You need a carry-on suitcase for a trip next Friday. It must fit Air Canada’s cabin limits, arrive within four days, cost less than $250 and have a reliable warranty.
Today, you might open Google, scan a few retailers, compare dimensions, read reviews and check delivery dates. An AI shopping agent can handle much of that work. You give it the constraints. It searches available products, removes the options that do not fit, compares the remaining choices and asks for approval before placing the order.
You may never visit the store’s homepage. You may not see its navigation, campaign banner or carefully designed category page. The agent might encounter the brand first, evaluate the offer and decide whether the product belongs on your shortlist.
The agent doesn’t browse. It picks.
That buying model is called agentic commerce. Payment companies, commerce platforms and major retailers are building the systems needed to support it. Shoppers have also started using AI for product research at a rate that businesses can measure.
The shift deserves attention, but it also needs a sober explanation. Most purchases still involve a person. Many AI shopping experiences still stop before payment. Trust, data quality and merchant support remain uneven. Businesses have time to prepare, though waiting for the channel to mature before cleaning up product data creates a harder job later.
What is agentic commerce?
Agentic commerce is a form of buying and selling in which an AI agent acts on behalf of a consumer or business. The agent can research options, compare products, check availability, build a cart and, with the buyer’s permission, complete a transaction.
The person still sets the goal and the boundaries. Those instructions might include a budget, required features, preferred brands, delivery timing or an approval threshold. The agent handles the work inside those limits.
Visa describes agentic commerce as a range of activities that can include product discovery, comparison, checkout initiation and completed transactions with user permission. Stripe’s current Agentic Commerce Protocol documentation covers catalog browsing, cart management, delegated payment, buyer authorization, orders and refunds.
An AI shopping assistant helps you decide. An AI shopping agent can take action after the decision.
How agentic commerce differs from online shopping today
Traditional ecommerce asks the shopper to operate the interface. The shopper searches, clicks, filters, compares and checks out. Retailers optimize each step to keep that person moving toward a sale.
Agentic commerce gives part of that work to software.
| Traditional ecommerce | Agentic commerce |
|---|---|
| The shopper enters keywords | The shopper describes an outcome and constraints |
| The shopper visits several pages or stores | The agent gathers options across available sources |
| Filters narrow a catalog | The agent evaluates attributes against the request |
| The shopper compares products | The agent creates a shortlist and explains the tradeoffs |
| The shopper fills out checkout | The agent may prepare or complete checkout with permission |
| The retailer designs for clicks and conversion | The retailer must also provide data an agent can interpret and trust |
The old path does not disappear. Some shoppers enjoy browsing. Many purchases involve taste, emotion or uncertainty that a product feed cannot settle. Websites, stores, paid media and human recommendations will continue to influence demand.
The new path adds another buyer interface. A business may now need to persuade a person and supply an agent with enough evidence to recommend the same product.
How an AI shopping agent completes a purchase
The process varies by platform, merchant and payment provider, but most agentic shopping flows include five stages.
1. The buyer gives the agent a goal
The request sounds more like a brief than a search query:
Find a waterproof commuter backpack under $180 that fits a 16-inch laptop, has a luggage sleeve and can arrive in Vancouver by Thursday.
The agent extracts the hard constraints and softer preferences. It may ask a follow-up question if the buyer has left out a detail that could change the result.
2. The agent discovers eligible products
The agent reads product feeds, merchant data, web pages or connected commerce systems. It needs accurate information about price, availability, variants, shipping, specifications and policies.
Discovery can fail before comparison begins. An agent cannot recommend a product it cannot find, parse or verify.
3. The agent compares the choices
The agent removes products that fail the buyer’s requirements. It may weigh price against quality, delivery speed, ratings, warranty terms or brand preference. A strong answer gives the buyer a short list with a reason for each selection instead of a page of blue links.
4. The buyer authorizes the action
The agent may ask the buyer to approve the item, total price, shipping choice or payment. Buyers and businesses can also set rules in advance, such as a spending cap or mandatory approval above a set amount.
Visa’s June 2026 announcement with OpenAI describes controls such as spending limits and approval thresholds. These guardrails keep the buyer in command while the agent carries out the task.
5. The agent completes or hands off the purchase
Some systems let the agent create a cart and send the shopper to the merchant. Others support payment inside the AI experience. The merchant still needs to confirm the order, handle fulfillment and support returns.
The difference between a useful demo and a dependable sales channel sits in this last stage. Payments, identity, fraud controls and order updates must work across the agent, merchant and payment network.
Why agentic commerce is moving now
AI product research has existed for years. Three changes are turning it into a commerce channel.
AI shopping traffic has become measurable
Adobe analyzed more than one trillion visits to U.S. retail sites and found that traffic from AI sources grew 393% year over year in the first quarter of 2026. In March, AI-referred traffic converted 42% better than non-AI traffic. Adobe also found that 39% of surveyed consumers had used AI for online shopping.
