The feature, called Shopping Research, appears as an option in ChatGPT when users ask a shopping-related question or access it from a dedicated menu. To start, shoppers can write a general or specific request, such as help finding a dress under $300, comparing three bluetooth speakers or identifying the best gift for a particular kind of person. ChatGPT then generates a short quiz to clarify priorities like size, budget or features. Once the model begins pulling products, the interface becomes interactive: Users swipe right on items they like and left on items they don’t. The tool then delivers tailored product recommendations and a buyer’s guide, drawing on user preferences and previous chat history, as well as prices, specs, images, availability and reviews sourced from publicly available websites.

OpenAI says consumers already ask ChatGPT 50 million shopping-related queries a day — queries like, “How much are running shoes?” ChatGPT can answer basic shopping questions by summarizing product info from the web, but OpenAI’s new Shopping Research is more personalized and sophisticated. Shopping Research, now available on both web and mobile to all logged-in ChatGPT users, launches just in time for the peak holiday shopping season.

But Shopping Research won’t pull from every major retailer. OpenAI said that the tool only surfaces information from websites that permit its browsing agents to access their pages. When asked whether that includes Amazon, OpenAI clarified that the system follows each site’s rules.

“We respect all of the OpenAI robots.txt,” Isa Fulford, who leads OpenAI’s Deep Research and ChatGPT agent teams, told reporters at a press briefing last week. “Anything that allows us to access their site, our product will access, and anything that we don’t, we won’t access.”

That means Amazon product listings may be limited — if not entirely nonexistent — within ChatGPT’s Shopping Research experience. In recent months, Amazon has updated its robots.txt file to block several OpenAI crawlers — including the agents used for model training, real-time browsing and its SearchGPT engine — part of a broader push to prevent third-party AI tools from scraping its sprawling e-commerce site. While OpenAI did not name Amazon specifically, its remarks suggest that the new shopping research tool will only display Amazon product pages in cases where Amazon’s robots.txt allows it.

Currently, if a user asks ChatGPT for specific product recommendations on Amazon — say, laptops under $1,000 — the AI chatbot will surface a handful of options sold by other retailers and suggest the user manually “check if they’re available on Amazon,” according to a recent test conducted by Modern Retail.

In another test, when asking ChatGPT if it could “show me Amazon product listings” more broadly, the agent answered, “I can’t browse or display live product listings from Amazon.”

Although this may seem like a shortcoming of ChatGPT, including its new Shopping Research feature, independent e-commerce analyst Juozas Kaziukėnas said the impact is fairly limited.

“Luckily for OpenAI, there are plenty of other retailers and options to get this information from,” he said, pointing to Walmart, Best Buy and brands’ own websites. “There are very few products on Amazon that are not available elsewhere.”

‘Battle of the giants’

Instead of opening its doors to outside agents, Amazon is investing in its own AI tools. For instance, Amazon has rolled out features like “Auto Buy,” which automatically purchases items for customers when prices drop. Amazon also began testing an agent earlier this year called “Buy For Me” that lets shoppers buy products from other websites without leaving its app. Amazon says shoppers who use Rufus are 60% more likely to finish a purchase, and it expects the tool to generate more than $10 billion in yearly sales.

Meanwhile, other retailers are embracing shopping within ChatGPT. OpenAI has inked deals with Walmart, Etsy and Shopify merchants that make their wares available for purchase through ChatGPT’s Instant Checkout feature. Last week, OpenAI and Target also announced plans to launch the retailer’s app inside ChatGPT.

Despite these partnerships, OpenAI says Shopping Research won’t show any favoritism in its product recommendations. The tool displays product recommendations that are organic and unsponsored, meaning they are ranked primarily by relevance to the user’s query. Shopping Research is trained to prioritize high-quality, trustworthy websites, with a focus on organic content, including review sites and even Reddit.

For now, there’s no checkout feature within Shopping Research, but OpenAI said deeper integrations with ChatGPT’s existing in-chat checkout tools are possible later. “In the future, I can totally see them being very complementary,” Fulford said, “but we just wanted to focus first on making a really great discovery experience.”

To that end, the tool is trained to avoid sites that may be low quality, spammy or overly commercial. Fulford defined low-quality sources as “sites where there are pop-up ads” or “clearly advertisements,” as well as those with conflicting or suspicious review patterns.

Moreover, Shopping Research is less likely to surface results from certain e-commerce platforms. During a Q&A session at last week’s press briefing, a reporter specifically asked OpenAI researchers if the new Shopping Research tool would elicit suggestions from Temu. “If you ask for Temu options specifically, it will respect that constraint and find items from Temu,” said Manuka Stratta, a researcher on OpenAI’s shopping team in response. But by and large, Temu “isn’t going to be recommended that much,” and customers looking for such products will have to ask for it.

Kaziukėnas said sellers and marketplace merchants aren’t likely to feel meaningful effects as a result of Amazon’s decision to block third-party shopping agents, at least not yet. “For now, this is like a battle of the giants,” he said. While companies like OpenAI, Perplexity and Amazon are racing to develop the default AI shopping agent, consumer adoption remains low compared to traditional forms of e-commerce. “If you’re a smaller brand or retailer, it’s not yet that important.”