How ChatGPT recommends products, and how a store earns a place on the shortlist

The scale is why merchants suddenly care. ChatGPT reached 900 million weekly active users in February 2026, per TechCrunch, and Adobe data reported by TechCrunch shows AI traffic to US retail sites grew 393% year over year in Q1 2026. In our own audit of 181 ecommerce brands, 87% were absent from all four AI engines, so the traffic is real and most brands are not in it.

This guide covers where the carousel gets its candidates and what OpenAI says it ranks on, then sets out the order to fix things in. For the conversational side, see our guide on how AI decides which brands to recommend.

How does ChatGPT Shopping actually recommend products?

OpenAI says product selection considers structured metadata from first and third-party providers, the model’s own prior response, and user context like Memory and Custom instructions, filtered through its safety standards. Results match intent, not keywords, and no partnership with OpenAI changes what shows up.

The OpenAI Help Center puts it in one sentence: “Product results are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships.” That leaves every ranking factor you can influence in your data.

Your site needs to be reachable by OAI-SearchBot in robots.txt, or the reading path never sees you, and reviews matter on your product pages and on third-party sites because quality is one of the four merchant ranking factors OpenAI names. Our walkthrough of AEO in product searching and the new buyer path follows a shopper from first question to decision.

Ads are a separate surface. OpenAI began selling ChatGPT ads in 2026, and Productsup reported product feed ads built from a merchant’s catalog in May 2026, which is what the feed spec’s ads-eligibility flag is for. The not-ads statement above covers organic product results only.

What are the two systems behind ChatGPT shopping recommendations?

Two separate pipelines wear the same name. The quick shopping carousel re-ranks candidates pulled from product feeds and Google Shopping data, and it does this in seconds. Shopping research is a different system, a version of GPT-5 mini trained for shopping tasks that reads product pages, avoids weak sites, and cites its sources.

393%

year-over-year growth in AI traffic to US retail sites in Q1 2026, per Adobe data reported by TechCrunch.

The quick carousel

Ask ChatGPT for “best running shoes under $150” and the tile carousel that appears is a re-ranking job: the model scores candidates from feeds and shopping data against your query and context, then renders the tiles. That path is fast and shallow, and your feed quality decides it.

Shopping research

The deeper mode. OpenAI announced it on November 24, 2025 as running on a version of GPT-5 mini post-trained for shopping tasks, per Search Engine Journal. It reads trusted sites the way a careful buyer would, skipping low-quality or spammy ones, and it cites the sources it used. OpenAI also states these chats are never shared with retailers.

The carousel responds to structured data and shopping research responds to pages it can read and cite, so the playbook below sequences work for both.

Does ChatGPT use Google Shopping?

Yes, heavily, for the carousel. Peec AI matched 43,000 products from 5,000 ChatGPT carousels against Google and found over 83% sitting in Google’s top 40 organic shopping results, and Semrush found the top ChatGPT product in Google Shopping’s first 3 results 75% of the time, in a test it calls small.

This changes where you spend effort first. Three studies point the same way, though they count different things:

We treat 83% as the headline number and the 100% claim in the later Precis write-up of Peec’s analysis as the stricter reading. Either way the direction is the same.

The mechanics underneath come from the Peec data reported by Search Engine Land. When you ask a shopping question, ChatGPT fires shopping fan-out queries, about 1.16 per prompt, each around 7 words, and those queries pull the candidate pool the model then re-ranks. Semrush also found Base64-encoded Google Shopping queries in ChatGPT’s own network traffic, which shows the link directly.

Your Google Merchant Center feed is therefore a ChatGPT ranking input. If your products are absent from Google Shopping’s top 40 for a query, the carousel has nothing of yours to re-rank, and the gate sits in a feed many merchants have not opened in months.

How does ChatGPT select and rank products once you are in the pool?

OpenAI ranks merchants on availability, price, quality and whether they are the maker or primary seller of the item. On top of that sits intent matching: the model weighs your product data against the user’s query, their Memory, and their Custom instructions before anything reaches the carousel.

Per the Help Center, merchants are “ranked based on factors like availability, price, quality, and whether they are the maker or primary seller of that item.”

The four merchant factors, operationalized

• Availability. Keep stock data accurate in every feed, because an out-of-stock flag removes you from consideration and a stale in-stock flag that leads to a dead product page burns trust.
• Price. The price in your feed must match the price on the page, since mismatches read as unreliable data and uncompetitive pricing costs visibility too.
• Quality. Reviews and ratings, mostly. Feed fields exist for both, and third-party reviews feed the shopping research path as well.
• Maker or primary seller. The quiet edge for DTC brands, because you make the thing and resellers do not. Win your own brand-name queries first.

What the feed fields control

OpenAI publishes a full product feed spec at developers.openai.com, and the fields map directly to how products appear. Required: item_id, title (150 characters), description (up to 5,000 characters), url, brand, image_url, price with currency, availability and seller_name. Also in the spec: star_rating and review_count (optional) and group_id for variants. Three flags control enablement: is_eligible_search, is_eligible_checkout, and is_ads_eligible.

