The Next Buyer Your Business Meets May Be an AI Shopping Agent

Your next customer may send software to do the shopping. That software won't react to your homepage design, your tagline, or your best photography. It reads what you publish about the product and the terms of sale, then compares you with every rival that published the same kind of information.
That changes what a storefront is. For a growing number of purchases that involve AI shopping agents, the storefront is no longer the page a person sees. It's the data and policies a machine can read.
What the Holiday Forecasts Say About AI Shopping Agents
Matthew Shay, president and chief executive of the National Retail Federation, said his organization saw a significant jump last year in shoppers using AI tools to look for gifts. He expects those tools, agents included, to play a much larger role this season. The Boston Globe reported his remarks in Retailers project a record $1 trillion holiday shopping season. But who's doing the spending? The same article carries the federation's forecast that U.S. holiday retail sales will grow 4.2% to $1.013 trillion in November and December.
Adobe adds a measurement. Its holiday forecast, Adobe: U.S. Holiday Shopping Season to Hit Record $275.1 Billion Online, Rising 6.7% YoY, projects that AI traffic to U.S. retail sites will rise 130% over last year between November 1 and December 31. Adobe counts that traffic when a shopper clicks a link from an AI chat or browser service. The number tells you where shopping starts. It doesn't tell you how many purchases an agent completes without a person. Adobe also sells analytics and commerce software, so treat the figure as a projection from a company with a stake in the trend.
The evidence outside the United States points the same way. IBM surveyed more than 18,000 consumers in 23 countries in the third quarter of 2025, for a study released in partnership with the National Retail Federation. In Own the agentic commerce experience, IBM reports that 45% of those consumers turn to AI during their buying journeys. Some 41% use it to research products, 33% to look for reviews, and 31% to search for deals. Yet 72% still shop in stores.
These are forecasts and surveys, not counts of purchases agents completed alone. IBM draws the same line. It describes today's AI-assisted shopping as still relying on a person to confirm the purchase, with agents that buy on their own as the next phase. The sources agree on direction. The first look at your product is moving to a tool that reads rather than browses, while stores and websites stay in the mix. AI adds a step. It doesn't replace the rest.
Your Product Data and Policies Are the Storefront
A person browsing a site responds to brand, layout, and persuasion. An agent takes in attributes. Price, availability, specifications, delivery dates, return terms, warranty. When AI shopping agents meet a field that's missing or contradicts another page, they have nothing to weigh. A rival with complete information wins the comparison, and nobody on your side ever knew it happened.
Machines also miss more than you'd expect. In AI Traffic Drives 83% Higher Revenue Per Visit, Adobe reports that the average U.S. retail homepage scored 63% on its Citation Readability Score in September. That means more than a third of homepage content isn't visible to the AI systems sending shoppers to retailers. The same report says AI-referred visits generated 83% more revenue per visit than other traffic. Adobe sells brand-visibility software, so weigh the framing. The gap it describes is easy to test on your own pages.
Policies move into the same spotlight. Return windows, shipping promises, and cancellation terms used to sit in the footer. An agent treats them as criteria. Google's Checkout requirements and best practices shows how seriously one major platform takes this. For merchants in Google's shopping programs, availability must match across product data, product page, and checkout. The price can't rise at checkout. Mandatory fees must be shown clearly. Refund and return information must stay easy to find, and leaving out relevant details counts as misrepresentation that can lead to disapproved listings or a suspended account.
Those rules govern one company's programs, not every agent. They do show the standard a machine can enforce, which is the same facts everywhere, every time. A return policy buried under exceptions is likely to lose to a plain one. A delivery promise with no track record is likely to lose to one that has a record you can show.
Most companies have work to do here. In the same report, 54% of 200 surveyed retail, consumer goods, and e-commerce executives reported persistent challenges connecting data across channels and systems. Their companies all operate globally with annual revenue above $500 million, so read the result as a view of large organizations. IBM argues that incomplete, inconsistent, or unverifiable product information leads both shoppers and machine logic to reject a brand. Stanislas Vignon, Head of Insights for AI and Omnichannel at Louis Vuitton Moët Hennessy, put it bluntly. "If you don't have the right data, it doesn't work."
Checkout Can Move Beyond Your Website
Google announced on November 4 that direct checkout is expanding to eligible Canadian retailers through AI Mode in Search and the Gemini app. Sabrina Geremia, vice president and country managing director of Google Canada, described the rollout in A new way to shop: Introducing a more intuitive, agentic shopping experience for consumers and retailers in Canada. The experience runs on the Universal Commerce Protocol, which Google calls an open standard for agentic commerce. Geremia wrote that "Retailers remain the seller of record" even when the shopper buys inside Google's own interfaces.
