Shopping through an AI assistant changes where a retailer can influence a sale. A customer can describe a need, set a budget and delegate the search. By the time a merchant receives an order, the comparison that determined the choice may already be over. Brands therefore need to make their offers understandable and usable outside their own websites.
The change involves more than a new source of traffic. An assistant can retain the customer's preferences, evaluate competing offers and carry a decision through to payment. The retailer still supplies the product and service, but another platform may manage the conversation that shapes the purchase.
Shopping through an AI assistant changes where a retailer can influence a sale. A customer can describe a need, set a budget and delegate the search. By the time a merchant receives an order, the comparison that determined the choice may already be over. Brands therefore need to make their offers understandable and usable outside their own websites.
The change involves more than a new source of traffic. An assistant can retain the customer's preferences, evaluate competing offers and carry a decision through to payment. The retailer still supplies the product and service, but another platform may manage the conversation that shapes the purchase.
Shopping through an AI assistant changes where a retailer can influence a sale. A customer can describe a need, set a budget and delegate the search. By the time a merchant receives an order, the comparison that determined the choice may already be over. Brands therefore need to make their offers understandable and usable outside their own websites.
The change involves more than a new source of traffic. An assistant can retain the customer's preferences, evaluate competing offers and carry a decision through to payment. The retailer still supplies the product and service, but another platform may manage the conversation that shapes the purchase.
Shopping through an AI assistant changes where a retailer can influence a sale. A customer can describe a need, set a budget and delegate the search. By the time a merchant receives an order, the comparison that determined the choice may already be over. Brands therefore need to make their offers understandable and usable outside their own websites.
The change involves more than a new source of traffic. An assistant can retain the customer's preferences, evaluate competing offers and carry a decision through to payment. The retailer still supplies the product and service, but another platform may manage the conversation that shapes the purchase.
Consumers are delegating selectively
Interest in AI assistance does not mean shoppers are ready to surrender every decision. A Deloitte survey of 13,500 European consumers found that 31% were open to letting AI complete the final purchase, while 47% were unwilling. Research and product comparison remain more familiar uses than delegating payment.
Retailers nevertheless expect behavior to change. In Checkout.com research published in 2026, 72% of merchants in the UK and US agreed that consumers would adopt agent-led shopping faster than most merchants were prepared to accommodate. A cautious customer today can still become a regular agent user as the experience improves.
That makes partial delegation commercially important. A retailer may lose a place on the shortlist even when the customer ultimately presses the payment button. Preparation should cover product discovery and comparison as well as checkout.
Platforms are choosing different ways to participate
Meta introduced Muse on September 8, 2026. According to Meta, the assistant uses a dedicated cloud environment, remembers preferences and requests approval before sensitive actions, including purchases. The company says conversations and data from that environment are not shared with its advertising systems.
The commercial response has been divided. On September 20, Amazon warned Muse users that access through an unauthorized agent violated its Conditions of Use. GeekWire reported Amazon's concerns about identification and credential handling. Amazon also has its own Buy for Me service and generated more than $68 billion in advertising revenue in 2025.
The legal position also has limits. On August 4, 2026, the Ninth Circuit vacated a preliminary injunction in Amazon's case against Perplexity. On the facts considered, the court treated the customer as accessing Amazon with an AI tool. That decision did not settle every dispute over agents or remove contractual restrictions.
For a brand, the practical question is which activities it will support and on what terms. An access decision affects sales, customer experience and security, so it needs input from the teams responsible for all three.
Recognizing an agent can preserve access to customers
Merchants have alternatives to treating every automated visit alike. Visa's Trusted Agent Protocol, developed with Cloudflare, uses cryptographic signatures to help recognize trusted agents and communicate shopping intent. It provides a basis for distinguishing legitimate commercial activity from malicious automation.
Mastercard is also developing trusted agent payments. The company said in June 2026 that all its European issuers were enabled for Agent Pay at the network level. Network readiness does not mean that every merchant has implemented an identical checkout experience.
