Holiday shoppers in 2026 are asking AI tools what to buy and where, then finishing most purchases themselves. Two surveys published on Nov. 12, one from Salesforce and one from SAP, put numbers on that split, and both point marketers at the same job: the product data an AI reads before a shopper ever reaches a brand’s site.
What the two surveys found
Salesforce’s Consumer Shopping survey covered more than 2,500 shoppers in the U.S., the U.K., Canada and Australia. Almost 20% said AI is their first stop on the way to the store, and 74% said they trust the recommendations it gives them. Thirty percent of AI shoppers said these tools frequently introduce them to brands they did not know, and 11% said they would skip a purchase if an AI assistant advised against it.
Salesforce’s own projections, published in the same post, put AI-driven Cyber Week spending at $75.8 billion, up 13% from 2025, with AI recommendations expected to influence more than 1 in 5 holiday orders. Traffic from AI tools to ecommerce sites is up 104% year over year, by the company’s count.
SAP’s AI in Customer Experience Report 2027 surveyed 4,800 consumers, business buyers, CX practitioners and decision-makers in the U.S. and Germany. It found that half of U.S. consumers now use AI in at least some of their purchase journeys, and that nearly all business buyers surveyed had used AI for purchasing or vendor selection. SAP calls the result the hybrid AI buyer: discovery and shortlisting happen inside an AI tool, while purchase, fulfilment, service and returns still run through the business.
Shoppers hand agents the research and keep the payment
The Salesforce data shows where shoppers stop. Thirty-one percent of consumers have tried a personal AI agent, and Meta Muse was the most used at 12% of respondents. Among people who use agents, 28% give them tasks several times a day, mostly comparing prices, reading specifications and reviews, and brainstorming gift ideas.
Checkout is another matter. Only 19% of personal agent users have sent an agent to check out. Just 9% give their agent unrestricted access to their financial details, and 36% have set up a separate credit card for it. About 73% of shoppers are not comfortable giving an agent their payment information, and 49% are not interested in agents browsing or buying for them at all. The obstacles respondents named were unwanted sharing of purchase history or personal data, and accidental or unwanted charges.
“AI is no longer a niche tool just for tech-savvy shoppers,” said Caila Schwartz, Head of Agentic Commerce Shopper Insights at Salesforce. “Everyday shoppers are asking AI what to buy, where to buy it, and whether it’s worth the money before they ever land on a retailer’s site.”
Earlier surveys on agentic shopping produced numbers that looked contradictory, which we traced to surveys asking different questions. These two ask about different stages of the same journey, and the stages line up: heavy AI use at research and shortlisting, light use at payment.
Brand agents and the cost of mistakes
Salesforce also measured shoppers who use a retailer’s own shopper agent on its site. Thirty-nine percent had done so, 80% of them reported a good experience, and 75% of that group felt more loyal to the brand afterward. Those shoppers had an 8% higher average order value and bought 23% more items than the average order, and shoppers arriving through AI referrals were more than twice as likely to convert as typical ecommerce shoppers.
The tolerance for errors is low. Most shoppers will abandon an AI assistant after two or three mistakes. Among personal agent users, 27% said their agent had bought something too expensive, 22% had seen an agent blocked from completing a purchase, and only 32% reported no major errors.
The shortlist is built from catalog data
SAP’s report puts the marketing consequence plainly: if AI leads the research, the brand has to be findable and recommendable by the AI. It describes discoverability as a continuous job in which every attribute, image, specification, availability signal and proof point shapes whether a product is shortlisted.
Jessica Keehn, chief marketing officer for SAP Customer Experience, wrote in the report summary that discoverability is earned and earned again on a continuing basis, because AI models keep re-evaluating what they surface and recommend.
The same report shows how far most companies are from that standard. Ninety-three percent of U.S. CX decision-makers said they feel prepared for AI-enabled customers, and 6% said all relevant systems are connected end to end. A catalog an agent reads has to reflect current prices and stock, and that depends on inventory, pricing and commerce systems sharing data.
Site analytics are starting to show the agent side of this traffic. Microsoft Clarity reported earlier this month that checkout ranked as the top action across 5.2 million agent sessions it recorded, and retailers such as Walmart have begun writing profiles and terms for AI shopping agents that visit their sites.
What it means for the marketing leader
In our reading, the Salesforce and SAP numbers move a large share of holiday persuasion upstream, into AI answers the brand does not write, while leaving the purchase and everything after it with the brand. That changes where marketing effort pays off this season.
Schwartz’s own pre-Cyber Week advice in the Salesforce post is specific: “Shoppers are starting with AI this season, so make sure your product data is accurate, your prices and inventory are current, and your own agent is ready to answer the hard questions.”
What to do before Cyber Week
- Audit the feed an agent reads. Check product titles, attributes, sizing notes, prices and stock in the feeds and pages AI tools crawl, and fix gaps in the products you plan to promote most.
- Separate agent traffic in reporting. Track AI referrals and agent sessions as their own segment so conversion and order value for that traffic can be compared with the Salesforce benchmarks.
- Test your own shopper agent against bad outcomes. Given that most shoppers drop an assistant after two or three mistakes, test it on out-of-stock items, price questions and delivery cut-off dates before peak traffic arrives.
- Keep checkout human-friendly. With 73% of shoppers unwilling to give an agent payment details, the shopper who arrives after an AI shortlist still needs a fast, familiar checkout.
Ask commerce and PIM vendors how quickly a price or stock change reaches the feeds AI tools read, and how they report agent visits separately from human ones. Those two answers decide whether a brand shows up on the shortlist with the right price.
Source: Salesforce News
