Your customers are deciding before they reach your site.

Ecommerce has spent twenty years optimising the wrong moment.

The whole machine is built around the visit. Get them to the site, then convert. Traffic in, sales out. Every dashboard, every agency report, every Monday meeting starts from the same assumption: the decision happens on your website, so the job is to win the website. Better landing page. Faster checkout. Tighter funnel. The customer arrives undecided, and you persuade them.

That assumption is quietly dying, and most retailers have not noticed.

Stord’s 2026 ecommerce report puts a number on it. The share of consumers using generative AI to shop online went from 38% in 2024 to 51% in 2025. Half of shoppers are now asking an AI to help them buy. They are using it to shortlist, to compare, to hunt for the better price, and increasingly to decide. By the time someone lands on your product page, the AI has already done the comparison. The page is not where the decision gets made anymore. It is where the decision gets confirmed.

Stord also found that 20% of consumers are more likely to convert when a product or store is recommended by AI. Sit with that. A fifth of buyers will trust the machine’s pick over their own browsing. The recommendation is becoming the sale.

So here is the problem with measuring traffic. If more of the buying decision happens before the click, then traffic is a lagging measure of a contest you already lost or won somewhere else. You can have a beautiful site, a fast checkout and a conversion rate you are proud of, and still be slowly starved, because fewer of the right people are being sent your way in the first place. Your funnel looks healthy. The top of it is being quietly rerouted.

This is what people mean by zero-click buying, and it is not a conference slide anymore. Discovery, comparison, recommendation, and in some cases checkout, all compress into a conversation that never touches your homepage. The customer asks. The agent answers. A few candidates make the shortlist. The rest are invisible, and invisible is not a ranking problem you can see in your analytics. It is an absence. You cannot report on the sale you were never considered for.

So the real question is not “how do we get more traffic.” It is “when an AI is choosing what to recommend, are we one of the options, and can it trust us enough to say our name.”

That second part is where the work actually is.

A machine recommending a product is taking a small risk every time. If it suggests something that is out of stock, mispriced, slow to ship or returned in a huff, that reflects on the AI, not just on you. So these systems do the sensible thing. They favour the option they can verify. Clean data is not a nice-to-have to an agent. It is the difference between being a safe recommendation and being the one it quietly skips because your facts did not line up.

That is why the asset worth building is not more content. It is what I have started calling a product truth layer. One consistent, verifiable version of what you sell, saying the same thing in every place a machine can read it. The feed. The structured data. The page. The marketplace listing. The reviews. The price, the VAT, the stock, the delivery promise, the returns policy, the specs, the compatibility claims. When all of those agree, an agent can recommend you without taking a risk. When they contradict each other, and in most catalogues they do, the machine picks the competitor whose story holds together.

None of this is the version of AI that gets applause. There is no agent with a personality. There is no clever file you buy that fixes it overnight, whatever the AEO crowd is selling this month. The Pattern UK research lands on the same unglamorous truth from the operator side: 57% of ecommerce businesses are already exploring AI agent use cases and 33% are actively preparing to deploy, and the ones getting ahead are treating it as plumbing, not a campaign. Pattern also reports that 76% of ecommerce organisations have already cut customer acquisition costs as AI search and shopping agents reshape discovery. The savings are real, and they are going to the businesses whose data a machine can act on.

If you sell online, here is the move. Stop reporting traffic as if it is the scoreboard. Start asking whether you are eligible to be recommended at all. Pull twenty of your best products and check the boring fields side by side across your feed, your schema, your page and your marketplace listings. Price, stock, GTIN, category, delivery, returns. I would put money on you finding contradictions in the first ten, and on nobody currently owning the job of fixing them end to end.

The brands that win the next few years of ecommerce will not be the ones with the loudest AI claims or the biggest ad budgets. They will be the ones a machine can recommend without flinching, because their product truth is clean enough to bet on.

The decision is moving upstream of your website. Go and win it there.

---

Jason Sibley is the founder of Cleo, a post-agency marketing and AI company. JasonVsTheNoise is where he writes about what is actually happening with AI, marketing, and how businesses should be thinking about both.