How ChatGPT Recommends Products, and Where Labels Like "Best All-Rounder" Come From
ChatGPT recommends products by matching a shopper's request against structured product data from merchants and third-party feeds, plus whatever the model already believes about those products before it searches. The short labels it's now testing on those results, like "best all-rounder" and "best for beginners," are model-written summaries of reviews on public websites, which means the one-line positioning a shopper sees for your product mostly comes from what other people have written about it.
That's my read of OpenAI's own help documentation, set next to a test that showed up in ChatGPT last week. I wanted to put the pieces in one place, since "how does ChatGPT recommend products" is a question a lot of brand teams are about to care about.
The short version
ChatGPT chooses products from three inputs OpenAI names: structured product metadata, the model's own answer before it searches, and OpenAI's policies. Product results aren't ads.
The "best for" labels are generated by the model from public reviews, and OpenAI says they aren't verified.
Merchants get ranked on availability, price, quality, and whether they make or primarily sell the product.
You can influence the product data and the reviews behind a label, though there's no way to buy one.
What ChatGPT is testing on product results
On October 6, Brodie Clark spotted ChatGPT testing a new version of the labels it shows on organic product results: a green "best all-rounder" tag on one product and gray "best" tags on others, like "best for beginners" and "best cushioning" (Search Engine Roundtable, Brodie Clark's notes). Clark describes it as an expansion of an earlier version of the labels. Barry Schwartz's guess is that the labels should help click-through on the products that get them, and I'd agree with that, with the obvious flip side that they'll hurt click-through on the products sitting next to them without one.
It's a test, so it may change or disappear. I'm writing about it anyway because the mechanism underneath it is documented and isn't going anywhere, and because "which product is best for what" is the exact question a shopper brings to ChatGPT in November.
How ChatGPT recommends products, in OpenAI's words
OpenAI's help article on shopping is more specific than most people realize. A product shows up in the carousel, it says, "when ChatGPT perceives it to be relevant to your intent," and it names three things that go into that decision (OpenAI Help Center):
Structured metadata from first-party and third-party providers (price, description, availability, that kind of thing).
The model's own responses generated before it considers any new search results.
OpenAI's safety standards and product policies.
That second one gets skipped over a lot, and I think it matters more than the other two for most brands. Before ChatGPT goes looking, it already has an opinion, built from everything it read in training, and the search results get layered on top of that. So a brand that was well described across the web a year ago starts the conversation ahead.
Then, once a product is picked, there's a separate ranking for the merchants who sell it. OpenAI says merchants are ranked on "availability, price, quality, and whether they are the maker or primary seller." Shopify merchants are already integrated through Shopify Catalog, and everybody else can apply for a direct product feed. OpenAI is also flat about money: product results "are not ads, nor influenced by any OpenAI partnerships," and ads are kept separate.
How ChatGPT shopping works
Three decisions, three sets of inputs
Source: OpenAI Help Center, "Shopping with ChatGPT Search", read October 12, 2026. Label test: Search Engine Roundtable, Oct 6, 2026.
Where the "best for" labels come from
The label part is on the same help page. OpenAI describes its feature labels (its example is "Budget-friendly") as not "guarantees or verified statements," and says the review summaries are "model-generated" and "based on reviews from public websites." It also says plainly that "reviews and ratings are not verified by OpenAI."
So when ChatGPT decides your running shoe is "best for beginners" and the other one is "best cushioning," it's compressing what reviewers, retailers, and probably some forums have written into a few words. Those words aren't yours, and there's no field for them in a product feed, so your best shot is making sure the people who write about your product have a clear, consistent reason to say the thing you'd want the label to say.
I'm less sure about which reviews count. OpenAI doesn't say which review sites it reads, and my own data makes me cautious about guessing. In the AI Verdict Study I logged every citation four engines gave across 432 scripted buyer conversations, and ChatGPT was the engine that leaned hardest on company websites. Brand sites, rival sites, and other vendors were 84% of its classified citations, and independent reviewers were 2% (Where AI Engines Go to Decide About Your Brand). That was conversational advice, which isn't the same surface as a shopping carousel, so the review sources behind a label could look very different. I'd treat "which reviews count" as an open question for now and cover more than one kind of source.
