Better AI is now a reason to switch shops

Back in March I spent an evening on the 50th floor of 8 Bishopsgate, watching Vercel's people and a few good guests make the case that the web is quietly reorganising itself around agents instead of people. Tom Occhino's line that night was that UIs are becoming leaf nodes and the agent is the new trunk. It was a clarifying argument, and like a lot of clarifying arguments delivered over warm canapés, it arrived with no numbers attached. You nod along, admire the view, and go home wondering how much of it is actually true yet.

Vercel has now gone and found some numbers. Together with the World Retail Congress it surveyed 4,000 European shoppers, 1,350 of them here in the UK, and the result lands square on the thing that room was gesturing at in the spring. I should say up front that I deploy on Vercel for my own projects and am not especially sorry about it, and that this is Vercel's own research, so the eyebrow goes up before the notebook comes out. But the headline UK figure is a hard one to wave away.

43% of UK shoppers say they would choose one retailer over another because it makes better use of AI on its website. That reframes AI from a nice-to-have on the product page into something closer to a reason to leave. British shoppers are, if anything, mid-table on it: the same question gets 59% in southern Europe, 40% in Germany and 39% across the Nordics, according to the report. But almost half your customers willing to shop elsewhere over the quality of your AI is not a rounding error, and it is a very different animal from the usual "customers love our new chatbot" telemetry that lands on my desk.

For a Head of Technology that is a budget-line argument, not a customer-experience one. Churn has a cost you can put in a spreadsheet. "Delight" doesn't.

Discovery moved. Buying mostly didn't

Here is the other half of the report, and the reason I would not sprint to the board with the 43% and ask for a war chest. AI has changed how Britain finds things far more than how it buys them. One in five UK shoppers now start a shopping journey on an AI search tool like ChatGPT or Gemini. On the report's numbers that puts AI search ahead of the social feeds as a starting point, with Facebook on 19%, TikTok on 18% and Instagram trailing on 14%.

AI search (ChatGPT, Gemini)

20%

Facebook

19%

TikTok

18%

Instagram

14%

That is a real shift and retailers should take it literally. Whether it keeps climbing from here I honestly don't know; the report is a snapshot, not a trend line. But watch what happens as you move down the funnel. 87% of UK shoppers find AI recommendations useful and 55% have bought something on the back of one, and yet 79% still begin on the retailer's own website, 65% still want to stand in a shop and look at anything expensive before they pay for it, and only about one in ten would rather spend with an AI assistant than a human. The agents are running the research desk, not the till.

If that pattern sounds familiar, it is because it is almost exactly what CI&T found in its UK research this summer: plenty of agent use, nearly all of it to compare and shortlist rather than to hand over the wallet. We spent a whole piece on whether shoppers will actually let AI buy things for them, and the Vercel figures are another brick in the same wall. Adoption is real at the top of the funnel and thins out fast the closer you get to the money.

Brand.com is back, if it's any good

So what do you do with this. Vercel's answer, via CTO Malte Ubl, is the bit where the vendor's interest and the reader's interest happen to line up, so take it with the appropriate pinch of salt: "For most of the web's history, we optimised storefronts for two audiences: people and Google. There's now a third: AI agents that discover, compare, and increasingly recommend products on a shopper's behalf." In the report's foreword he goes further and declares that brand.com is back, on the argument that agents tend to walk straight up to a brand's own site rather than route through a marketplace listing.

Strip the salesmanship out and there is a genuine engineering point in there. I started out doing print artwork and packaging for food and drink brands, the kind that live on supermarket shelves, and the job then was making a pack legible at arm's length under bad lighting. The job now is making a product page legible to a machine that never actually renders it: clean structured data, honest specs, prices and stock it can read without guessing. We have written before about making the catalogue legible to agents, and this shopper survey is really just the demand-side version of that same argument.

And the unglamorous truth sitting underneath "brand.com is back" is that it only comes back for sites that are quick. An agent comparing ten retailers has even less patience than a human on a 14.4k modem ever had for a page that won't load; it will pick the shop that answers cleanly and move on. Vercel names Frasers Group among the UK retailers already on its stack, alongside the likes of Diageo and Breitling, which at least means the fast, agent-readable storefront pitch isn't purely theoretical on this side of the Channel.

Worth being honest about where we are

This is a vendor's survey, commissioned by a company that sells the exact fast storefront it concludes you need, and I would be doing you a disservice not to say so twice. The trust figures underneath the headline are their own caution: a third of UK shoppers already suspect the AI isn't showing them the best products, more than a third won't share their data, and nearly half would still rather deal with a person than a chatbot. (For the avoidance of doubt, that is the Vercel report's own trust data. A separate Braze study doing the rounds this month carries a coincidentally identical 43% about data-sharing; different survey, different question, easy to mix up.)

None of that says ignore it. It says the 43% is a real signal about where loyalty starts to leak, and the rest of the report is a reminder that the leak is at the discovery end, not the checkout.

If I were spending money on the back of this, I would not buy a shopper-facing agent and call it a strategy. I would make my own storefront fast, structured and legible to whatever turns up asking, because that pays off whether the thing asking is a shopper's agent, Google's crawler, or a person on a train with ten minutes to kill. The flashy version of AI commerce is still mostly waiting on shoppers who keep saying, politely, that they would rather browse than delegate. The boring version, the one that just makes your site answer quickly and honestly, is the one that 43% of them are already quietly rewarding. Worth not waiting for a press release to tell you which of the two to build first.

---

Data sources and further reading

• The Agent Era: How 4,000 European shoppers are shopping differently. Vercel and the World Retail Congress, 2026 (survey of 4,000 European shoppers, 1,350 in the UK, fielded May 2026). Vercel-commissioned; the report was supplied to us directly and has no public URL at the time of writing, so the figures below are cited to the trade coverage that carries them.
• AI recommendations reshape UK shopping but not spending. eCommerceNews UK / IT Brief, 30 July 2026 (the 43% switching figure, the useful/bought split, spend and trust numbers, and the Malte Ubl quote).
• New study reveals how AI is reshaping how people shop online. TechRadar Pro, 1 August 2026 (the discovery-source split of AI 20% / Facebook 19% / TikTok 18% / Instagram 14%, the 79% start-on-site and 65% in-store figures).
• Related reading on this site: Notes from Vercel London, will consumers let AI buy for them, CI&T's UK agent research, and making the catalogue legible to agents.