Access can be decided separately for browsing, building a cart and making payment. A merchant might allow an identified assistant to retrieve product details while requiring additional verification at checkout. The rules should reflect the action being requested and the authority the customer has granted.
Sales records will reveal less about the decision
An order and a browsing history are different kinds of evidence. The Keyword reports that Google Analytics and custom pixels do not fire in Meta's direct checkout on Shopify. Checkout-started and checkout-completed events are passed server to server. The merchant records the sale without receiving the full behavioral trail behind it.
Marketing teams should account for this when interpreting conversion and attribution reports. A missing website session does not establish that a customer had no contact with the brand. Equally, a completed order does not explain which comparison persuaded the assistant to recommend it.
Reporting needs to distinguish what the business directly observes from what it infers. Server-side order events and channel reporting become more useful when teams understand which parts of the journey each integration actually exposes.
Service terms become part of product selection
A machine can compare delivery commitments, return eligibility and warranty terms alongside price. Ambiguity makes an offer harder to assess. A retailer whose website, app and marketplace listings describe different conditions should make those differences explicit and reconcile discrepancies that serve no purpose.
Payment services can introduce their own expectations. Stripe describes purchase protections through Link for qualifying transactions, including damaged or lost items, price drops, no-fee returns and return guarantees. Such benefits give an assistant additional conditions to consider when evaluating a merchant's offer.
Accountability also shapes trust. In a Riskified survey of US and UK shoppers, 50.8% assigned responsibility for unauthorized or erroneous agent purchases to the AI platform, 23.2% to the retailer or brand and 18.7% to the consumer. These are consumer expectations, rather than a statement of legal liability.
Clear terms help before and after payment. Customers need to know how an order can be corrected, canceled or returned, including when an assistant initiated it. Product, checkout and service information should tell the same story.
Loyalty needs benefits an assistant can use
Brand preference still influences delegated shopping. Accenture's Talk to My AI Agent research found that 56% of consumers would specify brands for an agent to consider. Yet 37% of behaviorally loyal consumers would allow a switch if another brand better met their needs.
The implication is that recognition alone may not secure the order. A preferred brand needs an offer the assistant can evaluate: dependable availability, clear pricing and benefits the customer can actually redeem. Loyalty should remain useful when the customer delegates the transaction.
A portable loyalty identifier can help carry eligible benefits into an agent checkout. The same consideration applies to returns, exchanges and service requests. Maintaining an account relationship is more valuable when customers can use it through their chosen assistant.
What retailers should review first
Start with existing platform settings. Shopify products can already be shared with Meta through Catalog by default, and eligible stores can have direct checkout enabled. Merchants should establish which channels are active, what customers can do there and which controls are available.
Next, document access rules for each shopping stage. Agree how verified assistants will be recognized and what additional checks a payment requires. Review customer-facing terms at the same time, so technical permissions and commercial promises remain consistent.
Product data should describe the offer accurately enough for comparison. Delivery, returns and warranty information need structured, machine-readable representations alongside clear customer explanations. Loyalty benefits should be testable through the delegated journey, rather than assumed to work because they function on the website.
Finally, review the evidence each agent channel sends back. Establish how orders, cancellations and returns will appear in reporting before evaluating the channel's performance. The useful starting point is a purchase journey the brand can support, explain and measure with the information actually available.
SHOMEDIA
SHOMEDIA
SHOMEDIA
SHOMEDIA
Interest in AI assistance does not mean shoppers are ready to surrender every decision. A Deloitte survey of 13,500 European consumers found that 31% were open to letting AI complete the final purchase, while 47% were unwilling. Research and product comparison remain more familiar uses than delegating payment.
Retailers nevertheless expect behavior to change. In Checkout.com research published in 2026, 72% of merchants in the UK and US agreed that consumers would adopt agent-led shopping faster than most merchants were prepared to accommodate. A cautious customer today can still become a regular agent user as the experience improves.