Why the first answer matters so much
Labels sit on the product carousel, which is often the first thing a shopper sees, and the first answer turns out to carry a lot of weight. In the Verdict Study, when an engine named a brand in its first response, before the buyer brought it up, the conversation ended in a rejection 18% of the time. When it didn't, 44% (AI Verdict Study). I can't prove the first mention causes the better ending, and some of it is probably just that strong brands get named early and also get recommended late. Still, if a label nudges the shopper toward one product on that first screen, it's working on the moment that already predicts how the conversation ends.
The timing matters too. Adobe expects traffic from AI services to U.S. retail sites to rise 130% this holiday season compared to last year (Adobe), and last season's AI traffic to retail sites was up 693% year over year, with those visitors converting 31% better than other traffic and 33% less likely to leave right away (Adobe). More shoppers are about to see these carousels, and the ones who click through have been the ones more likely to buy.
Why the first screen matters
before the holidays, DO THESE FOUR THINGS
Number one, decide what label you'd want and check whether anyone is saying it. Pick the one "best for" phrase that's true of your product, whether that's best for wide feet or best for small kitchens or something else entirely, and then go read your reviews on retailer sites, the roundups that cover your category, and the Reddit threads where people argue about it, and see if that phrase, or something close to it, shows up. If it doesn't, the model has nothing to compress into the label you want, and no amount of on-site copy will fix that by itself.
Two, clean up the data that feeds the carousel. If you're on Shopify, your catalog is already flowing to ChatGPT, so the question is whether the titles, descriptions, prices, and stock status are right. If you're not, look at OpenAI's direct feed application. Merchants get ranked partly on availability and whether they're the maker, so a brand selling direct with accurate inventory has an edge it should use.
Three, give reviewers something specific to say. A review that only says the product was great and shipped fast doesn't give a model much to work with when it decides what you're best for. My guess is that post-purchase emails asking a specific question ("what did you use it for?") get more specific reviews back, and a product with a clear use case is easier for anybody to describe. I'd be careful here, though. Paying for or scripting reviews is a fast way into trouble, and a label built on reviews only means something if the reviews are real.
Four, don't forget your own help docs and comparison pages. In the Verdict Study, help center articles were 29% of the citations to a brand's own site, and on the "what are the limitations" question it was 46% (Where AI Engines Go to Decide About Your Brand). If the label says "best for beginners," a reasonable next question from the shopper is "what's the catch," and the engine will go read your docs to answer it.
Then last, check it yourself, on a schedule. Run the questions your buyers ask, in a logged-out session, and screenshot what comes back, labels included. Do it more than once, because answers move. In my study, 21 of 72 brand and engine pairings got a recommendation on some runs and a rejection on others, with the buyer's words identical every time (Same Buyer, Same Brand, Four Different Verdicts). If you want a structure for doing that, the free Quick Check walks you through a first pass, and the $199 Audit Kit goes further if you want to run the whole thing yourself.
Self-check
Is your product ready for ChatGPT shopping?
Six questions. Answer for one product you care about this holiday season.
Answer all six to see your score.
Questions based on OpenAI's shopping help article and the AI Verdict Study. Your answers stay in your browser.
THE UNKNOWN AT THIS POINT
A few things I'd keep an eye on. The label test might not ship, or might ship looking different, and I haven't seen anything from OpenAI about it directly. I don't know how often the labels get regenerated, so I can't tell you whether fixing your reviews this month changes the label next week or next quarter. I also don't know how labels interact with ChatGPT's ads, which OpenAI says are separate; I wrote about the self-serve ads manager back in June and the line between the two is something I'd watch.
I've argued for a while that the answer is where the shopper forms an opinion, and that being in it is the number to watch (The Click That Didn't Happen). A label squeezes that opinion into three or four words. Shoppers still tend to finish the purchase on the merchant's own site (The Click Survived Checkout), so the carousel and its labels are deciding which site gets the visit. Why the engines settle on some brands and pass on others is most of what my book, Explainable, is about, and a label is that decision showing up in one line of gray text.
If you've seen these labels on your own products, or on a competitor's, I'd love to hear what they said and whether it matched how you'd describe yourself. Send me a note through the contact page or find me on LinkedIn.
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services. Find him at jarredsmith.com.