Who is really shopping with an assistant this holiday season, and who is only asking it what to buy first? As of November 15, 2026, four surveys published within four days put the share of shoppers who plan to use AI somewhere between 13 and 28 percent, and all four agree on the part that matters for a checkout page: almost nobody is willing to hand over a card. The disagreement is not about demand. It is about what each survey actually asked, and about the distance between a shortlist and a payment.
Four Numbers, Four Different Questions
Gartner's figure, reported by CX Dive on 12 November, is 13 percent of consumers who say they will use AI for holiday gift inspiration, and those who do consult about 4.5 sources of inspiration against 3.2 for those who do not. Salesforce, in a survey of more than 2,500 shoppers across the United States, the United Kingdom, Canada and Australia published the same day, finds almost 20 percent naming AI as their first stop, with 74 percent saying they trust its recommendations.
Bain puts the share planning to start holiday shopping on AI platforms at 25 percent, up from 17 percent a year earlier. Bank of America, reported by Fortune on 14 November, finds 72 percent of Americans saying they do not plan to use AI tools for holiday shopping, up from 49 percent in 2025, which leaves at most 28 percent who do, down from at most 51 percent. Bain says the share is rising; Bank of America says it is falling.
The two claims can coexist because the questions differ. One asks where a shopper starts, the other asks whether AI is used at all. The samples differ too: Bank of America surveyed Americans, while Salesforce surveyed shoppers in four countries. Bank of America told Fortune it asked the question the same way in both years and that it measures intentions rather than behaviour, which is the honest boundary of every number in this set.
What each survey measured
| Source | Fieldwork | Headline share | Question asked |
|---|---|---|---|
| Gartner | Reported 12 November | 13 percent | Will you use AI for gift inspiration |
| Salesforce | 2,500 shoppers in four countries | Almost 20 percent | Is AI your first stop |
| Bain | Reported 12 November | 25 percent, from 17 | Will you start on an AI platform |
| Bank of America | Reported 14 November | 72 percent will not | Do you plan to use AI at all |
| impact.com | 967 AI-active UK consumers, May 2026 | 50 percent began with an AI answer | Where did your last product search begin |
Read together, the four estimates bracket a range rather than settle a number, and the range is wide enough that a planning team cannot use any single figure as a budget line. The practical reading is narrower: a meaningful minority of shoppers arrives at a product page with a shortlist already shaped by an assistant, and the size of that minority depends on how the question was framed rather than on how the market behaved. For category growth planning the useful split is between consumer demand for help and consumer demand for payment: the first is measurable and rising, the second is measurably absent.
Where the Purchase Actually Finishes
The clearest evidence on what happens after the shortlist comes from impact.com, which surveyed 967 AI-active UK consumers in May 2026. Half began their last product search with an AI-generated answer, against 28 percent who started with traditional search results, and 95 percent said they had bought a product an AI recommended. But three in five completed that purchase through a separate search, a brand or retailer site, or a physical store.
That split is the part a brand cannot see in a channel report. About 72 percent said they sometimes, usually or always click through from an AI tool to a brand or retailer site before deciding, and another 4 percent said they never use the embedded link, starting a fresh search instead. In those cases the influence is real but the referral is invisible, and the visit arrives looking organic or direct.
The refusal is about payment, not discovery
Payment is where the enthusiasm stops. In the Salesforce survey, 73 percent of shoppers are not comfortable giving a personal agent their payment details, and 49 percent do not want agents browsing or buying on their behalf at all. Of the 31 percent who have tried an agent, 19 percent have let it check out, 36 percent set up a separate card for it, and 27 percent say it bought something too expensive.
The behavioural detail matters more than the headline. A shopper who has set up a separate card has not refused the agent; they have built a spending limit around it, which is closer to how people treat a second wallet than to how they treat a checkout button. Teams reading the 73 percent as a flat rejection will misjudge both the risk and the timeline.
What a Team Can Check This Week
With the estimates this wide, the useful work is not another estimate but a set of checks on the pages an AI referral lands on. Upp.ai and Retail Economics, in a study of UK discovery behaviour, split shoppers into four groups, with convenience seekers at 38 percent of spend, informed buyers at 33 percent, impulse shoppers at 18 percent and digital explorers at 11 percent, and report that 93 percent of retailers say they need to deliver more personalised content.
None of these checks requires an agent to complete a checkout, and that is the point. They require a product page that is accurate when someone arrives already holding a shortlist, a review section that answers the objection the assistant raised, and analytics able to record that the visit began somewhere the brand does not own. Those three items are within a team's control this week; the size of the AI shopper population is not.
FAQ
Why do the surveys disagree so much?
They ask different questions. One asks where a shopper starts, another asks whether AI is used at all, and the samples cover different countries and different fieldwork dates.
Does an AI referral show up in analytics today?
Usually not. Most analytics packages record it as direct or organic traffic, which is why the first practical step is to add AI assistants as a named referral source.
Should a brand optimise its product page for assistants?
Yes, on the basics: accurate price, stock and delivery date, a structured product feed, and a review section that answers common objections before the shopper asks.
Is agent checkout worth preparing for now?
Not as a priority. The surveys show most shoppers will let an assistant research but not pay, so the return sits in the landing page rather than in the payment flow.
