AI starts the search, Amazon closes the sale: PYMNTS holiday data for CPG brands

The most useful sentence in the new PYMNTS Intelligence report on agentic shopping is a short one: "The research start moved, but completion remains concentrated." The report, titled Will the 2026 Shopping Season Go Agentic?, draws on consumer and merchant surveys PYMNTS fielded between September 2025 and August 2026. PYMNTS wrote it up on October 30, and the headline figure is that Amazon completed 59% of shoppers' most recent AI-assisted purchases.

We read the full report rather than the summary, because the details matter for a packaged goods brand deciding where to spend time before the holidays. Below are the numbers we think change a decision, charted from the report's own figures, with the caveats PYMNTS itself attaches to them.

One caveat first, because it applies to almost every chart. PYMNTS reports many of these numbers as separate measures on separate survey bases. Research starts, purchase completions and AI-assisted completions are, in its words, "independent distributions rather than a tracked funnel." So 29% and 59% do not describe the same shoppers moving from one step to the next. They describe where people tend to start and where people tend to finish, asked separately.

Half of adults have already bought with AI help

The scale of AI-assisted shopping is no longer small. PYMNTS estimates that 131.8 million US consumers, about half of adults, have bought something at retail with AI help. It projects these counts by applying survey shares to the adult population, so they are estimates, not counts of accounts or orders.

Show the data

Measure	Millions
Bought something with AI help	131.8M
Would let an agent shop and buy	62.9M
Start retail research with an AI tool	58.4M
Now research on AI what they used to search for on Amazon	57.5M
Still buy on Amazon when an AI tool found the product	54.5M
Would let the AI app hold their card	36.8M

Source: PYMNTS Intelligence, Will the 2026 Shopping Season Go Agentic?. Projections from survey shares. The first three bars project to 262.3 million US adults; the two Amazon bars project to 95.6 million AI-assisted buyers (N = 690). The card figure is from the report's text.

Two of those bars matter most for a brand. 57.5 million people say they now use AI to research things they used to search for on Amazon. And 54.5 million say they still buy on Amazon even when an AI tool found the product. Read together, they describe the pattern the whole report keeps returning to: the question moved to the assistant, the checkout did not.

What did people use AI for? On their most recent AI-assisted purchase, 46% said it helped find the best price or a deal, 44% said it compared specific products or brands, 36% said it checked or summarized reviews, 35% said it found where to buy, and 30% said it helped decide what type of product they needed. None of the top five is "bought it for me."

Research moves to AI, money stays with Amazon

This is the chart behind the headline. PYMNTS asked where consumers start online retail research and where they complete purchases, and separately asked AI-assisted buyers where they completed their latest AI-assisted purchase.

Show the data

Measure	Value
Main place research starts	29%
Main place purchases are completed	49%
Most recent AI-assisted purchase completed	59%

Source: PYMNTS Intelligence consumer survey, August 2026. N = 2,061 for research start and main completion; N = 952 for AI-assisted completion. Independent distributions, not a tracked funnel.

PYMNTS also reports that 22% of consumers now begin online retail research in an AI tool. Its hypothesis is blunt: "Discovery goes agentic before checkout does." The company that wins the near term, it argues, is the one that already holds the shopper's saved card, loyalty identity, delivery promise and returns relationship, rather than the AI service that started the search.

The timing of Amazon's own messaging fits. On its October 29 earnings call, Andy Jassy said, according to the Wall Street Journal as reported by PYMNTS, that Alexa for Shopping "is going to be the primary shopping agent customers use to shop Amazon." The same day, Amazon said Alexa for Shopping now handles more than 100 new shopping actions, from grocery lists to Subscribe & Save changes, and that over 400 million customers have used it in the last 12 months.

For a CPG brand, the implication is concrete. A shopper may ask ChatGPT or Gemini which electrolyte mix is lowest in sugar, read a comparison, and then open Amazon to buy. The assistant decided which three products were considered. Your Amazon listing, price and delivery date decided which one was bought. Both have to be right, and they are different jobs. We covered the Amazon half in our Alexa for Shopping playbook.

Trust in full delegation went down this year

The report's most surprising chart is a trend line. PYMNTS has asked consumers monthly whether they would want an AI agent to handle shopping and purchasing for them autonomously. The share fell from 32% in September 2025 to 24% in August 2026, with a low of 22% in June.

