Tell me which AI you use and I’ll tell you what you buy

Buying a jacket used to involve choosing one. Finding it in a shop, searching for it on Google, or spotting it on someone in the street. Now a message can be enough for someone to find it, compare the options and buy it for us. The curious thing is that someone doesn't even have to be a person.

Instinct, an AI assistant you can message on WhatsApp, can already do just that. It searches for products, browses online stores and completes purchases with your approval. Stripe has even developed a digital wallet that allows agents like Instinct and Muse, Meta's personal AI assistant, to pay on our behalf. We've gone from asking Google where to buy something to asking a machine to take care of the whole thing.

Not everyone is equally pleased about these new shoppers. Amazon blocked Muse from its store in September, citing unauthorized access. Walmart, meanwhile, is making it easier for third-party agents to shop on its platform. And Google has introduced the Universal Commerce Protocol, a standard designed to help AI assistants, retailers and payment providers work together. There's something rather curious about companies that built much of their business around our searches now preparing for a world in which we might no longer need to search at all.

But something else interests me even more: these machines don't always make the same choices. In August, researchers at Wharton ran around 26,000 shopping experiments using six AI models. They found that recommendations varied between models and that, under certain conditions, simply changing the order of the sources could change the outcome. Machines don't have personal taste, of course. But they don't need it to end up favouring certain brands over others.

For brands, this changes quite a lot. For years, they've invested in being found on Google, standing out on Amazon and making sure we remember their names. Now they'll also need to understand how machines interpret their products, which information they rely on and why one assistant recommends something another overlooks. Because there's a big difference between losing a sale to a competitor and never making it onto the shortlist in the first place.

I wonder whether, over time, we'll start adopting our assistants' preferences as our own. It wouldn't be entirely new. Much of what we like came from someone else: a borrowed book, a song a friend introduced us to, a beautiful coat we spotted on a stranger. Perhaps the difference is that this time, we won't always know where the recommendation came from.

Maybe one day I'll walk out wearing a jacket my AI picked for me. Someone will tell me I have great taste. I'll smile, say thank you, and wonder: whose taste am I wearing?