Shoppers Use Artificial Intelligence, and Then Check It. What That Asks of Shops in South Africa

Who decides what a shopper sees first: the shop, the search engine or the assistant? In 2026 the answer began to move. Euromonitor International published its Top Global Consumer Trends 2027 report on 4 November 2026. It found that only 19 per cent of consumers see generative artificial intelligence, the software that answers questions in plain sentences, as an extremely trustworthy source for shopping. Most still use it to search, compare and summarise. Euromonitor calls this “Reluctant Reliance”. The shopper asks the assistant for a shortlist, then checks the specifications, the return policy and the seller’s ratings before paying. A survey by the product information company Akeneo, published on 3 June 2026, fills in the picture. Sixty-two per cent of shoppers compare prices across retailers before buying. Only 9 per cent trust a deal without checking. Forty-three per cent have used tools such as ChatGPT or Gemini to shop or find deals. And 22 per cent have bought on an assistant’s recommendation.

What does an assistant look for?

Information, more than price. When a shopper asks for a laptop or a fridge, the software checks three things. Does the item fit the space described? Are the warranty and the cables included? Do recent reviews report faults? If a shop’s product page lacks those details in a form software can read, the product is left off the shortlist however deep its discount. Newegg, an electronics retailer, used artificial intelligence to standardise specifications across thousands of product lines and saw search visibility and sales conversion rise.

Amazon’s own assistant, Rufus, was used by more than 300 million customers in 2025 and generated nearly $12 billion in additional sales. Amazon Web Services now sells the same design to other retailers, under the name Agentic Shopping Assistant. The fashion brand Kate Spade used it to build a gift concierge, and put it live in about two and a half months. McKinsey estimates that assistants acting for shoppers could handle up to $1 trillion of consumer retail revenue in the United States by 2030.

Where is the boundary?

Shoppers have drawn it, and the retailers doing well have respected it. Acceptable uses: comparing prices, translating, summarising specifications, tracking an order. Rejected uses: checkouts completed without the shopper, behaviour profiles built without consent, and prices adjusted to a person’s apparent urgency. Amazon’s assistant, renamed Alexa for Shopping in May 2026, compares and evaluates but leaves adding to the basket and paying under the shopper’s hand. Walmart’s assistant shows why a product was suggested, allows a one-click opt-out from tracking, and connects to a person when asked. Majid Al Futtaim, which runs Carrefour across the Gulf, says artificial intelligence influences 9 per cent of its transactions. It handles customer data under Saudi Arabia’s Personal Data Protection Law and the United Arab Emirates’ federal data protection law. It does not use that data to train outside models.

The retailers are also drawing a boundary against each other’s assistants. Amazon’s chief executive, Andy Jassy, spoke to investors on 29 October 2026. More than 400 million customers had used Alexa for Shopping in the past year, he said, and those customers spend 50 per cent more. Amazon expects its own assistant to be “the primary shopping agent customers use to shop Amazon”. Amazon has blocked outside agents, including ChatGPT, from buying on its site. An industry audit reported in November 2026 found that only 11 of 51 large American retailers had published rules on outside shopping agents. JPMorgan Chase and Amazon block them. Walmart allows them, at standard prices.

What about the adverts shops sell?

A shopper who checks everything has told the shop exactly what to show them.

Retail media is the business of selling advertising space on a retailer’s app, website and in-store screens. Retail World puts global spending at close to $200 billion a year. Its promise is that an advert can be matched to a real purchase. Its problem is credit. Most platforms credit a sale to an advert if the shopper clicked within seven days. So a detergent brand pays to reach shoppers who opened the app to buy that detergent anyway. The answer large advertisers in the Gulf now demand is a controlled test: show the advert to 95 per cent of an audience, withhold it from 5 per cent, and measure the difference in purchases. That difference is the advertising’s real effect. Smaller brands feel the cost first, because bidding prices rise with every new advertiser, and shoppers notice when sponsored listings push the best-rated products down the page.

What does this ask of shops in South Africa?

The same data points the other way here. Arthur Goldstuck of World Wide Worx describes South African retailers building advertising into the same system as their stores. Checkers, Pick n Pay, Takealot and TFG now have the online scale, R159 billion across the market in 2026, at which retail media and assistants are worth building. South Africa also has the Protection of Personal Information Act, in force since 2021. It sets the consent rules that Saudi Arabia and the Emirates are only now writing into their retail systems. The boundary shoppers want is already law here.

For the shop without a technology team, the questions are plainer and just as live. When an assistant describes a product to a shopper in Johannesburg, does any South African listing appear, and in what detail? If a spaza’s prices and stock exist only on a handwritten board, no assistant can recommend it. The first platform to list those shops will own the relationship. And when a retail media platform tells a South African brand that its advert worked, who checks against the 5 per cent who never saw it? Shops here have an advantage that the global players are spending billions to build: the customer is known in person, and trust was earned before any software asked for it.