ChatGPT Shopping: How Its Product Recommendations Are Picked

ChatGPT Shopping picks product recommendations organically: it ranks items by relevance to the user's request and conversation, using structured merchant feeds and signals such as price, availability, quality, reviews and whether the merchant is the primary seller, with no paid placement in the current feature. The result is a short list of products with images, prices, reviews and merchant options. For some items in the U.S., it also offers a Buy button.

As of October 2026, the job for retailers is to give ChatGPT clean, complete, current product data, and to support it with the reviews and third-party content the model reads. This article covers how the recommendation pipeline works, what OpenAI has confirmed and what is inferred, where the data comes from, how ChatGPT Shopping differs from classic ecommerce SEO, and a prioritized checklist with measurement steps.

What ChatGPT Shopping is

ChatGPT Shopping is a product discovery and buying feature built into ChatGPT. It has no standalone shopping website and no separate app. It works in two ways:

1. Discovery in chat. A shopping question returns products with images, prices, reviews, availability and merchant options. The list is ranked against preferences such as budget or style.
2. Buying in chat. Some products support Instant Checkout. The user taps Buy and confirms shipping and payment in ChatGPT, which passes only the necessary information to the merchant. The merchant fulfills the order and keeps the customer relationship. Products without Instant Checkout send the user to the merchant's site or app.

OpenAI also offers a shopping research feature that compares products, finds lookalikes, tracks deals and personalizes recommendations. Third-party apps run inside ChatGPT too: Sephora's app for beauty recommendations, and Klarna's Shopping Search app for visual results, prices, availability and offers. Shopify and Etsy are already integrated as merchant platforms.

How ChatGPT picks product recommendations

ChatGPT turns the conversation into a set of constraints, matches those against structured product data, and ranks the eligible items by relevance. The recommendations come from an underlying OpenAI model trained on a lot of data, so the model's reading of intent shapes the list as much as the feed does. The pipeline runs roughly like this:

1. Intent detection. The model recognizes a shopping query ("best trail running shoes for wide feet").
2. Constraint extraction. It pulls explicit limits from the prompt and earlier turns: budget, size, color, brand, delivery needs.
3. Retrieval. It matches products from feeds and structured sources. Third-party analyses describe semantic matching and vector search here, which means descriptive copy helps a product match even when the user's wording differs. OpenAI has not confirmed this detail.
4. Filtering. It drops items that are out of stock or that break a constraint.
5. Ranking. It orders the remaining products by relevance, then shows the merchant options for each one.

The ranking factors fall into two groups:

• Reported in launch coverage: relevance to the query, price, real-time availability, quality, primary-seller status, and Instant Checkout support.
• Inferred from how the system works: attribute completeness, the quantity and recency of reviews, and how well title and description language matches the way people describe needs.

Product ranking and merchant ordering are separate steps. A product can rank first while its merchant appears second in that product's seller list, behind a primary seller or one with Instant Checkout enabled.

Same prompt, different products

The table below uses illustrative outcomes. It shows how one follow-up line changes which product wins a query for "a carry-on suitcase".

Where the shopping data comes from and what merchants control

Shopping data comes mainly from merchant product feeds submitted to OpenAI's specifications, from integrated platforms such as Shopify and Etsy, and from other structured sources across the web. Whatever the model reads about a product also feeds its judgment of quality, which is why how large language models decide what to cite matters for shopping as well.

ChatGPT Shopping overlaps with classic ecommerce SEO and marketplace optimization, but the competition works differently:

• Overlap: clean feeds (as for Google Merchant Center), schema.org markup, strong reviews and accurate stock all carry over.
• Difference: there are no keyword bids and no ten-link page. A product either fits the constraints of the conversation or it does not appear.
• Difference: descriptive, need-based language ("wide toe box", "machine-washable") matters more than exact-match keywords.

Are ChatGPT Shopping recommendations ads?

No. The current feature ranks results organically, based on ChatGPT's research and product data, and it has no in-chat ad model or sponsored placements. OpenAI has discussed possible future "tasteful advertising" or affiliate fees, but neither exists in the product today.

Checkout is still changing. OpenAI announced Instant Checkout for supported U.S. items, with explicit user consent at each important step. Later reports describe most purchases going to merchant sites instead. Plan for both paths: a feed ready for in-chat checkout, and landing pages that convert a ChatGPT visitor who arrives with high intent.

Merchant checklist, prioritized by impact

Start with feed data that stops products from matching constraints. On-page data, reviews and operations come after that.

Products usually fail because of incomplete feeds, inconsistent taxonomy, duplicate listings, stale inventory, missing identifiers, poor images or titles that do not match the product page. Any one of these can drop a product from a constrained prompt. The work of getting your product into model answers starts with these fixes.

Measuring results

• Segment GA4 sessions by referrer domains such as chatgpt.com, and compare their conversion rate and order value with other channels.
• Run a fixed set of category prompts each week and log which of your products appear, their position and the merchant shown.
• Test one data change at a time, for example adding dimensions to 50 SKUs, then compare how often those SKUs appear against an unchanged control group over four weeks.

Don't rely on a single channel. Research from the Cloud Security Alliance records AI service disruption events across ChatGPT, Claude, Gemini and Microsoft Copilot, so track your visibility across several AI surfaces.

How Tellr helps brands get into AI shopping answers

Tellr runs a managed earned-visibility program for enterprise brands. A senior team works on the sources ChatGPT reads beyond your own feed:

• Comparison pages, reviews and answer-shaped articles written to be quoted by ChatGPT, Perplexity and Google AI Overviews, published to your CMS.
• A daily Reddit thread radar, with guideline-checked replies that pass an approval gate before posting.
• Weekly tracking of who Google and its AI Overviews cite for your category's queries.
• Category ad intelligence and ready-to-run creative.

Tellr is priced for companies spending $10k+ a month, so a small merchant with one Shopify catalog will get more from fixing its own feed first.

Key points for retailers

ChatGPT ranks products by how well they fit the conversation, and it can only do that with structured, current and complete data. Fix identifiers, attributes, variants and stock freshness first. Then build review volume and third-party coverage, and measure AI referrals against a fixed prompt set. Recommendations are unpaid today, so the brands that win in ChatGPT Shopping are the ones whose data and reputation give the model clear reasons to pick them.

FAQ