The short answer: AI shopping assistants still get prices, availability and product details wrong often enough that their makers tell shoppers to double-check. For an independent jewelry brand, the recommendation is only the first step. The sale is decided on the live product page.
OpenAI says so itself. When it launched shopping research in ChatGPT in November 2025, the company wrote that the feature “might make mistakes about product details like price and availability, and we encourage you to visit the merchant site for the most accurate details.” The brand’s job is to make the facts on that site impossible to misread.
How often do AI shopping tools get product details wrong?
Often enough to measure. In a benchmark reported by Android Headlines on September 22, 2026, Product.ai put 220 straightforward e-commerce questions across nine categories to ChatGPT, Claude, Gemini and Perplexity. Gemini performed worst on price: 56% of its responses contained at least one pricing mistake. Perplexity’s paid tier did best, at 14%. Every model also contradicted itself at times when asked the same question minutes apart.
Since these language models operate on probabilities, their inability to give consistent answers minutes apart shows how risky it is to treat them as automated shopping advisors.
Product.ai sells a product-verification service for shoppers and AI agents, so it has a commercial interest in showing that unverified answers are risky.
A separate test by data company Smarter Sorting, published January 28, 2026, followed 100 products through 2,500 interaction steps on five assistants. It found that 71.8% of steps failed its full product-accuracy check, and that availability checks failed in more than half of cases. Its most common error was “variant confusion”: choosing the wrong size or version of the right product. Smarter Sorting also sells product-data services.
Why do the details matter more for fine jewelry?
Because the details are the product. A shopper deciding on a ring may need to know whether it is solid 14K gold or vermeil, whether the stone is natural or lab grown, which sizes exist, what the final price is in their currency, and whether the brand ships to their country. Earrings add another trap: is the listed price for a single earring or a pair?
These are the kinds of attributes the benchmarks show AI tools blurring. A made-to-order piece can be presented as ready to ship. A price can be old, or quoted in the wrong currency. “Ships internationally” can hide the fact that the shopper’s country is excluded, or that duties and return costs change the real total.
A recommendation is a starting point
Check the live product page before deciding.
| AI recommendation says | Live product page confirms |
|---|---|
| Price and currency | Current total and currency |
| Metal and stone | Exact material specifications |
| Single or pair | What the listed price includes |
| Available to buy | Stock and production time |
| Ships to my country | Destination, duties and returns |
Who gets blamed when a recommendation is wrong?
Usually the brand. In a survey of 1,046 online shoppers in the U.S. and U.K., published by commerce platform Rithum on March 25, 2026, 58% said they blame the retailer or brand when an AI recommendation contains incorrect product information, and 16% said they would avoid buying the product entirely after a bad recommendation. Sixty-seven percent said price is the most important thing AI must get right. Rithum sells product-listing and data tools, so the finding also supports its pitch.
AI is making consumer trust a product data problem.
The stakes are rising because the traffic is. According to Adobe Digital Insights, AI-referred traffic to U.S. retail sites grew 62% year over year in July 2026, and those visitors converted at a rate 60% higher than non-AI traffic. Shoppers arriving from AI tools are close to buying. A mismatch on the product page is the last place a brand wants to lose them.
What can an independent jewelry brand control?
Not the sentence an AI assistant writes. The brand can control the source information the assistant draws on, and the page the shopper lands on.
Make the product page unambiguous. State metal and stone specifications in plain language. Show the current price and currency, and say whether earrings are sold singly or as a pair. Make sizes, production time, shipping destinations, duties and returns easy to find. If a variation changes the price or material, give it its own accurate description. Adobe reported in April 2026 that retail product pages were the least machine-readable page type it measured, averaging 66%.
Keep structured data and feeds current. Google’s merchant listing documentation says product markup can “highlight more specific data about a product, such as its price, availability, and shipping and return information.” OpenAI’s merchant page invites brands to share product feeds and says “richer data helps your products surface with more accurate information.” Neither guarantees a product will appear or be described correctly. Both make the brand’s own facts easier to find than a stale copy elsewhere.
Consistency across channels matters too. As we noted in How to Get Your Jewelry Brand Recommended by ChatGPT, an assistant repeats what it reads. If a marketplace listing and the brand’s own page describe a piece differently, it may pick the wrong one.
How should you test what AI tells shoppers about your pieces?
Test the experience as a customer would.
- 1
Ask like a customer. Put a specific request to two or three assistants, with a country, a budget and a material. For example: a solid gold garnet ring under $900 that ships to Canada.
- 2
Follow every link. Compare each answer with the live page: price and currency, metal and stone, single or pair, stock or production time, destination and returns.
- 3
Ask again. Repeat the same prompt a few minutes later. The benchmarks above show answers can change between runs.
- 4
Fix what you own. Note recurring errors, then clarify anything ambiguous on your side: page copy, variant names, structured data or your product feed.
If your page is already clear and the error persists, it may be beyond your control. You will still know what your customers are seeing, and what to clarify when they ask.
Shopping research might make mistakes about product details like price and availability, and we encourage you to visit the merchant site for the most accurate details.
FAQ
Are ChatGPT shopping recommendations accurate?
Not always. OpenAI says its shopping research “might make mistakes about product details like price and availability” and encourages shoppers to visit the merchant’s site. Independent benchmarks in 2026 found pricing errors and inconsistent answers across all the major assistants tested.
Who do shoppers blame when AI gives wrong product information?
Mostly the brand. In Rithum’s March 2026 survey of 1,046 U.S. and U.K. shoppers, 58% said they blame the retailer or brand when an AI recommendation contains incorrect product information.
What should a jewelry brand state on every product page?
Metal and stone specifications, the current price and currency, whether earrings are sold singly or as a pair, available sizes, stock or production time, shipping destinations, duties and returns.
Can a brand stop AI tools from getting its products wrong?
No brand controls what an assistant writes. Clear product pages, accurate structured data and a maintained product feed make the correct facts easier to find and harder to misread.
How do I check what AI says about my jewelry?
Ask two or three assistants for a specific piece, including a country, a budget and a material. Compare each answer with your live product page, repeat the prompt a few minutes later, and fix anything ambiguous on your side.