That makes partial delegation commercially important. A retailer may lose a place on the shortlist even when the customer ultimately presses the payment button. Preparation should cover product discovery and comparison as well as checkout.
Platforms are choosing different ways to participate
Meta introduced Muse on September 8, 2026. According to Meta, the assistant uses a dedicated cloud environment, remembers preferences and requests approval before sensitive actions, including purchases. The company says conversations and data from that environment are not shared with its advertising systems.
The commercial response has been divided. On September 20, Amazon warned Muse users that access through an unauthorized agent violated its Conditions of Use. GeekWire reported Amazon's concerns about identification and credential handling. Amazon also has its own Buy for Me service and generated more than $68 billion in advertising revenue in 2025.
The legal position also has limits. On August 4, 2026, the Ninth Circuit vacated a preliminary injunction in Amazon's case against Perplexity. On the facts considered, the court treated the customer as accessing Amazon with an AI tool. That decision did not settle every dispute over agents or remove contractual restrictions.
For a brand, the practical question is which activities it will support and on what terms. An access decision affects sales, customer experience and security, so it needs input from the teams responsible for all three.
Recognizing an agent can preserve access to customers
Merchants have alternatives to treating every automated visit alike. Visa's Trusted Agent Protocol, developed with Cloudflare, uses cryptographic signatures to help recognize trusted agents and communicate shopping intent. It provides a basis for distinguishing legitimate commercial activity from malicious automation.
Mastercard is also developing trusted agent payments. The company said in June 2026 that all its European issuers were enabled for Agent Pay at the network level. Network readiness does not mean that every merchant has implemented an identical checkout experience.
Access can be decided separately for browsing, building a cart and making payment. A merchant might allow an identified assistant to retrieve product details while requiring additional verification at checkout. The rules should reflect the action being requested and the authority the customer has granted.
Sales records will reveal less about the decision
An order and a browsing history are different kinds of evidence. The Keyword reports that Google Analytics and custom pixels do not fire in Meta's direct checkout on Shopify. Checkout-started and checkout-completed events are passed server to server. The merchant records the sale without receiving the full behavioral trail behind it.
Marketing teams should account for this when interpreting conversion and attribution reports. A missing website session does not establish that a customer had no contact with the brand. Equally, a completed order does not explain which comparison persuaded the assistant to recommend it.
Reporting needs to distinguish what the business directly observes from what it infers. Server-side order events and channel reporting become more useful when teams understand which parts of the journey each integration actually exposes.
Service terms become part of product selection
A machine can compare delivery commitments, return eligibility and warranty terms alongside price. Ambiguity makes an offer harder to assess. A retailer whose website, app and marketplace listings describe different conditions should make those differences explicit and reconcile discrepancies that serve no purpose.
Payment services can introduce their own expectations. Stripe describes purchase protections through Link for qualifying transactions, including damaged or lost items, price drops, no-fee returns and return guarantees. Such benefits give an assistant additional conditions to consider when evaluating a merchant's offer.
Accountability also shapes trust. In a Riskified survey of US and UK shoppers, 50.8% assigned responsibility for unauthorized or erroneous agent purchases to the AI platform, 23.2% to the retailer or brand and 18.7% to the consumer. These are consumer expectations, rather than a statement of legal liability.
Clear terms help before and after payment. Customers need to know how an order can be corrected, canceled or returned, including when an assistant initiated it. Product, checkout and service information should tell the same story.
Loyalty needs benefits an assistant can use
Brand preference still influences delegated shopping. Accenture's Talk to My AI Agent research found that 56% of consumers would specify brands for an agent to consider. Yet 37% of behaviorally loyal consumers would allow a switch if another brand better met their needs.
The implication is that recognition alone may not secure the order. A preferred brand needs an offer the assistant can evaluate: dependable availability, clear pricing and benefits the customer can actually redeem. Loyalty should remain useful when the customer delegates the transaction.
A portable loyalty identifier can help carry eligible benefits into an agent checkout. The same consideration applies to returns, exchanges and service requests. Maintaining an account relationship is more valuable when customers can use it through their chosen assistant.
