When AI shops for us: what agentic commerce means for luxury and jewellery

Until recently, online shopping meant searching, opening tabs and making the decision yourself. Agentic commerce introduces a new intermediary: software that does not merely answer, but can assemble options, reconcile specifications and—when authorised—move closer to purchase.

What changed: from search result to recommended decision

The signals are no longer experimental. Reuters reported on 7 August 2026 that retailers including Walmart, Ulta and Wayfair were adapting product content for traffic from ChatGPT and Gemini, while trying to keep ownership of customer relationships and data. That tension is the centre of the story: visibility inside an AI answer is attractive; becoming an anonymous row in a comparison table is not.

Adobe Analytics, drawing on more than one trillion visits to U.S. retail sites, found that generative-AI referrals in February 2025 were about 1,200% above July 2024. Jewellery, alongside electronics, was among the categories where conversion from that traffic was strongest. This is U.S. evidence, not a forecast for Iran, but it makes jewellery an early test case for machine-mediated recommendation.

Google’s I/O 2026 Universal Cart is designed to let an agent coordinate products across merchants. Mastercard and Visa are building authentication and control layers for agent-initiated payments. This does not mean unrestricted autonomous buying. The central design problem is identifying the authorised agent, defining its mandate and preserving transaction accountability.

In agentic commerce, a product page must be legible to a machine and believable to a person.

Why jewellery is not shampoo or a charging cable

An agent is good at comparing explicit constraints: budget, fineness, weight, size, stone colour, delivery and returns. A valuable jewel, however, combines hard and soft truth. Is the stone natural or laboratory-grown? Will the chain survive daily wear? Can the ring be resized? Who repairs it? Most importantly, why should the buyer trust the claim?

For high-value purchases, an agent can create a shortlist but cannot remove accountability. Missing data may cause a brand to be excluded—or invite the system to infer. Exaggerated imagery creates a different failure when delivery meets reality. Both are trust costs.

What does a machine-legible brand look like?

It is not a page stuffed with keywords. Agents need consistent facts: name, fineness, weight or range, dimensions, stone type and origin, availability, country of manufacture, care, resizing, warranty and returns. The visible page and structured data must mean the same thing. Contradictions between a caption, product page and invoice are as damaging to a machine as to a customer.

The second layer is identity. Who is the brand? What does it make? Who authored its specialist guidance, and which evidence supports its claims? Sourced editorial, a clear About page, real contact details, canonical URLs and Product/Article/Organization data form a network that lets an answer engine attribute information to a real entity.

The strategic risk: the agent knows the customer; the brand sees the order

In luxury selling, relationship data is part of the value: occasion, taste, size, repair history and explicit exclusions. If that context remains with an AI intermediary, the brand may receive only an SKU and delivery address. Reuters’ reporting shows retailers negotiating precisely this tradeoff between access to traffic and control of the relationship.

The answer is not to block agents. It is to give customers a reason to continue directly: credible consultation, fitting, personalisation, a digital product passport, care and repair. An agent can compare price. The relationship must offer something that the lowest number cannot replace.

A 30-day playbook for a gold or jewellery brand

Agent readiness does not begin with buying software. In week one, audit twenty important product pages as a demanding buyer. In week two, reconcile visible facts and Product/Organization/Article markup. In week three, publish concise, sourced answers to ten real customer questions. In week four, measure the path from AI discovery to human contact, appointment or purchase.

In Iran, autonomous payment and platform integrations may arrive later or take a different form; that is a practical inference, not a measured forecast. Discovery matters now. A brand unable to answer precise questions or substantiate claims today is unlikely to make an agent’s shortlist tomorrow.

Frequently asked questions