Remember when AI search traffic was the ecommerce equivalent of window shoppers who forgot their wallets? Visitors who'd browse extensively, read every product description with genuine enthusiasm, and then vanish like they'd witnessed a crime? That was eighteen months ago. The data has since done something rather inconvenient for anyone who used it as an excuse to postpone their AI search strategy.

It flipped. Completely.

Here's what happened, and I genuinely need you to sit with these numbers for a second. In July 2024, Adobe's analysis of over a trillion visits to US retail sites showed that shoppers arriving from AI chatbots were roughly 43% less likely to buy than people coming through conventional channels. Search, email, social — all of them outperformed AI referrals by a comfortable margin. The traffic looked enthusiastic but fundamentally unserious, like someone who shows up to every open house but has no intention of buying property.

By the 2025 holiday season, those same AI-referred shoppers were converting around 31% more than conventional traffic. By Q1 2026, the premium had widened to roughly 42% better per visit. If you're keeping score at home (and if you sell things online, you absolutely should be), that's close to an eighty-point swing in about a year.

The visit that used to waste your time is now, on a per-session basis, among the most valuable traffic arriving at your store. Which is wonderful news, unless you spent the last eighteen months telling your team that AI search optimisation could wait until Q3.

Two things happened simultaneously, and separating them matters because it determines whether this trend is durable or a statistical mirage.

The first is habit. Asking an AI assistant a question stopped feeling like a novelty and started feeling like the obvious first step. Adobe's consumer surveys through early 2026 show that close to four in five shoppers who use AI to shop now report feeling more confident in the purchase they eventually make. Confidence is doing a lot of heavy lifting in that sentence. A confident shopper doesn't need to re-run your entire product comparison grid before committing. They've already decided. Your checkout page is a formality.

The second force is that the assistants got competent at the specific, unglamorous work of actually shopping. Early chatbots were enthusiastic but unreliable — like handing your shopping list to someone who's never been inside a supermarket but has read extensively about the concept. They couldn't hold budget, size, compatibility and delivery window in mind simultaneously. The current generation can, and it pulls live product data rather than hallucinating specifications from 2023. When the tool narrows competently, the human trusts the output. Trust converts. Neither force works without the other.

To understand why these numbers look so dramatic, you need to understand what's happening to the shopping journey itself, because the mechanism has a very specific shape.

In a conventional search journey, a shopper might run three to five separate searches, open eight to ten tabs (across sites they'll never visit again), drift away to make dinner, forget what they were doing, return two days later, and finally buy in a completely separate session. Discovery, comparison and decision are spread across time, across your pages, and across your competitor's pages. Most visits start on a homepage or category page, not on the product they'll eventually buy.

In an AI-mediated journey, that entire deliberation collapses into a single conversation. The shopper states the need. The assistant asks clarifying questions. It narrows the field and surfaces a specific product. By the time someone clicks through to your store, the deciding is largely done. They're arriving pre-qualified, pre-convinced, and pointed directly at a product page rather than wandering in through the front door wondering what's on offer.

This is journey compression — the collapse of the entire purchase funnel into one exchange that deposits a near-buyer onto your checkout doorstep. And yes, it flatters the conversion numbers. An assistant that only sends people through once they've resolved most of their doubts will naturally produce high conversion rates (that's partly a description of who's being sent, not proof that the channel manufactures demand from thin air). Read the numbers with that caveat in mind.

But the caveat doesn't weaken the strategic conclusion. If anything, it sharpens it.

If most of the deliberation now happens inside a conversation you cannot see, on a platform you do not control, then the moment that decides whether your product gets bought is no longer a moment on your website. It's a moment in the chat, when the assistant assembles its shortlist.

You are either on that shortlist or you are not. There is no "page two."

For twenty-five years, the discipline of getting found online was about ranking — appear high enough in a list of blue links, and a human would click. The emerging discipline (variously called answer engine optimisation or generative engine optimisation, depending on which conference you last attended) plays by entirely different rules. The assistant isn't sending a human to browse your options. It has already browsed them on the shopper's behalf, by reading whatever it could parse of your catalogue, your specifications, and what the wider internet says about you.

If it couldn't read you cleanly, it couldn't confidently recommend you. And a product an assistant can't quote with confidence is a product it quietly leaves off the list. No notification. No "sorry, we couldn't crawl your page" email. Just silence, and a sale that went to whoever was machine-readable.

Here's where the data gets genuinely uncomfortable. Adobe's own analysis found that roughly a third of product pages are not reliably accessible to AI systems, and a meaningful portion of homepage content isn't structured in a way large language models can use. A third. That's not a rounding error — that's an entire aisle of the shop with the lights off.

The honest caveat: in absolute terms, this traffic is still small. AI-referred visits remain a low single-digit share of total sessions at most retailers. A large conversion premium on a small base is still a small number in today's revenue column. Anyone claiming this channel already pays the mortgage is overselling it.

But that is precisely the argument for acting now rather than later. The case for waiting rested on a specific belief — that the channel didn't convert, so effort spent there was wasted. That belief was reasonable in 2024. It is simply no longer true.

What remains is a channel growing at a steep rate, converting and monetising better per visit than the search traffic we've collectively optimised around for two decades, and whose entry requirement is largely about being machine-readable rather than being outspent. The work itself — clean structured product data, complete and specific descriptions, content that answers questions buyers actually ask, pages an assistant can render without a fight — is concrete, mostly one-time per template, and overlaps heavily with the technical fundamentals of good conventional SEO anyway.

The deeper reason "later" fails is structural. You cannot optimise your way onto a shortlist after it's been drawn. If the deliberation has moved into a conversation, and the assistant assembles its recommendation from what it can read of you today, then being late isn't arriving to a slightly more crowded field. It's being systematically excluded from the room where the choice is made, over and over, without ever seeing the sessions you lost.

The cost of delay used to be opportunity cost. It's starting to look more like absence.

The conversion gap has flipped. The channel that once justified waiting now justifies urgency. The work required — structured data, rich product attributes, content that answers real questions in natural language — is neither exotic nor expensive. Most of it is work that should have been done anyway (and if that stings slightly, good — it means you already know what needs doing).

The less comfortable truth is that the next phase may not even involve a click-through to your site at all. Agents completing transactions inside the platform, without ever sending the visit that conversion was measured on, is already on the roadmap. The metric that just moved in our favour could stop describing the thing that matters before we've finished celebrating it.

But that's a problem for the merchants who are on the shortlist. The more immediate problem is for those who aren't.

P.S. If a third of product pages can't be read by AI systems, and you haven't checked whether yours are among them, you might want to do that before reading the next paragraph.

P.P.S. There is no next paragraph. Go check your pages.

✒️ Parsio - Extract structured data from your PDFs, emails and other documents, automatically.

🌄 Pikzels AI - Create AI-generated thumbnails effortlessly

⚙️ Smartly - Achieve better marketing campaign performance and business results with a single workflow using AI.

🤖 D-ID Agents - Create and deploy digital people that work for you.

😍It was great, all I needed

🙂Good, not great

🙄It sucked