By May 2026, Adobe reported that AI traffic to U.S. retail sites had grown 138% year over year. Those visits converted 54% better and produced 53% more value than visits from non-AI sources.
These numbers cover AI-referred visits, not purchases completed by autonomous agents. They still show that shoppers who use AI arrive with stronger intent and growing confidence.
Commerce platforms now have shared protocols
An AI agent cannot build a separate integration for each retailer it encounters. Shared protocols give agents, merchants and payment providers a common way to exchange product, cart, identity and transaction information.
Google launched the Universal Commerce Protocol in January 2026 with support from Shopify, Etsy, Wayfair, Target, Walmart and payment companies. Google designed UCP to cover discovery, purchase and post-purchase support.
Google expanded UCP in March with catalog access for current price, inventory and variant details, multi-item carts and identity linking for loyalty benefits. Stripe’s ACP supports another set of building blocks for agent-driven checkout and delegated payment.
These standards will change. Their existence still lowers the cost of connecting agents and merchants.
Payment networks are building trust controls
A retailer once tried to block bots from checkout. Agentic commerce requires the retailer to recognize an authorized agent while stopping fraud and unwanted automation.
Visa, Mastercard, Stripe and other payment providers now work on agent identity, tokenized payment credentials, buyer intent and transaction controls. A payment system needs to answer three questions: Which agent initiated the purchase? Did the buyer authorize it? Did the agent stay inside the buyer’s limits?
The checkout cannot scale without clear answers.
Is agentic commerce real or hype?
Both concerns have merit.
The infrastructure is real. Google, Stripe, Visa and Mastercard have released protocols, commerce tools or payment systems. Retailers and commerce platforms have joined those programs. AI-referred shopping traffic has grown across retail and travel.
Adoption remains early. Many shoppers use AI to research products and then finish the purchase on a retailer’s site. Merchant coverage varies. Product data goes stale. An agent may fail to understand a subjective preference or recommend an option from an incomplete set.
The most useful planning assumption sits between dismissal and panic: agentic commerce has become a new channel, but it has not replaced ecommerce.
Businesses do not need to rebuild every system this quarter. They should fix the data and content problems that hurt both human shoppers and AI agents. Accurate product specifications, visible policies and reliable inventory improve the current website while preparing it for agent-driven discovery.
How AI shopping agents choose which products to recommend
No universal ranking formula controls every AI shopping agent. Each platform uses its own models, data sources, commercial relationships and trust systems. Still, agents need a common set of inputs to make a defensible recommendation.
Clear product attributes
An agent needs more than a product name and a lifestyle photo. It needs dimensions, materials, compatibility, sizing, ingredients, model numbers, use cases and limitations. A shopper might ask for “a stroller that fits in a compact-car trunk,” even if no retailer uses that exact keyword.
Detailed attributes help the agent connect the request to the product.
Current price and availability
Stale inventory makes a recommendation useless. Agents need the current price, available variants, delivery estimates and regional restrictions. Google’s UCP catalog capability includes access to price, inventory and variant information for this reason.
Policies the agent can read
Shipping fees, return windows, warranties, subscriptions and cancellation rules can change the best choice. Hiding those details inside an image, accordion script or vague legal page makes comparison harder.
Evidence of quality and trust
Reviews, ratings, expert coverage, certifications and consistent business information help an agent assess risk. A product claim carries more weight when the merchant supports it with specifications and independent evidence.
A merchant the system can verify
Agents and payment networks need to know who sells the product and who will fulfill the order. Clear company details, secure checkout, support information and dependable order handling contribute to that trust.
The best-looking page does not guarantee selection. The cheapest product does not guarantee it either. The agent needs enough reliable information to explain why one option fits the buyer’s request.
What agentic commerce changes for marketing
Marketers have spent years optimizing for attention. Agentic commerce adds a second audience that does not respond to a clever hero image or an emotional headline in the same way.
Product data becomes part of marketing
Catalog work often sits with ecommerce or operations teams. Agentic discovery turns product attributes, feed accuracy and policy content into acquisition inputs. A missing specification can remove a product from consideration before the buyer sees it.
Marketing, ecommerce and technical teams need shared ownership of that information.
Brand still matters, but its role changes
A buyer can tell an agent to prefer a brand. Reviews and past experience can influence the shortlist. Strong brands also earn mentions, coverage and demand that AI systems can observe.
The agent may compress the shopping experience, though. It can summarize five brands into three recommendations and reduce the number of direct brand interactions before purchase. Businesses need recognizable demand and evidence the agent can retrieve.
SEO expands beyond the search results page
Search engines, retailer feeds, review sites, publisher coverage and product pages can all supply information to an AI agent. