Agentic Commerce Explained: How AI Shopping Agents Will Reshape Search, Marketing and Sales
The next customer to approach your business may be software acting for a person.
Instead of opening a search engine and comparing websites manually, a customer can tell an AI agent: “Find a reliable office chair under £500, available this week, with strong reviews and a five-year warranty.” The agent can interpret the requirement, research the market, compare evidence and, with permission, complete the purchase.
This is agentic commerce, sometimes called agent commerce: autonomous or semi-autonomous AI agents undertaking parts of the buying journey for customers or businesses. The supplied background paper identifies the essential shift accurately, from search-led journeys to intent-led execution, with AI mediating discovery, evaluation, purchase and service. The infrastructure is no longer hypothetical.
What is agentic commerce?
A conventional chatbot answers questions. An AI commerce agent pursues a goal.
The customer defines an outcome and constraints such as price, quality, delivery time, preferred brands or sustainability standards. The agent then uses connected tools, product feeds, APIs and payment services to complete a sequence of actions.
Autonomy varies. An agent might reorder routine products within an agreed budget, while a high-value purchase may require approval before checkout. Agentic commerce covers a spectrum from assisted shopping to delegated purchasing.
OpenAI’s Agentic Commerce Protocol connects merchants, AI interfaces and buyers. Google’s Universal Commerce Protocol connects consumer surfaces, retailer systems and payment providers. Stripe, Visa and Mastercard are developing the payment tokens, permissions and controls agents need to transact safely. Santander and Mastercard completed a live end-to-end payment through an AI agent in March 2026, within predefined limits and a regulated banking framework.
Why agentic commerce matters now
AI-assisted shopping is already influencing measurable demand. Adobe reported that traffic from generative AI tools to US retail websites rose 693.4 per cent year on year during the 2025 holiday season. Those referrals converted 31 per cent better than other traffic sources, while revenue per visit increased 254 per cent. In the first quarter of 2026, AI-sourced retail traffic remained 393 per cent higher year on year.
People arriving through an AI assistant may carry greater intent because a detailed conversation has already clarified their needs and filtered unsuitable options.
Willingness comes with conditions. Visa’s research across the United States, Australia and New Zealand found that roughly one in three respondents expected to use AI shopping agents regularly. Around two-thirds valued time and price benefits, but nearly nine in ten wanted transparency about agent decisions, and about half said they would stop using them if they lost control. The opportunity and the trust requirement are arriving together.
Discovery changes from keywords to specifications
Search marketing has traditionally focused on the words people type. Agentic discovery begins with the outcome they describe. A search for “best running shoes” might become an agent brief covering budget, gait, terrain, delivery date and material restrictions. The agent can eliminate unsuitable products before the customer sees them.
For marketers, visibility increasingly depends on whether an AI system can understand the offer accurately. OpenAI’s merchant guidance asks for structured feeds containing titles, descriptions, images, price and availability. Google’s UCP is intended to support real-time inventory, dynamic pricing and transactions within conversational experiences.
SEO remains important, but it gains a machine-commerce layer. Product schema, catalogue quality, consistent identifiers, clear policies, accurate stock, review data and accessible APIs become commercial media assets. A brilliant campaign cannot compensate for an agent being unable to verify whether a product is suitable or deliverable.
Consideration becomes evidence-led
Agents are highly effective at comparing measurable attributes. Price, ratings, delivery speed, warranty, returns, certifications, service levels and total cost of ownership can be evaluated side by side. Vague claims such as “premium quality” become less useful unless the business supplies evidence showing what premium means.
Branding is not obsolete. Its job expands. Strong brands must create preference among people and legibility among machines. Reputation, positioning and emotional meaning can enter the customer’s instructions, while structured proof helps the agent justify its recommendation.
This changes the relationship between brand promise and operational performance. If a business claims to provide next-day delivery but repeatedly misses the deadline, an agent can identify the discrepancy. If a product is promoted as sustainable without recognised evidence, the claim may be excluded from the agent’s evaluation. In agent commerce, marketing claims increasingly need to survive machine scrutiny.
Purchase becomes permissioned action
The greatest shift arrives when an agent moves from recommendation to transaction. Emerging payment systems are designed around restricted authority rather than unlimited access. OpenAI describes payment tokens limited to a named merchant and amount. Google’s UCP includes cryptographic proof of consent, while Stripe’s Shared Payment Tokens can be constrained by seller, time and value.
Customers need to understand what an agent may buy, how much it may spend, which data it may share and how an action can be corrected. A person might authorise an agent to buy household essentials up to £100 a month, for example, while still requiring approval for new brands or price increases.
For businesses, fulfilment becomes part of marketing. A late delivery, difficult return or disputed charge can influence the next recommendation. Customer experience is no longer simply remembered by the customer. It may become data used by their agent.
Loyalty faces a new test
Agentic commerce can strengthen loyalty when an agent recognises memberships, saved preferences and positive experiences. It can also weaken habitual purchasing by comparing the market every time.
A customer may like a brand, but their agent may find a comparable product with better availability, lower lifetime cost or simpler returns. Loyalty programmes offering genuine value will be easier for agents to defend than schemes built around friction.
This means loyalty benefits must become machine-readable. Member prices, free delivery, warranties, points, access privileges and personalised offers need to be visible within the commercial data available to the agent.
The same logic is likely to influence B2B purchasing, where procurement agents could compare approved suppliers, request quotations and reorder components. Businesses whose offer, availability and terms are machine-readable will have an advantage. This is an inference from the emerging consumer and payment infrastructure.
Seven actions marketers should take now
- Audit agent visibility. Ask leading AI assistants to recommend products or services in your category using realistic customer requirements. Record whether your brand appears, what is said and which facts are missing.
- Improve the commercial data layer. Align product feeds, structured data, descriptions, pricing, availability, delivery, returns and service information. Stripe advises keeping inventory and prices current because stale feeds can create failed purchases and mismatched quotations.
- Make the proposition easier to evaluate. Define target use cases, eligibility, packages, outcomes, evidence and exclusions. Service businesses need structured offers too, even where the final sale requires consultation.
- Build verifiable trust. Strengthen review quality, credentials, guarantees, sustainability evidence, case studies and service standards. Remove claims an agent cannot substantiate.
- Design permission and escalation rules. Decide where autonomous purchase is acceptable, where confirmation is required and how refunds, exceptions, fraud and human intervention will work.
- Update measurement. Track AI referral traffic, recommendation inclusion, catalogue errors, agent-assisted conversions, checkout completion, returns, disputes and customer lifetime value. Sessions and clicks tell less of the story when decisions happen inside an agent.
- Give someone ownership. Agentic commerce crosses marketing, ecommerce, IT, finance, legal, customer service and operations. Treating it as a minor experimental channel will leave gaps between promise and fulfilment.
The next distribution layer
Agentic commerce is more than ecommerce with an AI assistant added to the front. It creates a new distribution layer between customer demand and business supply.
The winning brands will still need creativity, reputation and human relevance. They must also be easy for machines to discover, understand, verify and transact with. That means better data, clearer propositions, stronger evidence, dependable service and carefully designed permissions.
An AI agent does not owe a brand attention. It earns recommendations by matching the customer’s intent.
Businesses that prepare now can influence that emerging decision layer. Those that wait may discover that their website is still online, their campaigns are still running and their next customer has quietly delegated the choice elsewhere.