ChatGPT Shopping: How Products Get Recommended

People now shop inside the answer. They ask ChatGPT for the best running shoes under a budget, a gift for a coffee lover, or a quieter vacuum, and get a short list of products with images, prices and links, often without visiting a search results page at all. If your product is not on that list, the sale went somewhere else before your site had a chance.

Quick answer

ChatGPT recommends products based on the question, structured product data from merchants and third-party providers, and what reviews and other sources say. Organic results are not ads. Merchants influence them by keeping product data complete and consistent, submitting their catalog where OpenAI’s merchant program allows it, earning reviews and coverage, and publishing pages that answer buying questions. Paid placements exist as a separate, labeled lane.

Here is how the selection works, what you control, and how to measure whether it is working.

How ChatGPT shopping works

When a question has shopping intent, ChatGPT can answer with product results: a visual card per product with an image, a price, a short summary of what reviewers say, and links to the merchants that sell it. Ask a follow-up, such as “only the ones under $150” or “which is best for wide feet”, and the list updates.

OpenAI has described the selection as independent: products are chosen on relevance to the question, structured metadata such as price and description, and the reviews and sources the model reads. The organic product results are not paid placements.

OpenAI has also opened ways for merchants to take part directly. It runs a merchant program for submitting product feeds, and it introduced Instant Checkout, which lets some shoppers buy from participating merchants without leaving the chat, built on the Agentic Commerce Protocol it published with Stripe. Availability varies by country and merchant, so check OpenAI’s merchant pages for what applies to you.

What actually gets a product recommended

Think of the recommendation as the answer to three questions the model is trying to settle.

Does this product fit the question?

The model can only match what it can read. Complete, structured product data is the entry ticket: an accurate title, category, price, availability, sizes or variants, and a description that says who the product is for. Product schema on your pages and a clean feed to the places that syndicate your catalog make your products legible.

Consistency matters as much as completeness. When your own site, your marketplace listings and retailer pages disagree on price or specs, the model has to guess, and a product it has to guess about is easier to leave out.

What do other people say about it?

Recommendations lean on reviews and third-party sources: review sites, publishers’ buying guides, comparison articles and forum threads. A product that appears in the roundups and discussions the model reads is a product it can justify recommending. This is why earning coverage, and reviews on the retailers that syndicate them, moves AI recommendations more than any change to your own copy.

Can the model explain the choice?

The answers that win are the ones that give the model something specific to say: “best for wide feet”, “quietest in its class”, “holds up to daily use”. If your product pages and the articles about your product state those differences plainly, the model can repeat them. If they only say “premium quality”, there is nothing to lift.

A practical checklist for merchants

1. Audit the buying questions. List the questions your customers ask before they buy, the way they would ask a person, and include constraints like budget, size and use.
2. Check what ChatGPT answers today. Ask those questions and note which products and which sources it names.
3. Fix your product data. Complete schema and feeds; identical prices and specs everywhere.
4. Submit your catalog where OpenAI’s merchant program is available to you.
5. Earn coverage where the model reads. The review sites, roundups and threads the answers cite for your category.
6. Publish comparison and guide content that answers the buying questions directly and names clear differences.
7. Measure every week, because answers change as the model, your competitors and the web change.

One of our own clients shows how far a single well-aimed page can go: Stealth Health Containers published one targeted listicle and its own site became the most cited source in its category, ahead of Amazon, with a 14.3% citation share.

Paid placements are a separate lane

Organic recommendations cannot be bought. What can be bought is a labeled placement next to the answer: OpenAI Ads runs sponsored chat cards that sit apart from the products ChatGPT picks on its own. For merchants, the two work best together. Paid placements cover the prompts where the organic answer leaves you out today, while content, coverage and product data work on winning the organic spot.

Rank Prompt connects to OpenAI Ads with an Advertiser API key, so you can manage chat-card campaigns next to the organic visibility data that tells you where to bid.

Measuring whether ChatGPT recommends you

You cannot improve what you only check by hand once a quarter. Rank Prompt runs your buying prompts on ChatGPT and five other AI engines every week, records which products and brands the answers name, which competitors appear instead of you, and which pages they cite as sources. For stores, the ecommerce playbook shows how to set up the prompts by category and price point, and the citation data shows exactly which review sites and roundups to earn a place in next.