Getty Images/Malte Mueller
Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- AI agents can place orders for you and find deals, saving time.
- But experts wonder whether the risks outweigh the benefits.
- AI agents can also impact retailer traffic.
In late October, OpenAI launched ChatGPT Atlas, a web browser with ChatGPT at its core. In Agent Mode, it can perform actions on your behalf, such as pulling together an online order. Elyse Betters Picaro, Senior Contributing Editor at ZDNET, tried it that day, successfully using it to place a same-day delivery order from Walmart. The experience was so positive that she then used it to buy Disney on Ice Tickets, which she said saved her some money and hassle.
Also: Use an AI browser? 5 ways to protect yourself from prompt injections – before it’s too late
“It found me a deal on front row seats for $50 each. But what was really cool is that it flagged that the original showtime I wanted was French only and suggested another time in English,” said Betters Picaro. “I would’ve missed that if I bought them myself.” She added that given her successful experiences, she may use it for the holiday shopping season.
Betters Picaro isn’t alone: the Shopper Snapshot report from Mastercard found that 42% of shoppers are already using AI tools to assist with gift buying this holiday season, with a majority of Gen Z (61%) and millennials (57%) among them.
At the same time, experts are concerned about several risks associated with these tools. Should you give them a try? Here’s what to know.
AI is driving shopping
New data from Salesforce, first reported by TechCrunch, showed that between Thanksgiving and Black Friday this year, AI and AI agents influenced $22 billion in global sales. Adobe found that AI-driven e-commerce traffic for November shows a 770% increase YoY from the 2024 holidays.
Betters Picaro’s successful test helps explain the appeal: after some brief setup and prompt perfection, Agent Mode used her Walmart purchase history to quickly add her preferred brands of wood putty, paintable caulk, and 2-inch screws to her cart, which she then just had to confirm payment for and received within the hour.
In addition to using AI agents to place orders, people are also utilizing AI tools for product research, which helps them decide which product first meets their specific criteria, including the best deals and prices.
Also: Amazon puts ChatGPT on the naughty list, blocking shopping access – what we know
“In both cases, it’s not about replacing how people shop, but enhancing it through personalization and time savings,” Pablo Fourez, Mastercard Chief Digital Officer, told ZDNET.
OpenAI just launched a shopping research feature designed to act as an assistant that can create a personalized shopping guide tailored to your specific criteria in just a few minutes. Most AI chatbots, including Google’s Gemini and Anthropic’s Claude, can also help identify products that match your criteria based on a conversational search query, acting both like a personal shopper for the user and a referral site for the retailer.
Also: I’ve been testing the top AI browsers – here’s which ones actually impressed me
Greg Zakowicz, a digital marketing consultant, utilizes Rufus, Amazon’s AI shopping assistant, to track a product’s price history and help evaluate whether a sale is actually a worthwhile deal. In this way, AI tools can shape customer behavior beyond just making purchases for them by further informing their shopping decisions.
Zakowicz has also utilized AI chatbots like ChatGPT to assist him in gift shopping. His wife’s birthday is coming up, and she mentioned wanting a cappuccino machine months ago — he’s a black coffee drinker with no real idea of what she might want, so he asked the chatbot for specific criteria, including a budget, ease of cleaning, compatibility with at-home ground coffee beans, and as little waste as possible.
“I know nothing about cappuccino machines, but I’m going through and using all these AI tools and Amazon, looking at like, ‘Hey, help me find a freaking cappuccino machine within my budget that has these criteria’,” Zakowicz told ZDNET.
Downsides and blind spots
Despite using AI to inform his shopping decisions, Zakowicz has not let an AI tool build an order for him the way Betters Picaro did. He said he doesn’t want to relinquish the control that placing a manual order at checkout offers him, including double-checking that he’s entered the right credit card and shipping address, and is using credit card points where available.
While AI tools can’t fully click “confirm” on an order for you yet, it’s unclear if ChatGPT’s Instant Checkout feature, for example, obscures or automates over some of these details Zakowicz prefers to handle himself. His hesitation highlights a simple yet major challenge of using AI tools for shopping: missing out on opportunities that the agent may not have visibility into.
