AI Beats Social for UK Shopping Discovery - But Brands Are Invisible Where It Counts

The Scroll Has a New Boss

One in five UK consumers now begin their shopping journey on AI-powered search platforms such as ChatGPT and Google Gemini, putting those tools ahead of Facebook, TikTok, and Instagram for retail discovery, according to research published last week by Vercel and the World Retail Congress. That ranking alone should jolt any DTC operator still pouring the bulk of their budget into social reach.

The numbers stack up. Eighty-seven percent of UK shoppers find AI-powered product recommendations useful, and 51% already use AI to compare prices before buying. More than half say they have purchased a product after an AI recommendation. That is conversion-funnel territory, not novelty.

There is a hard floor underneath the optimism. Only 10% of UK consumers said they would spend more with an AI shopping assistant than with a human customer service agent. Retail Economics CEO Richard Lim put it plainly in a separate report: "In 2026, the focus shifts from experimentation to execution." Discovery intent is there. Wallet release is not, yet.

The Readiness Gap Is Real

Here is where it gets uncomfortable for operators. The same week those consumer numbers dropped, separate research from Varn Search Marketing found that only 12% of the UK's 100 biggest ecommerce brands are configured for AI systems to complete purchases directly on shoppers' behalf. Among the unprepared: M&S, Currys, and Barbour. Among the slim prepared cohort: Gymshark and Castore.

Varn's head of search and innovation warned it would "catch unprepared brands completely off guard." The structural point is sound.

The core problem is data architecture, not ambition. An AI shopping agent does not see your lifestyle photography or your branded checkout flow. It reads your data: price, availability, variants, return policy, and whether it can actually transact. Forty percent of ecommerce businesses are still standardising product pages for agentic AI, and 33% have not started at all, according to readiness research from earlier in the year. Awareness is high; execution is not.

Shopify has been moving faster than most merchants realise. Its Agentic Storefronts feature lets merchants sell inside ChatGPT, Microsoft Copilot, Google AI Mode, and Gemini simultaneously via a single admin toggle. One setup syndicates a full catalogue across multiple AI platforms, with attribution flowing back per channel. A small boutique in Brighton, in theory, now has the same AI channel access as a London enterprise retailer.

John Lewis has gone a different route, enabling shoppers to use Google Gemini and ChatGPT within its own shopping journey. Kingfisher deployed AI-driven assistants across B&Q in partnership with Google Cloud: an agent that can plan a full home improvement project, build the shopping list, and complete the purchase in a single conversation.

Those are the big guns. For the other 88%?

The IP Blind Spot Nobody Talks About

Alongside agentic readiness, almost nobody is discussing trademark coverage. If an AI agent retrieves products by matching brand names, product descriptors, and retail service categories to structured data, a brand with weak or missing IP protection is handing that real estate to whoever has filed.

Class 35, the Nice Classification covering retail services, advertising, and business management, is the formation code for most ecommerce operators. It is also where the filing data turns alarming. According to AIBD analysis of IPO (TMD) data, UK Class 35 trademark filings reached just 4,947 in Q3 2026, down 54.2% versus the prior period. As of August 2026, the trademark filing rate in the UK's most commercially critical class has been roughly halved.

That drop sits against a backdrop where, per TrademarkDashboard, the UK register receives 70,000 to 90,000 new applications annually and Class 35 remains one of the three most popular classes on the register. Something has compressed the pipeline sharply. Whether that is economic caution, a surge in deferred filings, or brands assuming their Google presence constitutes protection, the gap is real.

Brands are investing in AI visibility at the storefront level while simultaneously thinning out their IP defences at the registry. When an AI agent surfaces your product name in a shopping recommendation and a competitor with a valid Class 35 registration objects, the conversation stops being about conversion rates.

What to Do With This

Three calls to action, in order of urgency.

First, audit your product data. Agent-readable means specific attributes, clear pricing, machine-parseable return terms, and a working API or structured feed. This is not a redesign; it is a data hygiene job.

Second, check your AI channel presence. If you are on Shopify, Agentic Storefronts is the fastest path to appearing inside ChatGPT and Gemini shopping flows. If you are not, map which protocols your platform supports: Google's Universal Commerce Protocol and OpenAI's Agentic Commerce Protocol (co-created with Stripe, used by Shopify, Etsy, and others) are where the buying is moving.

Third, file the mark. With Class 35 filings down 54% in Q3, the queue at the UK IPO is shorter than it has been in years. That is a filing window, not a reason to relax.

The AI agent is not coming. It is already here, reading your data and deciding whether you are worth surfacing. Most brands are not ready to be found.