Show the data

Wave	Value
Sep 2025	32%
Jun 2026 (low)	22%
Aug 2026	24%

Source: PYMNTS Intelligence consumer surveys, September 2025 to August 2026. Three waves shown from the monthly series. August 2026 N = 2,061; wave bases range from 1,969 to 3,288.

PYMNTS notes that August is still below every month from September through March, and calls it "not the growth curve the market expected a year ago." Consumers remain more willing to delegate shopping than banking, which the report reads as people ranking the risk of the job rather than rejecting delegation outright.

The card question is sharper still. Only 14% of adults would let the AI app hold their card, a figure PYMNTS says is down 12% from a year earlier. About 52% prefer a payment experience that keeps the card outside the AI app, either through a linked wallet or by going back to the store to pay. In a May 2026 wave, PYMNTS asked whom consumers trust most when buying an item they found on a generative AI platform. The digital wallet they would pay with got 31%, the issuing bank 21% and the card network 16%. The AI platform itself got 9%.

This lines up with what we wrote about agent orders being small while agent research is not. The agent that decides your shelf position this season is mostly an adviser, not a buyer. That should shape where a brand spends effort: on the facts an adviser reads, more than on an agent checkout integration few shoppers will use yet.

What would pull a shopper to your own site

The report has one figure built for DTC brands. PYMNTS asked consumers who prefer large online marketplaces what would move them to buy from a brand's own site instead.

Show the data

Reason	Value
A lower price	53%
Shipping as fast as the marketplace's	49%
Returns as easy as the marketplace's	35%
Loyalty points from the brand	33%
A saved wallet that needs no new account	23%

Source: PYMNTS Intelligence consumer survey, August 2026. N = 1,212. Multiple responses allowed.

There is no surprise at the top: price and shipping. The more interesting result sits next to it. PYMNTS also asked people to pick a destination with price, shipping and returns held equal, once choosing themselves and once letting an AI assistant choose. A large online marketplace got about 40% either way. The brand's own site moved from 15.3% to 16.0%. In other words, handing the choice to an assistant barely changed where people expected to buy.

There is a door, though. 43% of AI shoppers say they have bought from an online store they had not used before since they started shopping with AI. AI widens the set of stores people consider. Whether a brand site converts that visit still comes down to the boring parts: price parity, a delivery promise that matches Amazon's, and returns that do not require an email to customer service.

Merchants are preparing, but not for agents yet

The merchant half of the survey is smaller, 60 US retail merchants, so we would treat it as directional. Still, the pattern is clear.

Table: US retail merchants on agentic commerce, August 2026

Measure	Share of merchants
See agentic commerce as an opportunity	87%
Have started product and catalog work	90%
Say customers arrive having already researched with an AI assistant	85%
Say customers ask whether they can shop with an AI assistant	13%
Expect a fee or commission on each agent-completed sale	80%
Very or fully prepared to integrate with AI platforms or agent networks	23%

Source: PYMNTS Intelligence merchant survey, August 2026, N = 60.

Merchants report spending an average of 2% of annual revenue on preparing for agentic commerce. Their strongest capabilities are real-time inventory through APIs and tokenized payments, which they had built for other reasons. Integrating with AI platforms or agent networks is the one capability where the unprepared outnumber the ready.

The gap between 85% and 13% is the line we would underline. Shoppers bring AI research to the store, but they almost never ask for an agent at checkout. PYMNTS offers two readings: little demand for agents at checkout, or agents working upstream where the merchant cannot see them. With no merchant in the study reporting full attribution, both are possible. Either way, it is another reason to check what assistants say about your products before the shopper arrives. That can be a monthly list of questions run by hand, or a tool such as SearchPanel from XENA Intelligence, which asks assistants shopper-style questions and scores how each product shows up.

What we would do with this before Black Friday

The data points to a short list for CPG brands in November:

• Treat AI research and Amazon checkout as one journey. If assistants recommend your product but the Amazon listing has a weaker title, older images or a later delivery date than a competitor, the assistant's work is lost at the last step.
• Make the price question easy to answer. Price and deals were the top reason people used AI on a purchase. A product page and listing that state size, count and unit price clearly give an assistant something exact to compare.
• Do not build for full autonomy this season. With 24% willing to let an agent shop and buy and 14% willing to let an AI app hold a card, a polished agent checkout is a later project for most brands. Our holiday playbook covers what to fix first.
• Match the marketplace before asking shoppers to leave it. If you want AI-referred shoppers on your own site, the PYMNTS list is the brief: price, shipping speed, easy returns, then loyalty.

PYMNTS says it will remeasure before the holidays. We will chart the update when it lands.