ChatGPT now says how it ranks stores for a product. Here’s what CPG brands should fix

OpenAI rewrote its ChatGPT shopping help page at the end of October. Most of the coverage went to virtual try-on. The more useful part for a CPG brand is a short section near the bottom called “How merchants are selected.” For the first time, OpenAI names the factors it uses to order the list of stores under a product.

Here’s what it says, what two new studies add, and what we would fix first.

How does ChatGPT decide which store to show for a product?#

When a shopper taps a product in ChatGPT, it can show a list of merchants that carry it. OpenAI says that list is built from merchant and product metadata that comes from third-party providers or straight from merchants. Then the key line: “Merchants are ranked based on factors like availability, price, quality, and whether they are the maker or primary seller of that item.”

Think about what that means for a protein bar or a laundry pod. The same item is sold by Amazon, Walmart, Target, a handful of resellers and maybe your own site. Four things decide who gets the click: in stock, price, a quality signal, and whether the seller is the maker. Only one of those is a brand advantage that a reseller can’t copy. You only get credit for it if your data says you are the maker, consistently, everywhere the product appears.

The page also explains how products get into the carousel in the first place. ChatGPT looks at structured metadata from first-party and third-party providers, other third-party content, and “model responses generated by ChatGPT before it considers any new search results.” That last input is the pre-search shortlist we wrote about in how ChatGPT picks brands before it searches. A brand the model already knows starts ahead.

Does ChatGPT rewrite product titles and labels?#

Yes, and the page says so plainly. ChatGPT may generate simplified titles and descriptions from third-party data because merchants describe the same product in different ways. It adds its own labels such as “Budget-friendly” or “Most popular.” OpenAI’s example: “Budget-friendly” might mean reviewers often mention good value, not that the price is the lowest. Review summaries are written by the model from public reviews, which OpenAI does not verify.

Two pricing details matter for promotions. The price in ChatGPT’s first answer typically comes from the first merchant listed, which may not be the lowest. And when merchants change price or shipping terms, “there may be some delay” before ChatGPT shows it. If you run a two-day Target Circle deal, assume ChatGPT may still show the old price for part of it.

The practical read: whatever your reviews say most often becomes your label. If shoppers keep writing “great value” about your premium SKU, ChatGPT may tag it as the cheap option. Watch review language the way you watch star ratings.

What goes into an OpenAI product feed?#

Shopify merchants are covered automatically through Shopify Catalog, and OpenAI says no extra work is needed. Everyone else can apply for direct feed access. The feed specification lists nine required fields for discovery: item ID, title, description, URL, brand, seller name, image URL, availability and price. Titles should stay under about 150 characters and descriptions under 5,000, in plain text.

A few rules in that spec trip up CPG catalogs:

• GTIN is optional, but if you send one it must be 8, 12, 13 or 14 digits with a valid check digit and its leading zeros kept.
• Seller name must be real, not a placeholder. For a marketplace offer, the seller and the marketplace are separate fields.
• Availability has to be one of five values. A blank or unrecognized value rejects the row.
• Shipping price left empty means unknown, not free.
• Reviews go in as a count and an average rating for the same set of reviews.

None of this is glamorous. It is exactly the input to the ranking factors above.

How often do AI shopping answers name a brand at all?#

Not often, according to two studies published in the last week. Found by Flux asked ChatGPT, Perplexity and Google AI Overviews the same 150 shopping questions per category every week since August, across 66 category lists in the U.S., U.K. and Australia. Across 2,541 brands and 313,600 answers, 72.6% of brands appeared in fewer than 5% of answers for their category. The median brand appeared in 1.9%. The average category leader appeared in 26.6%.

The platforms don’t agree either. In 35 of 56 categories, the three engines named a different top brand. On average ChatGPT named a given brand in 5.4% of answers, Perplexity in 5.1% and AI Overviews in 4.3%. Top-10 brands also swung an average of 4.2 points between weekly scans, so one check tells you little.

Yarnit’s fall study is smaller, 1,100 ChatGPT and Gemini answers about 56 consumer brands, but it splits results by the kind of question asked. Brands were named in 37% of direct recommendation answers, 17% of “what should I look for” answers and 8% of answers to questions about concerns.

Two Yarnit findings hit CPG directly. Beauty and personal care was the least visible of the larger segments: named in 16% of answers and cited in 3%. And the engines behaved like separate channels. ChatGPT linked to the brand’s site in 70% of its mentions, Gemini in 8%. Neither study is CPG-only, and both are run by vendors that sell visibility work, so treat them as directional. They point the same way as the OpenAI page: the engines name products they can describe and match with confidence.

Where does Amazon fit?#

Mostly outside this picture, by choice. The Wall Street Journal reported on November 1 that Amazon blocked Meta’s Muse from shopping its site while Walmart and Shopify opened up to agents, according to PYMNTS. PYMNTS Intelligence also found Amazon accounts for 29% of research starts in its survey but 59% of the most recent AI-assisted purchase. Shoppers research in an assistant and still buy at Amazon, where Alexa for Shopping does the picking (we covered that in our earnings week post).

So on Amazon, the detail page and reviews decide. Everywhere else, the feed and the open web decide.

What should a CPG brand fix first?#

Our order of work: identity, then availability and price, then reviews, then measurement. Score yourself:

The terms here, from agentic commerce to AI share of voice, are defined in our glossary.

If you want to see which of your SKUs ChatGPT, Gemini and Perplexity name today, and which store they send shoppers to, run a free SKU audit or book a demo. How we score each product is on the SearchPanel homepage.