AI agents could influence the shopping journeys of 1.3 billion users by 2031, but consumers are not yet looking to hand them control of the entire purchase.
Juniper Research expects the number of agentic commerce users to rise from fewer than 300 million in 2026, driven by retailer support, greater familiarity with AI and the development of payment infrastructure.
Its Agentic Commerce Market 2026–2031 study also forecasts that spending through the channel will reach $1.5 trillion in 2030.
Separate research from payments consultancy PSE Consulting suggests much of the early demand may be concentrated at the beginning of the shopping journey.
Among 4,250 consumers in the UK, US, France and Germany who already use AI for online shopping, 74% said they preferred an independent assistant for discovering and comparing products.
However, most still expected recognised marketplaces and platforms to complete transactions, arrange fulfilment and deal with problems after the purchase.
AI may find the product without completing the purchase
PSE’s survey draws a distinction between the service consumers want from an AI assistant and the role they continue to assign to established commerce platforms.
Forty-one per cent preferred a universal assistant such as ChatGPT or Gemini that could search across different providers. A further 33% favoured an independent assistant specialising in an area such as travel, finance or healthcare.
Only 10% wanted the shopping assistant embedded directly inside a retail, travel or delivery platform.
Price was the largest motivation for using AI during shopping, with respondents primarily interested in comparing offers across several merchants rather than searching within the catalogue of one retailer.
That gives independent assistants a potential advantage during product discovery. They can search across several commercial ecosystems without being tied to the inventory or priorities of one platform.
The preference did not carry through to the rest of the transaction.
Nine in 10 respondents expected their use of retail marketplaces and travel platforms to remain steady or increase as AI adoption grew. The corresponding figure for food delivery platforms was 88%.
Chris Jones, managing director at PSE Consulting, said the results challenged predictions that agentic AI would remove marketplaces from the relationship between merchants and consumers.
“There has been a narrative that agentic AI would make marketplaces less relevant by allowing consumers to curate the internet on their terms, rather than relying on marketplaces,” he said.
“What the research suggests instead is that consumers increasingly see discovery and execution as distinct stages of the shopping journey.”
Consumers may welcome help navigating prices and product choices, he added, while still depending on marketplaces for “fulfilment, payments, logistics, customer service and operational trust”.
Cards lead the early payment market
Juniper expects greater availability of payment infrastructure to help agentic commerce reach a larger audience. Card networks have taken an early lead by adapting tokenisation, authentication and fraud controls to transactions initiated by software.
Juniper ranked Mastercard, Visa and Stripe as the three leading agentic commerce payment infrastructure providers after assessing 14 companies.
Mastercard and Visa have both developed systems intended to identify AI agents, protect payment credentials and retain evidence of what the customer authorised an agent to do.
Existing card acceptance also gives AI developers a way to connect with large numbers of merchants without creating an entirely new payment network.
Juniper warned that treating cards as the default could hold back adoption in markets where consumers prefer digital wallets, bank payments or domestic payment systems.
“Cards increasingly support agent payments through tokenisation, but card domination within agentic commerce is not in the market’s best interests, given how important payment preferences are within eCommerce,” said Nick Maynard, vice-president of research at Juniper Research.
“If agentic commerce fails to support local payments, such as digital wallets and account-to-account payments, the market will limit its overall growth potential.”
An agent intended to simplify shopping may offer little benefit if it cannot use the payment method a customer would ordinarily select at checkout.
Payment providers therefore face a choice between integrating a broad range of regional methods and concentrating on the options that cover the largest initial markets.
Trust rises when an agent can spend
The PSE findings also provide some indication of where consumer trust currently stops. Using an AI assistant to search for a product carries limited financial risk. Allowing it to select a seller, approve a payment or resolve a disputed delivery requires a different level of confidence.
Juniper identifies trust as the leading barrier to wider adoption. Consumers must believe that an agent will interpret their instructions correctly, protect personal and financial information and remain within the authority it has been given.
The consequences of an error also become more serious once the agent can complete a transaction.
Incorrect product information or a misunderstood returns policy can lead to an unsuitable purchase. A compromised agent could expose payment credentials or place orders the customer never intended to make.
Merchants will need evidence that an agent is genuine, that the customer gave permission and that the transaction stayed within any limits attached to that permission.
Clear responsibility will also be needed when something goes wrong. Liability could involve the customer, merchant, AI provider, marketplace, wallet or payment company, depending on how the transaction was initiated and approved.
The continued reliance on marketplaces in PSE’s research suggests consumers still associate established platforms with the ability to resolve those problems.
Discovery becomes a new battleground
The shift towards AI-led product discovery could affect merchants before fully autonomous payments become common.
Jones said the “competitive battleground is moving upstream into the AI discovery layer”, with merchants increasingly dependent on whether an assistant includes their products before the consumer visits a marketplace.
Agents need structured and current information about prices, availability, delivery times, product specifications and returns.
Larger retailers with established application programming interfaces and well-organised catalogues may find it easier to supply that information. Smaller merchants could be overlooked when their product data is incomplete, inconsistently labelled or difficult for an agent to access.
Juniper’s research similarly warns that poor data can produce inaccurate rankings and unsuitable recommendations, even when the agent itself has sophisticated reasoning capabilities.
The development of protocols connecting agents, merchants and payment providers is intended to reduce that fragmentation.
Google’s Universal Commerce Protocol, for example, provides a shared format for functions including product discovery, shopping baskets, identity, checkout and ordering.
The protocol could allow an assistant to connect with multiple merchants without each pairing requiring a separate technical integration.
Autonomy is likely to arrive in stages
Juniper describes agentic commerce as a progression rather than a sudden move to purchases completed without human involvement.
An assistant may initially search for products and prepare a basket before asking the customer for approval. Later uses could include negotiating offers, managing recurring orders or completing purchases within agreed limits.
Regular and lower-value transactions may offer an easier starting point. An agent could amend a grocery delivery when a customer is away, replace an unavailable household product or find a cheaper subscription without receiving fresh instructions each time.
Consumers are likely to retain more direct control over expensive, personal or difficult-to-reverse purchases. PSE’s research points to a similar division of responsibilities: independent AI assistants may gain influence over what consumers consider, while marketplaces continue to handle the transaction and everything that follows it.
The immediate contest is therefore over which products AI presents before the customer reaches a familiar checkout.
