Chatbot makers and large retailers are clashing over user information as both sides build systems meant to let AI agents complete online purchases on behalf of shoppers.
Ask OpenAI’s ChatGPT about an item on Etsy and you may be able to type in payment details and buy it without leaving the chat. Instant Checkout was among the first features to appear in a recent wave of deals between major AI firms and ecommerce platforms. The idea is to encourage people to hand off portions of browsing and ordering to automated assistants and push toward more agent-driven shopping. These agents are appearing in more places, though they are still far from acting as full-time virtual buyers.
Executives at seven tech and ecommerce companies who spoke with reporters say OpenAI, Google, Amazon and others are still working through how to limit costly agent mistakes and how much product data and chat history must move between partners for these systems to work well. That means many of the tools available now need heavy user input, move slowly, or handle only a small set of items. With talks and pilots still under way, shoppers hoping to hand off holiday purchases to automation may be disappointed.
“I haven’t yet felt a super magical agentic experience in commerce,” says Talia Goldberg, a partner at the venture capital firm Bessemer, who has invested in AI companies including the search and browser startup Perplexity and the coding platform Cursor. “There are big questions that have to be solved around a true functional experience.”
Recent surveys of U.S. consumers suggest interest in assistance from AI is strong: 60 percent say they plan to use AI to help with shopping, 20 percent would let an AI agent fully handle everyday purchases, and just 25 percent say they would prefer to shop without any AI help. Long-range forecasts are optimistic. McKinsey projects agent-enabled commerce could generate as much as $1 trillion in sales in the U.S. by 2030.
To try to make that projection real, OpenAI has a partnership with Walmart that will let ChatGPT users buy Walmart items inside the chat window. OpenAI and Perplexity announced arrangements with PayPal and Shopify, the platform that hosts many online stores. Google recently rolled out AI agents that can populate checkout forms and place calls to stores to ask about pricing.
Some experimental integrations are producing promising outcomes. Expedia’s ChatGPT app returns live airfare and hotel pricing when users ask for it. Customers still have to complete bookings themselves; agents are not checking out for them yet. The result has nevertheless boosted sales more than Expedia expected. “That means there's something in these tools that works,” says Clayton Nelson, a vice president who oversees Expedia’s strategic alliances with AI giants.
Social commerce through apps such as TikTok and Instagram has not taken off in the U.S. to the degree seen in other markets, in part because consumers and major retailers remain wary of tech platforms. Payment processors like Visa and startups that help shops build or partner with chatbots are trying to broker terms that balance security, privacy and convenience. “We do think that a service provider like us will be faster to earn the trust of retailers, which is pretty important,” says New Generation CEO Adam Behrens.
Retailers are keen to join these efforts because chatbots are already a key part of how shoppers research and validate purchases. Agreements between AI firms and merchants could make sure assistants deliver accurate product details and use less compute when completing orders. Those efficiencies could raise profits for both sides if a workable commercial model is agreed.
Amazon’s CEO Andy Jassy offered one of the bluntest assessments yet of how agent-driven shopping performs on other platforms. “I would say the customer experience is not good,” Jassy said on an earnings call last month. “There’s no personalization. There’s no shopping history. The delivery estimates are frequently wrong. The prices are often wrong. We have got to find a way to make the customer experience better and have the right exchange of value. ”
Practical problems show up in everyday tests. In a recent trial, Opera’s browser AI agent took 45 seconds to add a carton of eggs to an Amazon cart. Doing the same task manually in Amazon’s app took a third of the time. Opera has invited possible partners to workshops to provide input on security and design choices. “If our agent doesn’t work with the biggest websites people go to, it will be a suboptimal experience,” says Per Wetterdal, an executive vice president who leads Opera’s commercial partnerships. “No one benefits if [a purchase] is ending up at the wrong place or in the wrong quantity.”
Money and data lie at the center of ongoing talks. If an AI service actually helps drive sales, companies such as Opera expect to be paid for generating incremental purchases. “If we do something that adds incrementality, it’s very fair to be compensated for that,” Wetterdal says. OpenAI has signaled one approach by taking what it calls “a small fee” from partners such as Etsy for purchases placed through Instant Checkout.
Data-sharing arrangements complicate potential deals. Retailers tightly guard pricing, inventory and customer records to keep an edge over competitors. AI companies want to protect chat histories so conversations with users feel private and personalized. At the same time, assistants require real-time store data to act on orders, and merchants prefer more shopper context to build lasting relationships.
OpenAI’s apps feature provides partners such as Expedia the user’s IP address and their relevant chat queries, according to a permission screen on ChatGPT. Nelson says he is content with that initial trade but would like deeper access if users agree to it. “I want to know the full conversation,” he says. “I know they're looking for a hotel room in Las Vegas right now for two guests, but I want to know, are the guests friends? Have they traveled before? Do they have other things that they like?”
Negotiations have turned tense at times. This month Amazon sued Perplexity, alleging the startup used AI agents to make purchases on customers’ behalf in ways that interfered with the ecommerce giant’s advertising and Prime subscription businesses. Perplexity said it will fight back.
Amazon is building its own version of an assistant, too. The company has tested a feature called Buy for Me that lets its agent complete purchases on third-party sites when Amazon lacks stock of a desired item. The tool adds goods to a cart automatically; shoppers check out with their Amazon payment details. Third-party sellers do not receive shoppers’ real email addresses and may contact Amazon to block the assistant if they prefer.
Many companies are keeping performance data private. An executive responsible for a large apparel retailer’s online operations in California says their business is eager to cut deals because bots are sending a lot of traffic to product pages, but the commercial offers so far feel underdeveloped. “Up to today, no one has a solid solution,” says the executive, who wasn’t authorized by their employer to speak to the media. “Everyone is just making marketing announcements.”
Some smaller developers are stepping back from agent shopping partnerships for now. Archit Karandikar, CEO of CoInvent AI, which builds the travel planning chatbot Airial, points to the basic challenge of getting AI to produce genuinely useful suggestions. Pursuing automatic purchases would be too much on top of that given current limits. “You can’t be spending someone’s money without being sure you’re making the right transactions,” Kandikar says. Airial instead links users to booking sites and takes a commission when a customer completes a purchase.
Expedia’s Nelson summarizes the largest technical risk in blunt terms. “My goodness, no one wants to mess up their vacation for their entire family because a bot went left instead of right, or didn't follow the specific prompt that was given,” he says. “It's up to us and our partner to make sure that we never leave travelers astray. And so that's the big thing that's holding us back on fully agentic booking experiences.”
This holiday season some shoppers will use chatbots to pick gifts, put items in carts and, in certain cases, complete checkout instantly. For now, people remain the final decision-makers. Maybe next year they can blame the bots if a gift misses the mark.