What retailers should review first
Start with existing platform settings. Shopify products can already be shared with Meta through Catalog by default, and eligible stores can have direct checkout enabled. Merchants should establish which channels are active, what customers can do there and which controls are available.
Next, document access rules for each shopping stage. Agree how verified assistants will be recognized and what additional checks a payment requires. Review customer-facing terms at the same time, so technical permissions and commercial promises remain consistent.
Product data should describe the offer accurately enough for comparison. Delivery, returns and warranty information need structured, machine-readable representations alongside clear customer explanations. Loyalty benefits should be testable through the delegated journey, rather than assumed to work because they function on the website.
Finally, review the evidence each agent channel sends back. Establish how orders, cancellations and returns will appear in reporting before evaluating the channel's performance. The useful starting point is a purchase journey the brand can support, explain and measure with the information actually available.
SHOMEDIA
SHOMEDIA
SHOMEDIA
SHOMEDIA
Interest in AI assistance does not mean shoppers are ready to surrender every decision. A Deloitte survey of 13,500 European consumers found that 31% were open to letting AI complete the final purchase, while 47% were unwilling. Research and product comparison remain more familiar uses than delegating payment.
Retailers nevertheless expect behavior to change. In Checkout.com research published in 2026, 72% of merchants in the UK and US agreed that consumers would adopt agent-led shopping faster than most merchants were prepared to accommodate. A cautious customer today can still become a regular agent user as the experience improves.
That makes partial delegation commercially important. A retailer may lose a place on the shortlist even when the customer ultimately presses the payment button. Preparation should cover product discovery and comparison as well as checkout.
Platforms are choosing different ways to participate
Meta introduced Muse on September 8, 2026. According to Meta, the assistant uses a dedicated cloud environment, remembers preferences and requests approval before sensitive actions, including purchases. The company says conversations and data from that environment are not shared with its advertising systems.
The commercial response has been divided. On September 20, Amazon warned Muse users that access through an unauthorized agent violated its Conditions of Use. GeekWire reported Amazon's concerns about identification and credential handling. Amazon also has its own Buy for Me service and generated more than $68 billion in advertising revenue in 2025.
The legal position also has limits. On August 4, 2026, the Ninth Circuit vacated a preliminary injunction in Amazon's case against Perplexity. On the facts considered, the court treated the customer as accessing Amazon with an AI tool. That decision did not settle every dispute over agents or remove contractual restrictions.
For a brand, the practical question is which activities it will support and on what terms. An access decision affects sales, customer experience and security, so it needs input from the teams responsible for all three.
Recognizing an agent can preserve access to customers
Merchants have alternatives to treating every automated visit alike. Visa's Trusted Agent Protocol, developed with Cloudflare, uses cryptographic signatures to help recognize trusted agents and communicate shopping intent. It provides a basis for distinguishing legitimate commercial activity from malicious automation.
Mastercard is also developing trusted agent payments. The company said in June 2026 that all its European issuers were enabled for Agent Pay at the network level. Network readiness does not mean that every merchant has implemented an identical checkout experience.
Access can be decided separately for browsing, building a cart and making payment. A merchant might allow an identified assistant to retrieve product details while requiring additional verification at checkout. The rules should reflect the action being requested and the authority the customer has granted.
Sales records will reveal less about the decision
An order and a browsing history are different kinds of evidence. The Keyword reports that Google Analytics and custom pixels do not fire in Meta's direct checkout on Shopify. Checkout-started and checkout-completed events are passed server to server. The merchant records the sale without receiving the full behavioral trail behind it.
Marketing teams should account for this when interpreting conversion and attribution reports. A missing website session does not establish that a customer had no contact with the brand. Equally, a completed order does not explain which comparison persuaded the assistant to recommend it.
Reporting needs to distinguish what the business directly observes from what it infers. Server-side order events and channel reporting become more useful when teams understand which parts of the journey each integration actually exposes.
