AI becomes a major Cyber Week referral channel

Salesforce forecasts that the 2026 Cyber Week period will generate US$357 billion in global online sales, with US$75.8 billion, or 21 per cent, influenced by AI and agents. The company says AI-driven traffic to retail sites is up 104 per cent year over year and expects one in five holiday shopping orders to be influenced by AI recommendations. Its Shopping Index draws on activity from more than 1.5 billion global shoppers and 1.5 trillion page views across Salesforce commerce, service, marketing and agent products.

Influenced sales are not the same as agent purchases

A transaction can be influenced by an AI recommendation without an agent completing checkout. A shopper may ask for ideas, click a referral and buy later through a conventional session. Retailers should understand Salesforce’s attribution window and channel definitions before treating US$75.8 billion as autonomous commerce. Overlapping marketing touchpoints can claim the same sale. Incrementality tests remain necessary to establish whether AI created demand, shifted the chosen product or merely appeared in a journey that would have converted anyway.

AI referrals may bring higher intent

Salesforce says visitors arriving through AI referrals are twice as likely to convert as typical ecommerce traffic. Conversational research can narrow choices before a shopper reaches the retailer, creating a high-intent visit. That does not prove the channel will retain the same performance as it scales. Early users may be more motivated or comfortable with technology. Retailers should compare order value, returns, support contacts and repeat purchases, not only conversion rate. A recommendation that drives a quick sale but a high return is not an efficient customer outcome.

Nonhuman traffic changes storefront engineering

Salesforce expects 20 per cent of holiday shopping traffic to be nonhuman. Some of that activity may be legitimate agents checking products, price and availability; some will be scrapers, fraud or automated load. Retailers need ways to authenticate approved agents, rate-limit abusive behaviour and protect inventory without blocking useful discovery. Product pages and APIs should expose consistent structured data, while checkout retains strong confirmation and anti-fraud controls. Capacity planning must account for machines that can generate many searches in seconds even when only a few lead to purchases.

Recommendations carry market power

Salesforce reports that one in five shoppers will begin holiday shopping with AI, one in three is introduced to new brands through assistants and 11 per cent will not buy when AI advises against it. These figures suggest recommendation systems can shape which merchants enter consideration. Assistants should explain whether results reflect paid placement, available integrations or a broad market search. Retailers need accurate feeds and clear product policies, but they should not be required to join every proprietary agent ecosystem simply to remain visible. Transparent ranking is a consumer and competition issue.

Agent-ready commerce needs reliable semantics

A shopping agent requires more than a descriptive product page. It needs current price, stock, delivery date, variant, return terms and a safe way to identify the merchant. Inconsistent data can create a recommendation that cannot be fulfilled. Retailers should use structured catalogues, stable identifiers and explicit checkout states. An agent must not infer that an item is available because a page was cached. Before payment, the user should see the merchant, items, total, delivery, substitutions and card, with a receipt that links to the retailer’s order system.

Fraud and support must evolve alongside discovery

Agentic traffic can automate beneficial shopping and malicious testing of promotions, accounts and payment methods. Fraud systems need to distinguish a user-authorised agent from bot abuse without relying solely on browser signals. Customer support should preserve the action history when an order was assembled by an assistant, so disputes do not require the shopper to reconstruct a conversation. Responsibility needs to be clear when incorrect product data, agent reasoning or retailer fulfilment causes the problem. Trust will be built through recovery, not just successful recommendations.

The forecast is a readiness signal

Salesforce’s figures point to a holiday season in which AI is no longer a marginal discovery channel. They are forecasts based on the company’s data and methodology, not guaranteed market totals. Retailers can use them as a reason to test structured product data, agent authentication, capacity and attribution before traffic peaks. The strategic opportunity is a shorter path from intent to a suitable purchase. The risk is concentrating influence in opaque assistants and measuring every AI-touched order as success. Durable agentic commerce will require transparent recommendations, dependable transactions and evidence that the channel improves outcomes for shoppers as well as retailers.

After Cyber Week, the most valuable report will compare predictions with observed behaviour. Salesforce should publish how AI influence, nonhuman traffic and conversion were ultimately measured, including revisions and regional variation. Retailers can compare those aggregates with their own referral and order data. Transparent post-event analysis would help the industry separate a structural shift from a seasonal spike and improve planning for 2027. It would also reveal which agent experiences generated repeatable customer value rather than novelty-driven clicks.