Skip to content

AI agent

AI

An AI agent is an AI that acts on someone's behalf: it understands a goal, picks the steps and carries them out instead of only answering. In e-commerce, it anticipates the shopper's intent and guides them the way an advisor would. An AI shopping agent is an AI agent that acts for a shopper: it searches, compares and can buy products on their behalf.

How it works

An AI agent chains three actions.

  • It understands the goal, even when it is phrased vaguely: “find me a gift for a friend who loves cooking”.
  • It decides on the steps: which questions to ask, which data to check, which products to compare.
  • It acts with the tools it has access to: catalog, stock, cart, order tracking.

That is what sets it apart from an AI that only generates an answer: an AI agent goes all the way to the result.

An AI shopping agent follows the same logic for the consumer, from an assistant such as ChatGPT, Claude, Gemini or Perplexity. The consumer describes what they need in plain language: "find me a compact stroller that fits in a small car trunk, under $400." The agent searches the web and the catalogs it can access, compares offers and comes back with a reasoned shortlist. Some agents place the order themselves, with the user's approval, at retailers that support it.

To choose, the agent relies on what it can read: product pages, prices, availability, reviews, shipping and return terms. Missing or poorly structured information gets left out.

Why it matters in e-commerce

AI agents are changing how shopping happens. On the brand side, an agent can guide every shopper the way a good sales associate would: ask the right questions, recommend, clear doubts, add to cart. On the buyer side, AI shopping agents search and compare products for them, so part of the search happens away from the brand's site. Shoppers arrive later in their decision, sometimes with their mind already made up, or not at all.

That has two consequences for brands. Their product data has to be complete and machine-readable, or the agent will recommend a better-described competitor. And shoppers who still come to the site expect answers as good as the ones their agent gives them.

Shoppers are already willing to hand over the choice, less so the payment. In the 2026 barometer iAdvize ran with Ifop (in French), 35% of online shoppers would let AI preselect a product for them, as long as they approve the payment themselves. Among regular generative AI users, the figure is 60%. The same barometer shows shopping assistants are mostly used on marketplaces (61% of users) and far less on brand sites (26%). In the barometer, iAdvize CEO Julien Hervouët says brands have a window to build their competitive advantage today.

At the round table iAdvize hosted in Lille (in French), Camille Aâssila, Product & Customer Marketing Director at Nosto, summed up the first step: be present, findable and discoverable in these agents. What a brand can start now:

  • Structure its catalog and fill every product page with the attributes customers ask about.
  • Work on product reviews, content and FAQs, which agents read too.
  • Get the site technically ready so agents can read it.
  • Expect searches on its own site to become more conversational and less keyword-driven.

Further reading: Agentic Commerce Is Here, and What To Do About It

Real-life example

On a home appliance storefront, a shopper asks which washing machine to choose for a family of five with a small laundry room. The AI agent asks how much width is available, compares the catalog models that fit the space, checks stock and suggests two options, explaining how they differ in capacity and energy use.

On the buyer side, a shopper asks her AI assistant for a lightweight rain jacket for a trip to Scotland. The agent compares five brands. The one whose product page lists the waterproof rating, the weight and a size guide comes out on top. The one whose page only says "waterproof jacket, navy" gets dropped, even if the product would have been a good fit.

 

  • 35%

    of online shoppers would let AI preselect a product for them, if they approve the payment themselves

  • 60%

    among regular generative AI users

2026 barometer on AI use in online shopping, run by iAdvize with Ifop (in French). Source: iAdvize x Ifop 2026 barometer.

Common confusion

“AI agent” is often used for any AI tool. An AI agent is defined by its ability to act: it does not just answer, it chains actions to reach a goal.

An AI shopping agent is also often confused with an AI Shopping Assistant, though they sit on opposite sides. The shopping agent belongs to the consumer and can recommend any brand. The AI Shopping Assistant is installed by the brand on its own site, knows its catalog and advises its shoppers.

The iAdvize approach

An AI agent working for sales

The iAdvize AI Shopping Assistant fetches the relevant product information and puts together the best answer to each question, without a fixed script. Connected to the brand's catalog and tools, it recommends, compares and lets shoppers add to cart, and hands over to a human agent when useful. iAdvize also builds the layer that connects a brand's e-commerce stack (its product catalog) to LLMs.

 

Frequently asked questions

What is an AI shopping agent?

It's an AI agent the consumer uses, from an assistant such as ChatGPT, Claude, Gemini or Perplexity. It searches, compares and can buy products on their behalf. It works for the buyer, whereas a brand's AI Shopping Assistant works for the brand's site.

What is the difference between an AI agent and a chatbot?

A chatbot answers questions, often from scripts. An AI agent understands a goal and chains actions to reach it.

Is an AI agent fully autonomous?

Rarely. Most AI agents act within a defined framework: the tools they can access, the rules set by the brand and the moments when a human takes over. An AI shopping agent, for example, only pays at some retailers and with the user's approval. In most cases today, it compares and recommends, and the purchase is completed on the retailer's site.

Is an AI Shopping Assistant an AI agent?

Partly. It finds information and puts together its answers on its own. Some steps, such as handing over to an agent, follow rules set by the brand.

How do I get recommended by an AI shopping agent?

Publish complete, up-to-date product data: specs, dimensions, price, stock, shipping and return terms. An agent only recommends what it can read.

Are shoppers ready to let an agent choose for them?

Some are. In the iAdvize barometer run with Ifop, 35% of online shoppers would let AI preselect a product as long as they keep the final payment approval. Only 21% would go as far as a fully autonomous purchase, mostly for simple purchases.