November 2024 for connecting models to software.

Checkout. Payment runs through the retailer's existing processor, with the agent filling in shipping and payment details the customer has stored. Because the merchant remains the seller of record, fraud screening, tax and returns policies are unchanged.

The same agent can also work away from the retailer's site. Through syndication, a retailer can expose its catalogue and checkout to outside assistants using protocols such as Google's Universal Commerce Protocol (UCP), announced in January 2026, or the Agentic Commerce Protocol (ACP) maintained by OpenAI and Stripe. Salesforce says native integrations with ChatGPT, Google Search including AI Mode, and the Gemini app are "arriving in the coming months", according to its July release. Walmart has taken the furthest step: from the week of March 25, 2026, its Sparky agent began running inside ChatGPT, with baskets synchronised across Walmart's website, app and the chatbot, a move PPC Land covered after Walmart's ChatGPT checkout underperformed.

Retailers' e-commerce, merchandising and service teams operate the agent, usually on a commerce platform such as Salesforce or Shopify. The counterpart is the consumer, who increasingly arrives through another agent.

From shopbots to Rufus

Software that shopped on a consumer's behalf predates LLMs by three decades. In 1995 Andersen Consulting's BargainFinder compared compact-disc prices across online stores, and several retailers blocked it. Researchers at the University of Washington described ShopBot, a "scalable comparison-shopping agent", in a 1997 paper, and the technology fed the price-comparison sites of the early 2000s. These shopbots worked for the buyer and against the merchant's margin.

The retailer-run version came later. Chat widgets and scripted bots on retail sites in the 2010s followed decision trees and broke on anything unexpected. Generative AI changed the economics. Amazon launched Rufus in beta in February 2024, and by November 18, 2025 it said more than 250 million customers had used the assistant that year, with monthly users up 149% year on year. That day Amazon added memory, price tracking and an auto-buy function that completes purchases once a target price is reached, with a 24-hour cancellation window.

Google moved in parallel. On November 13, 2025 it introduced agentic checkout and an AI feature that calls local stores, with Wayfair, Chewy and Quince among the first merchants. OpenAI launched Instant Checkout with Stripe on September 29, 2025, then shut it in March 2026. Walmart's data showed that purchases made inside ChatGPT converted three times worse than click-outs to Walmart's site, according to Daniel Danker, Walmart's executive vice president for product and design.

That result gave retailers an argument for owning the agent themselves. Salesforce made Shopper Agent generally available on July 6, 2026, alongside a business-to-business Buyer Agent and a back-office Merchant Agent. "The brands that win will have their Shopper Agent live on their own properties for the 2026 shopping season," said Nitin Mangtani, executive vice president and general manager of Agentforce Commerce, in the release.

Why it matters for marketers

A shopper agent sits on the path where search ads, retail media and onsite merchandising used to work. When a customer asks an agent rather than browsing a category page, the sponsored product tile and the banner slot lose some of their audience. The agent's ranking logic becomes the new shelf, and whoever writes that logic decides how much of it can be bought.

The traffic numbers explain the urgency. Adobe, which measures more than 1 trillion visits to US retail sites, reported on October 29, 2026 that AI-referred traffic grew 71% year on year in September and converted 90% better than other traffic, against 9% worse in August 2025. It forecast AI traffic increases of 159% on Thanksgiving, 95% on Black Friday and 88% on Cyber Monday. Those visits arrive from outside assistants; a retailer with its own agent can at least keep the conversation once they land.

Product data has become a media input. Google's Heiko Hotz has pitched the Shopping Graph's 60 billion listings as grounding data for agents, while OpenAI added product feed support to ChatGPT Ads Manager after abandoning checkout. Agents read attributes, not imagery, so the completeness of a feed affects whether a product is surfaced at all.

There is also a retail media angle. PPC Land's analysis of how Amazon's $69 billion ad business hangs on a browser lawsuit shows the risk for marketplaces when an outside agent shops without seeing sponsored listings. A retailer's own shopper agent can carry sponsored recommendations; an outside personal agent may ignore them. According to BCG's October 2026 report on AI shopping agents, 65% of surveyed brands expect retail media spend to grow as agents spread.

Limitations and disputes

Contested performance claims. Most figures come from vendors. Salesforce's July release said retailers running shopper agents grew sales 59% faster than those without; a Salesforce survey reported on November 12, 2026 said brands with shopper agents see seven times faster sales growth. Neither publication sets out a method, and the two figures cannot both describe the same comparison. Walmart says Sparky users spend about 35% more per order, which may reflect who chooses to use it as much as what the agent does.