Shopify and Etsy merchants skip the application entirely. Both catalogs are already integrated, per OpenAI’s merchant page.

How do I get my products recommended in ChatGPT?

Fix the inputs in the order the machine consumes them. Google Merchant Center feed first, because the carousel leans on Google Shopping’s top 40. Catalog and feed hygiene second. Product schema and review capture third. Third-party mentions last, because they feed the conversational path and, per Peec, probably nudge the carousel too.

Peec

84%

of carousel matches sat in Google’s top 20, per Peec, so a top-20 Google Shopping rank is the working target.

Photo: OleksandrPidvalnyi, Pixabay

Most guides list tactics without sequencing them by mechanism, so merchants do the satisfying work before the load-bearing work. For a DTC brand, the order looks like this:

1. Fix your Google Merchant Center feed. The top-40 dependency makes this the gate: accurate titles, complete attributes, live availability, competitive prices and no disapprovals. Peec found about 60% of carousel matches in Google’s top 10 and nearly 84% in the top 20, so treat a top-20 Google Shopping rank on your money queries as the working target. If you are invisible in Google Shopping for those queries, you are invisible in the ChatGPT carousel too.
2. Clean up your Shopify Catalog or submit the OpenAI feed. Titles within 150 characters that lead with what the product is. Descriptions that use the 5,000-character allowance to answer real buyer questions. Variants grouped with group_id. Populate star_rating and review_count; empty review fields concede the quality factor to whoever filled theirs in.
3. Product page schema and review capture. Product, Offer, and AggregateRating markup on every PDP. This feeds both surfaces: it is structured metadata for the carousel path and parseable trust for the reading path. Then capture reviews relentlessly, because they are the visible quality signal on both.
4. Earn third-party mentions and citable content. Shopping research reads trusted pages and cites sources, so the lever is presence on pages the model trusts and content structured to be quoted. Our playbook on how to get cited by ChatGPT covers this path in full.

Steps 1 and 2 target the carousel and steps 3 and 4 target shopping research, with the feed work first because it gates the surface with the most predictable mechanics.

Our RIPT Apparel case study shows the feed side on an apparel store: the rewritten product titles feed Merchant Center as well as organic search, and Merchant Center clicks over 28 days were up 317.8% year on year. Semrush’s AI Search panel for riptapparel.com on 25 September 2026 shows 157 cited pages and 298 mentions across ChatGPT, Google AI Overviews, AI Mode and Gemini, and we have not isolated which change produced that. For a much bigger catalog the same play scales, as in scaling 5,000 product titles with AI.

What changed with Instant Checkout and the March 2026 pivot?

On March 24, 2026, OpenAI said it would let merchants use their own checkout and concentrate on product discovery. That walked back standalone Instant Checkout as the endgame, and most guides published before March are stale on this.

The announcement was titled “Powering Product Discovery in ChatGPT.” Modern Retail reported that people researched products in ChatGPT but rarely bought there, and that Walmart saw in-chat sales convert at about one third the rate of sales that sent shoppers to its own site. Dataiads lists Target, Sephora and Nordstrom as integrated, with retailers submitting catalogs, prices and availability directly to OpenAI.

The money question is now “which products does ChatGPT surface at all,” so guides still pitching Instant Checkout as the goal are optimizing for last year’s product.

Can you pay to appear in ChatGPT shopping recommendations, and what else can’t you control?

Not in the organic results. OpenAI states product results are organic, selected independently, and not influenced by partnerships, and its ads run on a separate surface. Several inputs also sit outside your control: the model’s priors, Memory personalization, price-data lag and rollout timing. Any agency promising guaranteed carousel placement is selling something that does not exist.

The honest list of what you cannot fix:

• Model priors. The model arrives with opinions about categories and brands from training, and you influence them slowly through the third-party record rather than your own site.
• Memory personalization. Two users asking the identical question can get different carousels, because Memory and Custom instructions are ranking context, so there is no single result to screenshot.
• Price-data lag. Feeds refresh on their own schedule, so a flash sale may not reach the carousel while it is running.
• Rollout timing. OpenAI decides which features reach which markets, and no merchant action changes that.

What you can control, the feeds, the data quality, the schema and the mentions, is plenty.

How big is AI shopping traffic really?

Big, and it converts. Digital Commerce 360 reported that in May 2026 AI-referred traffic to US retail sites converted 54% better than non-AI traffic, and its August report put July 2026 at 60% higher, the 11th straight month AI traffic converted better, with 53% more revenue per visit. Both rest on Adobe data covering more than 1 trillion visits.

Earlier Adobe work points the same way. In its March 2025 analysis, AI-referred visitors viewed 12% more pages per visit and bounced 23% less than other traffic sources, and 39% of its 5,000 survey respondents had used generative AI for online shopping.

Want to know whether ChatGPT recommends your products today? Run the Free AI Visibility Audit. It checks your brand across ChatGPT, Perplexity, Gemini and Copilot and shows you exactly what buyers see when they ask about your category, free, with no card, and the report arrives within 48 hours.