As someone who spent years writing about e-commerce before I started covering AI, I’ve been fascinated by how AI shopping has evolved. And I have a personal stake in it: I hate shopping. So I couldn’t wait to see whether AI could make the experience better, and I’ve been experimenting with shopping in ChatGPT ever since.
That’s why I was so keen to have my former Fortune colleague Jason Del Rey on the Ground Level AI Podcast. Now the founder of The Aisle, Jason has been covering retail and e-commerce for well over a decade and is the author of Winner Sells All: Amazon, Walmart, and the Battle for Our Wallets. He is my go-to for all things AI and shopping.
For those who prefer to read, here are some highlights from our conversation:
AI shopping is having a reality check
One of the biggest themes of our conversation was just how much the expectations around AI shopping have shifted over the past year. Jason launched The Aisle in part because retail executives, e-commerce leaders and people inside AI labs were convinced AI would bring the biggest disruption to online shopping since mobile commerce—or perhaps even Amazon Prime.
But nine months later, the picture looks more complicated. Consumers really are turning to ChatGPT and other AI tools for product research and discovery, Jason said. What’s less clear is whether they want to hand over the actual decision-making and purchasing to agents.
“The last nine months has kind of been, I think, a wake-up call for a lot of both AI labs, but also retailers, about what consumers do and do want right now from AI.”
It turns out e-commerce was already pretty convenient
We also talked about something that gets lost in the race toward agentic commerce: buying things online isn’t actually all that difficult. Tech companies have been trying for years to persuade consumers to complete purchases somewhere other than a retailer’s own site—from Twitter’s old buy button to various efforts by Google, Instagram and Facebook—and largely failed.
Jason thinks AI companies are running into some of the same realities. Product data and real-time inventory are messy. Consumers have to trust a new checkout experience. And after using ChatGPT to research something, clicking through to a retailer designed specifically to sell it isn’t exactly an enormous burden.
“There’s a lot of talk about removing the friction of e-commerce through AI. I could sit in my pajamas on my couch and buy something from across the world. That’s pretty darn easy.”
And there’s another problem: shopping can be visual and entertaining. A chatbot full of text isn’t necessarily an improvement for people who actually enjoy browsing.
Retailers discovered the AI labs didn’t necessarily understand retail
Jason said large retailers publicly remain eager to experiment with the AI labs, particularly because consumers increasingly use LLMs to discover products and brands. They don’t want to repeat the mistake some retailers made by taking e-commerce too lightly in its early years.
But he added that retailers privately have described more frustration:
“We thought the LLM apps—and a lot of this is aimed at, like, OpenAI—we thought they got how hard e-commerce is, and they clearly didn’t.”
That includes challenges as basic—and difficult—as surfacing the right product at the right price and knowing what’s actually in stock.
Who gets to decide where your AI agent can shop?
Then there’s a question that goes well beyond retail. Jason walked me through the fight between Amazon and Perplexity over Comet, Perplexity’s browser agent, which was designed to navigate Amazon and shop on a user’s behalf. Amazon objected and ultimately sued Perplexity after it continued finding ways for the agent to access Amazon’s systems.
For Jason, the dispute raises a much larger issue for the agentic web:
“If I want my agent to shop on my behalf, should I be able to do that whether or not the retailer or site on the other end [agrees]?”
What Jason is watching next
For all the ways AI shopping has run into reality, Jason is far from bearish on what comes next. When I asked what he was genuinely excited about, we both immediately landed on a technology that e-commerce has been promising for years: virtual fitting rooms.
Virtual try-on has never quite lived up to its promise, in part because the technology has historically been expensive and difficult to deploy at scale. But Jason said there are companies now working on the actual fitting problem.
He’s also watching Shopify, which he thinks may be particularly well positioned for an AI-driven shift in product discovery. Shopify has co-developed the Universal Commerce Protocol with Google, designed to make it easier for retailers’ products to surface—and eventually be purchased—across AI apps and other platforms outside traditional e-commerce sites. And that could give smaller, independent brands a better chance of being discovered.
Finally, there’s AEO/GEO, the fast-growing effort to help brands get recommended and cited by ChatGPT and other AI services. Jason is watching the space closely as venture money pours in and retailers look for new sources of traffic. But he’s also skeptical about how much of the emerging optimization industry will prove durable as the underlying models keep changing.
A special offer for Ground Level AI readers: Jason is offering GLAI readers and listeners an extended free trial of The Aisle, his publication covering AI, commerce and consumer behavior. Try The Aisle here.
