In its initial responses, ChatGPT didn’t provide any links to products. But it easily supplied them when I asked, and while I didn’t click on every single one, none appeared to be a hallucination. Claude, on the other hand, apologized and said that it “cannot actually link to websites or products directly.” Anthropic hasn’t released a web search feature for Claude yet, but the company says it’s working on it.
That technically made Claude the least useful chatbot I tested for shopping. But it also means that Anthropic has so far avoided wading into the ethically murky territory of allowing its AI chatbots to scrape human-written product reviews from the web. Instead, Claude bases its product comparisons on its existing data set. Perplexity, on the other hand, says that thanks to Buy with Pro, people “no longer have to scroll through countless product reviews.”
When I asked Perplexity what I should get for my editor-musician friend, it recommended a solar bike light set (I had also noted he was a cyclist). It was a good idea, but not exactly a milestone-birthday-worthy gift. I kept tweaking my prompt. What about a personalized leather guitar strap? Down the rabbit hole I went.
Perplexity’s goal in hyping up its shopping features, I was beginning to understand, wasn’t just to help me brainstorm fresh ideas or come up with supremely thoughtful gifts. Perplexity is playing the long game, slowly siphoning our attention away from competing corners of the web, gaining a better understanding of how people like me are using its platform, and funneling that data into its ever-evolving AI models. Each time I needed to refine my searches because the initial results were often lacking, I remained in Perplexity’s app, which meant I was not on Amazon and not on Google, though I ended up on both of those sites eventually. Perplexity Pro is not a full-fledged ecommerce site, nor is it “agentic” in any real way yet, but I am one of millions of people supplying the information it needs to become those things.
When I turned to Google’s Gemini, I found the gifts it suggested for my 16-year-old niece weren’t bad, per se, just uncreative and in one instance confusing. It said I should buy her a “cat blanket for snuggling up with a good book”; it wasn’t clear if the blanket was for her or her cat. A Kindle was a fine idea. But I’m terrified of what she would text me if I sent her the SAT prep book Gemini suggested (probably “thx,” and nothing else). The app’s ideas for my editor-musician friend were equally uninspiring, among them “vinyl records” and “high-quality headphones.”
I was using the year-old version of Gemini, but earlier this month, Google started rolling out a newer version, Gemini 2.0, to developers and limited testers. The new AI model will “think multiple steps ahead and take action on your behalf,” the company says. For now, this means taking action on behalf of developers—executing the next step in their coding workflows—but I’m eagerly awaiting the day it can plow through my shopping list.
ChatGPT eventually led me to an online spice store where I bought a few specialty baking ingredients for my friend, who at this point I had built up in my mind to be a finalist in The Great British Bake-Off. In the end, I chatted with the AI bots for so long that many of the gifts I picked won’t arrive until after Christmas. My niece will be getting cash in a card. My search for a friend’s milestone birthday gift was inconclusive. I decided to kick the task down the road until January, a month full of newness and agentic resolve.