Service terms become part of product selection
A machine can compare delivery commitments, return eligibility and warranty terms alongside price. Ambiguity makes an offer harder to assess. A retailer whose website, app and marketplace listings describe different conditions should make those differences explicit and reconcile discrepancies that serve no purpose.
Payment services can introduce their own expectations. Stripe describes purchase protections through Link for qualifying transactions, including damaged or lost items, price drops, no-fee returns and return guarantees. Such benefits give an assistant additional conditions to consider when evaluating a merchant's offer.
Accountability also shapes trust. In a Riskified survey of US and UK shoppers, 50.8% assigned responsibility for unauthorized or erroneous agent purchases to the AI platform, 23.2% to the retailer or brand and 18.7% to the consumer. These are consumer expectations, rather than a statement of legal liability.
Clear terms help before and after payment. Customers need to know how an order can be corrected, canceled or returned, including when an assistant initiated it. Product, checkout and service information should tell the same story.
Loyalty needs benefits an assistant can use
Brand preference still influences delegated shopping. Accenture's Talk to My AI Agent research found that 56% of consumers would specify brands for an agent to consider. Yet 37% of behaviorally loyal consumers would allow a switch if another brand better met their needs.
The implication is that recognition alone may not secure the order. A preferred brand needs an offer the assistant can evaluate: dependable availability, clear pricing and benefits the customer can actually redeem. Loyalty should remain useful when the customer delegates the transaction.
A portable loyalty identifier can help carry eligible benefits into an agent checkout. The same consideration applies to returns, exchanges and service requests. Maintaining an account relationship is more valuable when customers can use it through their chosen assistant.
What retailers should review first
Start with existing platform settings. Shopify products can already be shared with Meta through Catalog by default, and eligible stores can have direct checkout enabled. Merchants should establish which channels are active, what customers can do there and which controls are available.
Next, document access rules for each shopping stage. Agree how verified assistants will be recognized and what additional checks a payment requires. Review customer-facing terms at the same time, so technical permissions and commercial promises remain consistent.
Product data should describe the offer accurately enough for comparison. Delivery, returns and warranty information need structured, machine-readable representations alongside clear customer explanations. Loyalty benefits should be testable through the delegated journey, rather than assumed to work because they function on the website.
Finally, review the evidence each agent channel sends back. Establish how orders, cancellations and returns will appear in reporting before evaluating the channel's performance. The useful starting point is a purchase journey the brand can support, explain and measure with the information actually available.
SHOMEDIA
SHOMEDIA
SHOMEDIA
SHOMEDIA
Interest in AI assistance does not mean shoppers are ready to surrender every decision. A Deloitte survey of 13,500 European consumers found that 31% were open to letting AI complete the final purchase, while 47% were unwilling. Research and product comparison remain more familiar uses than delegating payment.
Retailers nevertheless expect behavior to change. In Checkout.com research published in 2026, 72% of merchants in the UK and US agreed that consumers would adopt agent-led shopping faster than most merchants were prepared to accommodate. A cautious customer today can still become a regular agent user as the experience improves.
That makes partial delegation commercially important. A retailer may lose a place on the shortlist even when the customer ultimately presses the payment button. Preparation should cover product discovery and comparison as well as checkout.
Platforms are choosing different ways to participate
Meta introduced Muse on September 8, 2026. According to Meta, the assistant uses a dedicated cloud environment, remembers preferences and requests approval before sensitive actions, including purchases. The company says conversations and data from that environment are not shared with its advertising systems.
The commercial response has been divided. On September 20, Amazon warned Muse users that access through an unauthorized agent violated its Conditions of Use. GeekWire reported Amazon's concerns about identification and credential handling. Amazon also has its own Buy for Me service and generated more than $68 billion in advertising revenue in 2025.
The legal position also has limits. On August 4, 2026, the Ninth Circuit vacated a preliminary injunction in Amazon's case against Perplexity. On the facts considered, the court treated the customer as accessing Amazon with an AI tool. That decision did not settle every dispute over agents or remove contractual restrictions.