Consumer reluctance. The same Salesforce survey of more than 2,500 shoppers in the US, UK, Canada and Australia found 76% not interested in an AI agent buying on their behalf. BCG put the share of consumers willing to let an agent complete a purchase alone at 4%. Shopper agents that advise are accepted; agents that transact are not yet trusted.

Accuracy. Agents grounded in a catalogue can still misstate sizes, compatibility or delivery dates. Independent reviews of Rufus by Marketplace Pulse described it as "confident and wrong" on product questions.

Neutrality. An agent owned by a retailer recommends from that retailer's range and can be paid to favour certain brands. No regulator had set a disclosure standard for sponsored placement inside an AI agent's answer as of November 2026, although Brussels has begun targeting paid AI recommendations.

Fragmentation. Retailers face UCP, ACP and proprietary integrations at once. Each outside assistant adds a channel to maintain, and the March 2026 reversal by OpenAI shows a channel can disappear within months.

Personal agent. Software acting for one consumer across many merchants, with delegated authority over payments, such as Meta's Muse or Perplexity's Comet. A shopper agent acts for the merchant; a personal agent acts for the buyer, and the two increasingly negotiate with each other.

Buyer agent. In advertising, an agent that buys media for an advertiser under protocols such as the Ad Context Protocol (AdCP). Salesforce also uses "Buyer Agent" for a business-to-business purchasing assistant, a different product.

Shopbot. The 1990s term for price-comparison software working for the consumer, without conversation or checkout.

Conversational commerce. The broader practice of selling through chat and messaging, whether handled by people, scripts or agents.

Recent developments

The 2026 holiday season is the first test at scale. Salesforce projects AI will influence $75.8 billion of Cyber Week spending, up 13% from 2025, a figure published without a method, and Meta put a buying agent inside WhatsApp in September for US users. Retailers are building shopper agents to keep customers on their own properties while personal agents arrive from outside asking to transact. Some businesses are refusing them outright: United Airlines has barred AI agents from bookings. Whether the merchant's agent or the buyer's becomes the default interface for online shopping remains unsettled.

Timeline

  • 1995 - Andersen Consulting releases BargainFinder, comparing compact-disc prices across online stores
  • 1997 - University of Washington researchers publish ShopBot, a comparison-shopping agent
  • February 2024 - Amazon launches Rufus in beta in the United States
  • November 2024 - Anthropic publishes the Model Context Protocol
  • September 29, 2025 - OpenAI and Stripe release the Agentic Commerce Protocol and Instant Checkout in ChatGPT
  • November 13, 2025 - Google introduces agentic checkout and store-calling in AI Mode and Gemini
  • November 18, 2025 - Amazon adds memory, price tracking and auto-buy to Rufus, reporting 250 million users that year
  • January 2026 - Google announces the Universal Commerce Protocol
  • March 2026 - OpenAI shuts Instant Checkout after Walmart reports in-chat conversion three times lower than click-outs
  • Week of March 25, 2026 - Walmart's Sparky begins running inside ChatGPT
  • July 6, 2026 - Salesforce makes Shopper Agent generally available in Agentforce Commerce
  • September 8, 2026 - Meta launches the Muse agent, including WhatsApp purchasing, in the US
  • October 29, 2026 - Adobe forecasts AI retail traffic up 159% on Thanksgiving 2026
  • November 12, 2026 - Salesforce survey finds 76% of shoppers do not want AI agents buying for them

Summary

Who. Retailers and brands that deploy shopper agents, including Amazon (Rufus) and Walmart (Sparky); platform vendors such as Salesforce, whose Shopper Agent gave the term its product name; outside assistants from OpenAI, Google and Meta that compete for the same customer; and consumers, most of whom do not yet want agents to buy for them.

What. A conversational AI agent run by a merchant on its own properties, grounded in its catalogue, inventory and customer data, that guides a shopper from discovery to checkout and service. It can also be syndicated to outside assistants through UCP or ACP. In looser usage the term covers any AI agent that shops.

When. Comparison shopbots date from 1995. Retailer-run generative AI assistants arrived with Rufus in February 2024, gained the ability to transact in late 2025, and became a packaged product with Salesforce's general availability on July 6, 2026.

Where. On retailer websites, apps and messaging channels, and increasingly inside ChatGPT, Gemini and Google's AI Mode through protocol integrations, with most deployments in the United States as of November 2026.

Why. Retailers want to keep the customer relationship, first-party data and sponsored placements as shopping shifts to conversational interfaces. For marketers the agent's ranking logic becomes a new shelf, product feeds become media inputs, and the neutrality of recommendations becomes a regulatory question.