Shoppers are researching in AI and buying elsewhere. That is a data opportunity and not a channel to ignore
AI-referred traffic is growing faster than any acquisition channel in commerce, and most of it does not buy on the first visit. That is not a failure of AI. It is a signal about the job shoppers are using it for. Understanding that job is how you turn a research tool into a revenue one.
This article sets out what the current consumer data actually shows about how people use ChatGPT, Claude, Gemini and Perplexity when they shop, why those tools dominate discovery and comparison while rarely closing the sale themselves, and what UK retailers should fix now so their products get found, understood and bought wherever the decision starts.
The numbers say research, not checkout
Start with adoption, because it is no longer marginal. ChatGPT reached 900 million weekly active users by early 2026, and non-work use now accounts for more than 70% of consumer activity. Nearly 60% of US consumers already use generative AI tools for shopping tasks, and 73% of those who do cite AI as their primary source of product research. Awareness is close to universal and replacement behaviour, AI instead of Google, is already normal for a third of US adults.
What they are not doing, mostly, is buying inside the chat. A study of 973 retailers and 12 months of first-party data, covering nearly 50,000 ChatGPT-attributed transactions against 164 million from traditional channels, read the AI engagement pattern plainly: long sessions, deep browsing, lower conversion. That is research-stage behaviour, not purchase-stage. The researchers pointed to trust as the likely cause.
Consumers do not treat the AI as the last step before payment. They compare, then leave, then buy through search, direct or a route they already trust.
Adobe's data on what happens after the click tells the same story from the other side. Shoppers arriving from AI assistants spend up to 48% longer on site, view more pages, and bounce far less than non-AI traffic. They are more informed and more deliberate. They have done their comparison shopping in the conversation and arrive to confirm, not to start.
So both things are true at once. AI is now the front end of the buying decision for a large and growing share of shoppers. It is still rarely the checkout.
Why the conversion gap is closing anyway...
Here is the part retailers should not misread. "Rarely a purchase channel" is a snapshot, not a trend line.
Adobe tracked AI-referred traffic converting 38% worse than non-AI traffic in March 2025. By March 2026 the same channel converted 42% better, a swing of roughly 80 percentage points in a single year. Shopify's numbers point the same way: AI-referred sessions converting at higher rates than organic search on product pages, and AI-referred orders carrying around 14% higher average order values, because a pre-qualified buyer tends to arrive on the right, often higher-tier, product.
The direction is not in dispute. The research behaviour is compressing into purchase behaviour as the tools improve and trust builds. Retailers optimising for the AI front end today are not betting on a fringe channel. They are getting ahead of a curve that is bending fast.
Interestingly... what people buy decides where they research...
The averages hide the useful detail. AI does not lift every category equally, and knowing which is which changes where you focus.
The pattern is consistent across datasets: AI earns its keep in research-intensive, spec-led purchases. Consumer electronics and jewellery convert markedly better from AI traffic, because those are decisions made against specifications, compatibility, use case and tradeoffs. 87% of users say they are more likely to reach for AI on a large or complex purchase than a cheap one. When a shopper has to reason through which product fits a precise set of criteria, that is exactly what a conversational model is good at.
Taste-led categories behave differently. In apparel and grocery, where shoppers already have a brand, a style or a routine in mind, AI traffic can convert below other sources. People do not need an agent to tell them which t-shirt they like. But even here AI does one thing well: discovery. It surfaces brands a shopper would not have found alone.
Broad category labels flatter to deceive, though. Apparel as a whole looks unremarkable for AI, yet its spec-driven corners pull well ahead. A watch and a t-shirt sit in the same parent category and demand entirely different amounts of research before purchase. The lesson is not that some categories are "good for AI" and others are not. It is that shoppers reach for AI wherever its current strengths match the decision in front of them.
The fix is the same one that helps everywhere else
The reasonable objection: if AI still sends less traffic than organic search and closes fewer sales directly, why act now?
Because the work is shared. Google has been explicit that its AI experiences run on the same core quality and ranking systems as Search. The structured product data that makes your catalogue legible to an AI agent is the same data that helps those products rank in organic search, power on-site filtering and feed your merchandising. You are not building a separate playbook. You are strengthening the fundamentals that already carry most of your traffic.
This is the point most platforms cannot act on, because they are structurally incapable of it. When fit, compatibility, materials and use-case attributes live in free-text descriptions rather than structured fields, nothing downstream can use them: not the product page, not search, and not the AI deciding what to recommend. 70-93% of fashion returns come down to fit and sizing, which is the same attribute gap showing up as a returns cost. Retailers moving to structured PIM see return rates fall 15-40%, and stores with near-complete attribute coverage see 3-4x higher visibility in AI recommendations. Same catalogue, better data, three separate wins.
What Remarkable clients should do now
Treat AI and organic discovery as one roadmap, not two. A few high-leverage moves:
Enrich the catalogue as structured, machine-readable data.
Record fit, compatibility, materials, audience and "best for" as discrete attributes, not prose. That single source powers storefront filtering, organic search and how AI systems interpret your products. This is what Remarkable's PIM is built to do.
Build product pages for deeper-funnel arrivals.
Half of AI-referred visits land straight on a product page, past the homepage, arriving with intent and prior research. The page has to confirm the decision, not begin it: clear specs, use cases, compatibility, reviews, delivery and returns, comparison-friendly detail.
Make discovery match how people actually ask.
AI queries are longer and more conversational than keyword search. Remarkable Discover, our AI semantic search built on Google Vertex AI and Elasticsearch, reads voice, text, image or video in a single query and factors live inventory, pricing and promotions at query-time. It replaced a leading search provider across a client group's 19 sites in October 2025, improving performance while removing a £360,000 annual licence fee. Deployable on any platform.
Measure AI and organic side by side.
Most analytics setups miscount AI, folding Google AI Overviews into organic, so your AI number is almost certainly an undercount. Remarkable Signal, our agent visibility analytics across ChatGPT, Claude, Gemini, Copilot and Perplexity, is in development for Q3 2026 to close exactly that gap. Until then, build a custom channel grouping and watch it against organic by referrer.
Shopping behaviour is changing. The response is not binary
The market narrative wants a winner. AI replaces search, or AI is too small to matter. The data supports neither.
Organic search is still the larger channel and still growing. AI is the faster-growing one, concentrated today in research-heavy journeys and closing the conversion gap month on month. Shoppers now have more ways to find and buy than before, and that will keep evolving as the tools get more capable. The retailers best placed are the ones whose products are easy to discover, understand, trust and buy wherever the journey starts.
That is a data readiness problem, and it has one answer that serves both surfaces at once. One catalogue, structured properly, working everywhere.
We help retailers with underperforming sales become Remarkable retailers. If AI-referred traffic is arriving on your storefront and leaving without buying, talk to us about what your catalogue and product pages would need to convert it.