Also: Why Amazon really doesn’t want Perplexity’s AI browser shopping for you
“The disadvantage might be that shoppers could miss some things that are not cleanly shared with the agents — be it unique inventory that a retailer holds back, special loyalty benefits, or value-added services like installation that the retailer may not let go through the agent,” Aaron Cheris, a partner in Bain & Company’s retail, private equity and customer strategy and marketing practices, told ZDNET.
Risks for consumers
Of course, using autonomous AI tools that can facilitate orders on behalf of users also presents several risks. The space is nascent; not only are true agentic experiences rare to begin with, but all automated shopping services come with security issues.
“The key change has been the introduction of automated buying from chatbots and ChatGPT-driven sites, but these are not really AI Agents, but piggybacked payment engines,” said Ananda “Andy” Chakravarty, VP Research, IDC Retail Insights. “Some immediate risks are fraud, bad actors, limited capabilities, and untested shopping algorithms.”
Chakravarty added that these risks “will be ironed out over time,” but that “we’re not quite ready for autonomous agentic shopping just yet.”
The process of having an AI chatbot create an order for you involves handing over access to sensitive information, such as site logins, payment information, address, and more. Many credit card companies have begun laying the groundwork for agent-based transactions by implementing new protocols, frameworks, and measures designed to improve security for individual users and enterprises.
Also: I let Google’s new AI tools take on my holiday shopping list – here’s how it went
For example, Google launched its Agent Payments Protocol (AP2), which the company says is an “open protocol developed with leading payments and technology companies to securely initiate and transact agent-led payments across platforms.” Similarly, the Visa Intelligent Commerce initiative opens Visa’s payment network to developers and engineers building agent-based AI shopping experiences.
So should you use AI tools for shopping? The tools do have the potential to save people money and time, and ultimately, the amount of risk is equivalent to which tool the user picks, with AI agents posing the greatest risk for vulnerabilities.
“Consumers can absolutely find value and save time by using AI agents as part of their shopping process,” added Cheris. “But they will likely still want to hold onto the final approval step before fully handing over the keys to the agent.”
Consequences for retailers
The risks of AI shopping tools also extend beyond the consumer experience, affecting retailers as well. AI agents summarize information for a consumer, meaning that websites dedicated to facilitating shopping research for them (or the actual retailers themselves) may receive fewer page views during a transaction.
Also: How small businesses can survive AI shopping: 7 essential steps
“Shoppers may be less likely to scroll and monitor deal pages because they can ask AI agents to provide deals for the items they’re searching for,” Hannah Donoghue, head of consulting at Flywheel, told ZDNET. “Additionally, Rufus’ price history functionality and Amazon’s AI ‘help me decide’ feature may significantly impact conversion, with shoppers leveraging these tools to evaluate the actual ‘value’ of the ‘deal,’ or to outsource decision making.”
These AI tools present information to users by scraping the web. Some sites have chosen to block these crawlers, opting to keep their information protected from summarizing bots (and AI training) and therefore exclusive to shoppers who visit their sites directly.
However, Tomer Elias, Sr. Director of Product at HUMAN Security — which protects against bots and recently released a report on how Americans are using AI to holiday shop — noted that some AI companies are finding ways to circumvent these boundaries.
Also: AI’s free web scraping days may be over, thanks to this new licensing protocol
“The proliferation of AI crawlers and scrapers presents a new challenge for website owners and security teams,” said Elias. “Some of our recent research revealed that unauthorized scrapers and other malicious actors use AI crawler user agents to disguise their activity and bypass anti-bot measures.” Similarly, a March study from Columbia on chatbots and news found that several chatbots could still retrieve articles from publishers that had blocked their crawlers.
Of course, this is on a company and tool-by-tool basis. For example, OpenAI explicitly stated that it respects these boundaries and doesn’t populate sites on its shopping research tool that have blocked its crawlers. Regardless, this summer, Cloudflare started blocking AI web crawlers by default unless they have permission from the site itself or are compensating that site.