For a brand, the practical question is which activities it will support and on what terms. An access decision affects sales, customer experience and security, so it needs input from the teams responsible for all three.
Recognizing an agent can preserve access to customers
Merchants have alternatives to treating every automated visit alike. Visa's Trusted Agent Protocol, developed with Cloudflare, uses cryptographic signatures to help recognize trusted agents and communicate shopping intent. It provides a basis for distinguishing legitimate commercial activity from malicious automation.
Mastercard is also developing trusted agent payments. The company said in June 2026 that all its European issuers were enabled for Agent Pay at the network level. Network readiness does not mean that every merchant has implemented an identical checkout experience.
Access can be decided separately for browsing, building a cart and making payment. A merchant might allow an identified assistant to retrieve product details while requiring additional verification at checkout. The rules should reflect the action being requested and the authority the customer has granted.
Sales records will reveal less about the decision
An order and a browsing history are different kinds of evidence. The Keyword reports that Google Analytics and custom pixels do not fire in Meta's direct checkout on Shopify. Checkout-started and checkout-completed events are passed server to server. The merchant records the sale without receiving the full behavioral trail behind it.
Marketing teams should account for this when interpreting conversion and attribution reports. A missing website session does not establish that a customer had no contact with the brand. Equally, a completed order does not explain which comparison persuaded the assistant to recommend it.
Reporting needs to distinguish what the business directly observes from what it infers. Server-side order events and channel reporting become more useful when teams understand which parts of the journey each integration actually exposes.
Service terms become part of product selection
A machine can compare delivery commitments, return eligibility and warranty terms alongside price. Ambiguity makes an offer harder to assess. A retailer whose website, app and marketplace listings describe different conditions should make those differences explicit and reconcile discrepancies that serve no purpose.
Payment services can introduce their own expectations. Stripe describes purchase protections through Link for qualifying transactions, including damaged or lost items, price drops, no-fee returns and return guarantees. Such benefits give an assistant additional conditions to consider when evaluating a merchant's offer.
Accountability also shapes trust. In a Riskified survey of US and UK shoppers, 50.8% assigned responsibility for unauthorized or erroneous agent purchases to the AI platform, 23.2% to the retailer or brand and 18.7% to the consumer. These are consumer expectations, rather than a statement of legal liability.
Clear terms help before and after payment. Customers need to know how an order can be corrected, canceled or returned, including when an assistant initiated it. Product, checkout and service information should tell the same story.
Loyalty needs benefits an assistant can use
Brand preference still influences delegated shopping. Accenture's Talk to My AI Agent research found that 56% of consumers would specify brands for an agent to consider. Yet 37% of behaviorally loyal consumers would allow a switch if another brand better met their needs.
The implication is that recognition alone may not secure the order. A preferred brand needs an offer the assistant can evaluate: dependable availability, clear pricing and benefits the customer can actually redeem. Loyalty should remain useful when the customer delegates the transaction.
A portable loyalty identifier can help carry eligible benefits into an agent checkout. The same consideration applies to returns, exchanges and service requests. Maintaining an account relationship is more valuable when customers can use it through their chosen assistant.
What retailers should review first
Start with existing platform settings. Shopify products can already be shared with Meta through Catalog by default, and eligible stores can have direct checkout enabled. Merchants should establish which channels are active, what customers can do there and which controls are available.
Next, document access rules for each shopping stage. Agree how verified assistants will be recognized and what additional checks a payment requires. Review customer-facing terms at the same time, so technical permissions and commercial promises remain consistent.
Product data should describe the offer accurately enough for comparison. Delivery, returns and warranty information need structured, machine-readable representations alongside clear customer explanations. Loyalty benefits should be testable through the delegated journey, rather than assumed to work because they function on the website.
Finally, review the evidence each agent channel sends back. Establish how orders, cancellations and returns will appear in reporting before evaluating the channel's performance. The useful starting point is a purchase journey the brand can support, explain and measure with the information actually available.
Комментарии