By Carson Cobb. Research from November 4 to 17, 2026.
OpenAI doesn’t publish how ChatGPT picks products. But you can measure what it does. Over two weeks I asked it 48 shopping questions across six product areas, repeated questions to see what stays the same, checked its facts against real product pages, and scanned 617 Shopify brand stores to see what the brands it names have in common. This is what decides the pick, in order of how much it seems to matter.
1. What independent sources say
ChatGPT showed a source or store link in 45 of 48 answers. Review and testing sites appeared in 35% of answers, magazines and publishers in 33%, and medical, veterinary and research sources in 29%, mostly in pet, supplements and beauty. Reddit was a visible source once. Amazon never was. (Sources ChatGPT cites)
That’s why the Shopify brands it named looked like brands reviewers write about. In my scan, they published far more (a median of 142 blog articles against 66) and their product pages backed claims with testing and comparisons. (What AI named brands have in common)
2. Specific, checkable reasons
Every pick came with a reason. Value, materials and ingredients came up most, in 29 to 37 answers. Testing and editor reviews were a reason in 24 of 48 ChatGPT answers. Vague claims, like “premium quality,” didn’t get repeated. (Why AI recommends products)
3. The shopper’s situation
ChatGPT labeled picks in 44 of 48 answers, and 45% of its labels named a specific need, like “best for sensitive skin.” In 18 of 48 answers it asked a follow up question about the shopper first: skin type, budget, a dog’s weight, a child’s age. (Best for labels, follow up questions)
4. A short, stable list
ChatGPT named a median of 4 brands per answer and never more than 8. The first pick came back in 98% of repeat runs. But the lists barely repeat across questions: 193 different brands across 48 answers. Wording changes the list far more than asking again does. (How many brands, how often picks change)
5. Clean product data, for getting the facts right
Structured data didn’t decide who got named. Named and unnamed brands had Product structured data at the same rate. But it decided whether ChatGPT got the facts right: prices were right for all 20 products with product structured data, and ratings were often wrong, especially where a review app added a second copy of the product data. (What AI gets wrong)
What this means for a brand
| What decides it | What to work on |
|---|---|
| Independent sources | Earn coverage on review, testing and expert sites |
| Specific reasons | Put checkable claims on the product page: test results, materials, certifications |
| The shopper’s situation | Say exactly who each product is for |
| A short list | Test many question variations, not one |
| Facts | Clean structured data, one Product block, GTINs, policies |
My guide to getting Shopify products recommended by ChatGPT turns this into steps. Finding where ChatGPT names you and why is part of my AI Search and Conversion Audit. The AI Visibility Check is a quick first look.
How this was done
- Questions: 48 shopping questions across supplements, beauty, food and drink, pet, home and apparel, asked in ChatGPT in November 2026, plus repeat runs and reworded versions.
- Fact checks: ChatGPT’s stated prices and ratings compared with 24 product pages.
- Stores: 617 Shopify brand stores scanned November 16 and 17, 2026.
- Limits: this measures what ChatGPT does, not its internal rules, which OpenAI doesn’t publish and can change.
No store or brand is named in this report.