Ask whether AI shopping assistants are worth using in 2026 and you get two very different answers depending on which one you are talking about. Amazon says its assistant Rufus is on track to add $10 billion in extra sales this year, with 250 million shoppers already using it and purchase rates jumping 60% when people engage with it. Meanwhile OpenAI quietly pulled back its own in-chat checkout feature just months after launching it, because checkout completed inside ChatGPT converted at roughly a third the rate of just sending shoppers to the retailer's own site. Same idea, wildly different results, and the gap tells you almost everything about where this technology actually works right now.
Why Amazon's Rufus is a genuine success story
Rufus is Amazon's built-in AI shopping assistant, and the numbers behind it are unusually concrete for an AI product. During Amazon's Q3 2025 earnings call, CEO Andy Jassy told investors that Rufus is expected to generate more than $10 billion in annual incremental sales, based on a seven-day rolling attribution model that tracks purchases following a Rufus interaction. Monthly active users grew 140% year over year, and interactions with the assistant were up 210%.
The reason Rufus works is not really about the AI model behind it. It is about where it sits. Rufus lives inside an app people already use to buy things, with saved payment methods, order history, and one-click purchasing already in place. When a shopper asks Rufus to compare two blenders or find a gift under a certain budget, the assistant is doing research inside a checkout flow that already exists and already works. It is answering a question, not trying to replace an entire buying process.
Why OpenAI's shopping agent stumbled
OpenAI launched Instant Checkout inside ChatGPT in late September 2025, built on an Agentic Commerce Protocol developed with Stripe, letting people buy from Etsy sellers and, eventually, Shopify merchants without leaving the chat. By March 2026, OpenAI had scaled the feature back. According to reporting from CNBC and Modern Retail, onboarding merchants turned out to be far harder than expected, with only around 30 Shopify sellers actually live on the feature months after launch, and purchases completed inside ChatGPT converted at roughly one third the rate of sending the same shoppers to finish checkout on a retailer's own site.
OpenAI is now shifting toward letting purchases happen through individual retailer apps connected to ChatGPT, such as Instacart, Target, Expedia, and Booking.com, rather than trying to complete every purchase natively inside the chat window. Google is taking a similar path with its Universal Commerce Protocol, rolling out agentic checkout with specific retailers like Wayfair, Chewy, and Quince rather than building one universal buy button. The lesson from both companies is the same: shoppers will let an AI assistant answer questions, but getting them to trust a chatbot with their card details, shipping address, and order accuracy is a much higher bar, and one that a general-purpose chat app has to earn from zero.
Are AI shopping assistants worth using for everyday shoppers
Setting aside the corporate horse race, the more useful question is what these tools are actually good for right now. Based on the available research, the honest answer is: research and comparison, not necessarily the final purchase.
- Survey data from IBM's Institute for Business Value found AI-assisted shopping has grown 62% globally over the past two years, with most of that use going toward research (45%), comparing products (41%), and reading reviews (33%), rather than letting the AI complete the transaction.
- Age plays a large role in adoption. Roughly 74% of shoppers aged 18 to 29 say they have used AI to research a purchase, compared with about 20% of shoppers over 61, according to YouGov polling.
- Even among people who use these tools, trust is thin. Only around 17% of consumers say they trust AI product recommendations without double-checking them elsewhere.
That pattern, heavy use for research paired with low trust in the final recommendation, suggests most people are treating AI shopping assistants the way they might treat a knowledgeable friend rather than a financial decision-maker. It is useful for narrowing down ten options to three. It is not yet something most people want handling the actual payment.
The privacy tension nobody has fully solved
The more personalized these tools get, the more uneasy people seem to feel about them. Multiple 2026 surveys point to the same discomfort: a large share of consumers, some studies put it as high as 73%, say they feel uneasy about how AI might use their personal shopping data, and a meaningful minority, around 30% in one survey, say they would never let an AI handle a purchase or hold their payment information at all.
Part of the problem is that personalization and surveillance look identical from the outside. A recommendation that feels helpful when it is slightly right can feel invasive when it is exactly right, especially if a shopper does not remember giving that information to anyone. Retailers have also run into direct backlash over AI-driven dynamic pricing, where shoppers noticed different prices shown to different devices or accounts for the same product, which turns a personalization feature into something that looks a lot like being charged more because the system thinks you can afford it.
None of the major platforms have fully resolved this. The practical result is that AI shopping tools keep getting more capable while consumer trust in them stays roughly flat, which is an unusual gap for a technology this heavily promoted.
What actually changes for shoppers this year
The realistic shift in 2026 is not that AI takes over shopping. It is that AI becomes a bigger part of the research phase across more places at once, Amazon's own site, Google's AI Mode and Gemini app, and chat tools like ChatGPT and Perplexity, while actual checkout mostly stays where it already was, on the retailer's own site or app, at least for now. The agentic checkout experiments from OpenAI and Google are real and ongoing, but the current data suggests shoppers are not yet ready to hand over that last step, and the platforms are adjusting their products to match that reality rather than forcing it.
The takeaway
The real story here is not that AI shopping assistants are a success or a failure. It is that the version people actually trust looks less like a chatbot trying to be your entire shopping trip and more like a smarter search bar sitting inside a store you already use. Rufus works because Amazon did not ask people to trust a new relationship, it improved an old one. OpenAI's early struggles happened because it tried to build the trust and the transaction at the same time. If you are deciding whether to use one of these tools, the safest bet in 2026 is to let AI do the comparing and let yourself do the buying, at least until the trust numbers catch up with the capability.
