AI shopping in 2026: what African retailers should do now

AI shopping is becoming a practical sales channel, with different levels of access across markets. For African retailers, the immediate opportunity is to make products easier to find and evaluate, then give customers a dependable route to payment and delivery.

The distinction matters because an AI assistant recommending your product and a customer buying it inside the assistant are separate features.

As checked on 8 September 2026, Shopify’s documentation for Google AI Mode and Gemini direct checkout requires stores to be based in the United States and sell to US customers. It also describes an ongoing rollout. Those conditions apply to this specific integration, rather than every form of AI shopping. [1]

I would start with this detail before spending money on an integration. A retailer needs to know which services support its business location, customers, products and payment arrangements.

The opportunity is worth preparing for. The preparation should improve the business even before a new checkout becomes available.

What agentic commerce means

Agentic commerce describes shopping in which AI software performs tasks for a customer, such as comparing products, preparing a basket or arranging an authorised purchase. The level of independence depends on the system and the permission the customer gives it.

On 11 January 2026, Google announced the Universal Commerce Protocol, or UCP, developed with companies including Shopify, Etsy, Wayfair, Target and Walmart. It provides a shared way for AI systems and commerce services to communicate across shopping and order-related tasks. Google’s initial checkout announcement focused on eligible US retailers. [2]

Shopify’s current help pages describe agentic storefronts as active by default for eligible stores. They also distinguish product visibility from direct checkout inside an AI channel. [3]

These developments give retailers specific services to assess. They do not establish how much additional profit an individual shop will earn.

For a business owner, three separate checks matter: whether the product is visible, whether the customer has a working purchase route, and whether the resulting order earns enough to justify the cost.

Accurate product information comes first

An AI shopping service needs usable information about the product it recommends. Shopify describes catalogue data including titles, descriptions, options, images, prices and availability. Its guidance directs merchants towards complete, structured product information. [4]

Consider a hypothetical customer seeking a laptop bag which fits a 16-inch device, costs less than an agreed budget and arrives before a trip. A useful listing states the internal dimensions, material, available colours, stock position and delivery estimate.

I would want the retailer’s systems to give the same answer on the product page, in the product feed and at checkout. If the advertised size is unavailable, the system should report it. If delivery depends on the postcode, the customer should see the relevant estimate before paying.

This is useful work for retailers serving Lagos, Nairobi, Accra, Johannesburg or customers overseas. Each business still needs to establish its own delivery coverage, currency and payment options. An AI interface does not remove those operating requirements.

My first investment would be in the accuracy of the information the business already publishes. Better descriptions, current inventory and clear delivery terms also help customers arriving through ordinary search or social media.

Visibility and checkout need separate checks

Shopify says its Google AI Mode and Gemini integration uses Google Merchant Center for product information. Its documentation also explains a route in which customers find products in the AI experience and complete the purchase on the merchant’s own store. [1]

This gives an African retailer a more useful question to investigate: which supported discovery routes already lead customers to a checkout the business controls.

A platform’s international ambitions should be assessed alongside its current eligibility rules. Before committing to development work, I would ask the provider to confirm the supported countries, product types and payment methods in writing.

I would also check how stock changes reach the platform and how orders return to the shop’s own systems. A product recommendation has limited value if the merchant cannot fulfil the resulting order.

Permission needs a record

The payment stage introduces a separate requirement: evidence of what the customer authorised.

Google introduced the Agent Payments Protocol, or AP2, on 16 September 2025. Its published design uses cryptographically signed records called mandates to capture instructions and approvals. It describes both purchases approved while the customer is present and delegated purchases governed by instructions given in advance. [5]

The principle is understandable without studying the protocol. A system acting for a customer needs a defined purchase, spending limit and set of conditions. An instruction to compare prices does not provide permission to spend.

For example, a hypothetical customer authorises a bag costing no more than £80 including delivery. I would expect the service to seek fresh approval if the final total exceeds £80 or the selected item changes outside the agreed conditions.

A protocol supplies part of the technical process. The merchant still needs to establish who handles a failed order, how the customer requests a refund and which records the payment provider requires when something goes wrong. These are operational questions to settle before launch.

Measure the money left after the sale

I would evaluate an AI sales channel through completed orders and the contribution each order leaves towards overheads and profit.

Take an illustrative order with £100 in revenue after discounts and excluding tax. Suppose product cost is £55, payment and channel charges total £5, fulfilment costs £10 and the expected allowance for returns and support is £5. The remaining contribution is £25, before fixed overheads.

If the retailer adds a £10 discount and other costs stay unchanged, the contribution falls to £15. A 10% reduction in revenue has reduced the contribution by 40%.

These are hypothetical figures, not a quoted fee schedule. They show why I would track the quality of sales alongside the number of orders. A channel which creates more returns, expensive delivery requests or heavy discounting needs close examination.

The useful measures include completed purchases, cancellations, refunds, fulfilment cost and contribution per order. Referral traffic helps explain where customers arrive from. The order economics establish whether the channel deserves more investment.

Search and AI visibility share practical foundations

Retailers also need to distinguish being found in an AI answer from enabling software to place an order. Clear public information supports the first task. A supported commercial integration and appropriate permission are needed for the second.

Google’s Search guidance says established SEO practices remain relevant to AI Overviews and AI Mode. It specifies no special AI markup as a requirement for appearing in those features. Important content should be available as text, accessible to search systems and consistent with any structured data. Eligibility does not guarantee inclusion. [6]

I would spend first on useful product pages, accurate information and a reliable mobile checkout. Any proposed AI visibility service should explain what it changes, which platform supports the approach and how results will be measured.

What I would do next

I would begin with a small group of products the business fulfils reliably. For each one, I would check the description, variant, price, stock, delivery coverage and return terms against the actual service offered.

I would then review the sales channels available to the business and test the complete customer journey through a supported route. The test should include an unavailable item, an unsupported delivery address and a request to cancel or return an order.

Any initial trial would have a fixed budget and an agreed review point. I would expand it only after the orders demonstrate acceptable margins, manageable support and dependable fulfilment.

My view is the strongest preparation for AI shopping is disciplined retail operations. African businesses should assess each new service on its availability and results, while making their own product information and checkout easier to use today.

I will judge progress by customers receiving what they ordered, merchants understanding what they earned and both sides retaining a clear record of the transaction.