How AI assistants choose which products to recommend

By 7BE ResearchPublished 24 October 2026Updated 25 October 2026

When a shopper asks ChatGPT, Perplexity or Google AI Mode what to buy, the answer usually names a short list of products, each with a reason. No one outside these companies sees exactly how that list is chosen, and each assistant works in its own way. But the inputs are visible, and they are the same kinds of things in every assistant. This guide walks through them, then covers what a store can change.

It starts with the question

Shoppers ask assistants in full sentences. “Best linen sheets” is a broad question. “Linen sheets for a hot sleeper under $200 that don't wrinkle much” carries a budget, a need and a preference. The assistant reads all of it and narrows the field before it names anything.

Two things follow. First, the same store can be recommended for one phrasing and missing from another. Second, the products that win specific questions are the ones whose pages state the facts the question asks about: the price, the material, who it suits. A product page that never says it suits hot sleepers gives the assistant nothing to match.

Answers also vary. Ask the same question twice and the list can change, because these systems write a new answer each time rather than look up a fixed ranking. One answer is a sample, not a verdict.

Product data and feeds

To recommend a product, an assistant needs its facts: name, brand, price, availability, images, ratings. It gets them from product feeds and catalogs as well as from web pages.

• Google AI Mode draws on Google's Shopping Graph, which combines the product data merchants send through Google Merchant Center with what Google finds on the web.
• ChatGPT builds its shopping results from structured product data, such as prices, descriptions and reviews from merchants and data providers, together with what its web search finds.
• Perplexity answers shopping questions with product cards and lists the sources it used under the answer.
• Shopify announced Agentic Storefronts on 10 December 2025: Shopify Catalog sends products from Shopify stores to ChatGPT, Perplexity and Microsoft Copilot. Since 24 March 2026, products on Shopify stores are discoverable in ChatGPT by default.

Being in a feed means the assistant knows the product exists. It does not mean the assistant will recommend it. Every competitor in the category is in the same feeds, and an answer names only a few products.

The web sources it reads

For most shopping questions the assistants also search the web and read what they find. Perplexity lists its sources under every answer, ChatGPT shows them when it searches, and Google AI Mode links to the pages it drew on. The pages that come up are the ones a careful shopper would read too:

• editorial reviews and buying guides;
• forums and community threads where people compare what they bought;
• marketplaces, with their ratings and review counts;
• brand and store pages, including yours.

A product that is praised on the pages an assistant relies on for a category has more chances to be named. If those pages never mention a store, the store depends entirely on its own pages and feeds.

Reviews and ratings

Ratings and reviews are among the facts assistants repeat most. An answer may mention a product's rating and how many people left one, or what reviewers say about its fit or how it holds up. These reach the assistant in three ways: review text on your product pages, ratings in your structured data and feeds, and reviews elsewhere, such as marketplaces and forums.

A product with no reviews the assistant can read gives it one less reason to pick it. Reviews that load only through a script after the page opens may be missed by crawlers that do not run scripts. Review apps that also add rating data to the page's structured data are easier to read.

The reasons it gives

Most shopping answers give a reason for each pick: the material, a spec, the price, a rating, who it suits. The assistant takes those reasons from what it has read. That is useful to a store, because the reasons show what the assistant needed to know.

Say an assistant picks a competitor because its sheets are pre-washed and come with a 30-night trial. The question for you is simple: does your page say the same things about your product, in plain words? If your sheets are pre-washed too but the page never says so, the assistant cannot use it.

What a store can influence

You cannot write the answer. You can change most of what it is built from.

1. Let the assistants read your store. Check that robots.txt does not shut out the crawlers assistants use for search, such as OAI-SearchBot for ChatGPT and PerplexityBot for Perplexity. Google AI Mode is part of Google Search and uses Google's regular crawler.
2. Give clean product data. A product type or category, price, availability, brand, identifiers such as GTINs where you have them, and Product structured data on every product page.
3. Collect reviews and make them readable. Ask buyers after delivery, and use a review app that adds rating data to the page.
4. Write descriptions that state facts. Who the product is for, the specs shoppers compare, the use cases, and answers to common questions.
5. Publish your policies. Clear shipping and returns pages let an assistant answer questions about free returns or delivery to a country.
6. Be present where your category is discussed. Reviews, buying guides, forums and marketplaces. This is earned, not bought, and it takes time.

The AI-readiness checklist covers the first five in detail, with steps for Shopify and WooCommerce. For the fourth, see how to write product descriptions AI can recommend.

What no store controls

Some things sit outside any store's reach: how each assistant weighs its sources, the shopper's own history and location, and the fact that answers change from one run to the next. Be wary of anyone who promises you a place in the answers. The honest approach is to measure: ask the same questions on a schedule, and see whether your changes move the results.

See where you stand

The free 7BE scan builds 10 shopper questions from your catalog, asks ChatGPT, Perplexity and Google AI Mode, and shows the products they recommend, the brands named instead of you and the sites they relied on. It also scores your store on the 8 readiness checks. It reads only public pages and needs no signup.