That sentence deserves attention. A sale can start and finish in another company's interface while you still stand behind the order. The same post describes agents completing tasks for a shopper with the shopper's permission. Before you enable agent purchases, ask your commerce and payment providers how authorized activity is identified, how confirmations reach customers, and how disputed or duplicate orders are handled.
Geography matters too. Eligibility, rollout, and features vary, and a launch with participating Canadian retailers doesn't establish access for merchants elsewhere. If you sell in several markets, currency, measurements, delivery coverage, and return terms can change by destination. Translating a description while leaving the promise unclear invites a wrong recommendation. The announcement shows what the platform can do. It doesn't prove higher margins or fewer abandoned carts, so test those results against your own orders and costs.
B2B Procurement Is Heading the Same Way
Gartner predicts that by 2028, AI agents will sit between buyers and sellers in 90% of B2B purchases, with more than $15 trillion in spend passing through those exchanges. The prediction appears in Gartner Unveils Top Predictions for IT Organizations and Users in 2026 and Beyond. Its analysts say that in this new ecosystem, "verifiable operational data becomes a currency." Gartner also expects products built API-first and cloud-native to gain a lasting lead.
That's a prediction, and I'd hold the 2028 date loosely. The pattern behind it is easier to trust. Procurement teams already screen suppliers on specifications, terms, certifications, and delivery performance before agreeing to a meeting. An agent does that screening faster, across more suppliers, and never loses patience with a badly organized catalog. A supplier whose key facts sit in PDFs, email threads, and a sales rep's head may never reach the shortlist.
Three gaps are most likely to cost you, and this is my read of how automated screening works, not a measured result. Pricing that only resolves after a sales call gives a screen nothing to compare. Stock and lead-time data that disagree across your website, internal systems, and distributors invite a buyer's system to treat your supply as uncertain. Delivery, warranty, and compliance terms revealed only at final checkout can't be weighed early. Unit basis matters as well. A quote priced for a single part and a quote priced for a pack of 10 can look identical to a screen that misses the unit.
The consumer case and the business case share one mechanism. A person decides what matters, then hands the screening to a tool. Brand and relationships still count with the person who sets the criteria and handles the exceptions. They count for less with the tool that makes the first cut.
What Leaders Should Do Before the Next Buying Season
Start by reading your business the way an agent would. Pick 10 of your most important products or services and have a small team evaluate each one using only what an outside buyer can access. Ask the AI tools your customers use about your offers, your terms, and your rivals. Record where the team can't establish fit, total cost, delivery, or the next step. Note what the tools get wrong, what they leave out, and who they recommend instead.
Then fix the source. Compare what's public with your product feeds, your current policies, and your actual checkout or quotation process. Check that attributes match across your own site, marketplace listings, and distributor catalogs. Where they disagree, decide which version is true and correct the rest. Rewrite policies with dates, conditions, and costs. Don't respond by making every policy more generous, which can wreck the economics of the offer. Make each promise explicit, then decide which commitments you can profitably support.
Ownership comes next. This work crosses marketing, technology, operations, finance, and legal, so it needs one executive who answers for it. Commercial leadership should own the quality of the offer. Product teams supply specifications, operations supports availability and delivery, and finance tests the economics of each promise. Treat it as a go-to-market decision, not a catalog cleanup.
Then make your strengths readable. Service quality, reliability, and trust don't register with an agent unless they show up as evidence. Delivery records, service commitments, certifications, and verified reviews all count when a machine can read them. If your price reflects installation, a longer warranty, or a material that matters, say so plainly. Otherwise an automated comparison may reduce a real difference to a higher price. A business with a strong reputation and thin data carries the biggest risk. The reputation lives in your customers' heads, and the agent can't see it.
Last, measure what counts. Track completed orders, contribution after fulfillment and returns, incorrect recommendations, cancellations, and requests for clarification. More referrals help. Profitable orders fulfilled as promised are the stronger test.
The work isn't glamorous. It won't get a launch event. But it decides whether your business appears when AI shopping agents build the list, and that is a growth question as much as a technical one.
Most figures in this piece describe the United States. Payment systems, consumer protection rules, and AI adoption differ from market to market, so test these ideas against the places where you sell.
If an agent read everything you publish tomorrow, would it pick you?
When Buyers Change How They Choose Suppliers
Growth stalls when the way customers evaluate you shifts faster than your go-to-market model. Aspirations Consulting Group works with executives at those inflection points, whether a company is entering a new channel, resetting its pricing and terms, or reworking how its data, digital presence, and sales approach fit together. We help leadership teams agree on priorities across functions and act on them.
Follow the Shift in Buying as It Develops
Request a complimentary subscription to ACG Strategic Insights™ and join more than 10 million current and aspiring executives who receive practical analysis every weekday.
Thanks for reading!
~ Jerry Justice
Living to Serve, Serving to Lead™




Comments