Too Many Bots in the Kitchen? Williams Sonoma Bans AI Shoppers
San Francisco-based Williams Sonoma and Pottery Barn sent emails to customers this week, updating terms of service to bar customer AI agents from shopping on their websites without the company’s written permission.
The new terms, which apply across all brands under parent company Williams-Sonoma Inc., forbid customers from letting any “agent,” “assistant,” bot or AI-enabled software use their accounts or buy things on their behalf. Any automated tool that does interact with the sites must identify itself as automated. And a customer’s permission doesn’t count as the company’s: Williams Sonoma said it can block any third party “including where you have purported to authorize such access.”
The terms also make customers responsible for all activity on their accounts, “whether or not you authorized that activity.”
Williams-Sonoma Inc. did not respond to KQED’s request for comment in time for publication.
In August, a federal appeals court ruled in Amazon’s lawsuit against San Francisco-based Perplexity that the user, not the AI company, is the one accessing a website when an AI agent shops for them. That weakened retailers’ ability to use anti-hacking law against agents, but the court said companies could still set rules through their terms of service.
In plain English, the fight to control which AI agents get to guide our shopping decisions is on.
Loyola Marymount University law professor Lauren Willis said the new terms are likely legal, for now. She wants the law to require retailers to accept shoppers’ AI agents, and “a requirement that the agents act in the best interest of the consumer, rather than the provider of the agent.” She pointed to proposals from the European Law Institute to require businesses to permit consumers to use digital shopping assistants in their transactions with the business.
An analysis by the e-commerce marketing firm Productrise found Google’s AI offers up more expensive options than a regular Google search. Willis said that a growing number of AI chatbot developers like Meta and OpenAI are selling advertising, which creates a natural conflict of interest. “They’re able to track whether you’ve clicked on something that they’ve given you and therefore can be biased by that,” she said.
Some research has shown that disclosing a conflict of interest can backfire, leading consumers to trust the adviser more, not less.
“What’s the business case for blocking agentic AI?” Eric Goldman, associate dean for research and professor at the Santa Clara University School of Law, told KQED. When websites don’t explain themselves, he said, blocking “often looks just like an overreaction to AI.”
That’s despite the fact that the Williams Sonoma group uses AI on its website. The Williams Sonoma chatbot introduced itself to this reporter as “Olive, your AI Sous Chef.”
Goldman said AI agents can serve consumers, retailers or only the companies that make them, and it isn’t always easy to tell which. But “in general, we should be encouraging consumers to find the tools that work best for them and to let them use them,” he said.
“The battle over what consumers can use to access a website is almost as old as commercial internet,” Goldman said. Around 2000, websites battled Third Voice, a tool that let users post comments layered on top of any site. Two decades ago, retailers fought adware pop-ups that displayed competitors’ offers on their pages. More recently, critics have argued the Honey browser extension siphoned affiliate commissions away from websites.
Goldman thinks retailers will eventually lose the ability to single out AI agents at all. “Many of the things that might be considered to be problematic about agentic AI ultimately will likely be built into a standard web browser,” he said. Blocking “only works as long as it’s really separate from the standard technology users are using.
“And I think that’s going to collapse over time.”