TL;DR: A year ago, AI-referred visitors to US retail sites converted to purchases 38% worse than regular human traffic, according to Adobe Analytics. Last month, they apparently converted 42% better—and generated 37% more revenue per visit. The data suggests consumers are increasingly outsourcing the miserable part of online shopping—wading through sponsored clutter and endless options—to AI. The catch is that AI shopping's actual execution is still shaky.

What happened: Adobe's analysis of more than 1 trillion visits to US retail sites found that AI-referred traffic rose 393% year over year in Q1—but the more telling number is what those visitors are worth. A year ago, a regular human visitor generated 128% more revenue per visit than an AI-referred one; now it's inverted. AI visitors spend 48% longer on site and view 13% more pages (Adobe sells AI e-commerce products, though, so take all this with a grain of salt). The likely reason for the flip: AI selects products for engaged shoppers before they click, so they arrive already knowing roughly what they want—and now more people are using AI as a shopping tool.

Cutting through the noise: More than signaling a love affair with AI shopping, these numbers might reflect what most of us already know: Online shopping today is a drag, and we’re drowning in a sea of options. AI companies have positioned chatbots and agents as personalized shopping assistants that can help narrow down choices. Google’s AI Overviews now surface product recommendations at the top, so you’re not scrolling through a million individual sites (and its AI Mode can even help you find in-stock products). Amazon claims over 250 million people used Rufus, its AI shopping assistant, in 2025 and last October launched Help Me Decide, a Rufus feature that picks one single best product for you. Per Wired, Walmart’s own AI chatbot Sparky is bringing in new customers at twice the rate of its search engine referrals.

Still imperfect: The actual rollout of AI shopping has had real stumbles. Walmart’s original ChatGPT-integrated checkout was pulled after six months because it converted three times worse than its own site. Walmart’s executive VP told Wired his theory: Shoppers didn’t like that they had to buy items individually inside the chatbot. And Adobe’s survey shows that while 39% of respondents have tried AI shopping (flat from a year ago), the share saying it improved their shopping experience fell from 92% to 85%.

Bottom line: Adobe's data suggests the funnel is already shifting—AI is routing higher-quality shoppers to retail sites even before the experience of buying through AI has been figured out. Consumers seem willing to use AI to find things, but they're not yet ready to hand it their credit card. And the real test is whether AI can actually execute on beating online shopping’s clunky status quo. —WK

AI 101: Why your chatbot sometimes loses the plot

Last week I gave you the basics—the three main platforms, how to access them, and a few settings worth knowing about. Today, I want to explain something that trips up a lot of people: why your AI chatbot sometimes loses track of what you told it earlier, forgets a detail mid-conversation, or confidently gives you an answer that's just... wrong.

Every conversation you have with an AI has a maximum capacity—think of it like a whiteboard that can only hold so much at once. Your messages, the AI's responses, and anything else in the conversation all take up space on that whiteboard. Once you approach the limit (called a context window) the AI starts losing track of what was written at the top. It can't scroll back up.

The tricky part is knowing when you've hit the limit. Most platforms won't tell you (they’ll just automatically condense everything), which is why it's worth getting into the habit of regularly asking the chatbot to summarize everything important that's been discussed. Try this prompt:

"Please create a detailed summary of our conversation so far, including all key details, decisions, and context, and save it as a markdown file."

Depending on your platform and account type, you’ll either be able to download the file directly or copy and paste the text and save it yourself. Once it gives you the summary, review it—you know what's important better than the AI does, so check that nothing crucial got left out. If something seems missing, ask it: "What did you leave out and why?" It'll explain its reasoning, which can help you catch any gaps before you save it.

If the context window fills up mid-conversation, you can drop the file back in to keep going without losing the important stuff. Or start a fresh chat and paste it in from the beginning—that gives the AI a clean whiteboard with all the context it needs, which can produce better results. You can also keep these files organized in a "project" within some platforms, so the context is automatically available whenever you start a new chat in that project—more on that next week.

Next Friday, we're launching a new weekly column going deeper on all of this—from prompting to specific use cases at work and in your personal life. Reply to this email with whatever you've been wondering about. —SM

If you ever get the creeping suspicion that an indie band is all over your For You Page because someone paid for it—don’t blame the special gummies just yet. Wired reported earlier this week that a marketing firm helped engineer the TikTok campaign that made Brooklyn rock band Geese such a social media sensation. It’s a tactic the marketing firm’s co-founders call “trend simulation.”

They’re also not alone—there’s a whole industry of digital marketing outfits that pay creators to post social content for musicians, brands, Twitch streamers, probably even your neighbor who’s running to be HOA president. These firms sometimes buy hundreds of iPhones—enough to get “treated like VIPs at Verizon”—to run accounts that generate hype for clients, per TechCrunch.

To be fair, it’s clear Geese has plenty of bona fide fans, whose reactions to the paid marketing revelation have ranged from a limp shrug to some very ruffled feathers. Its marketing firm, for its part, told Wired it's “vehemently opposed to the use of bot farms” and employs only “genuine music fans.” Going “viral” used to imply an organic spread of social contagion—the reality, it turns out, is that it’s sometimes a bunch of paid marketers telling the crowd to cheer. —WK

Chaos Brewing Meter: /5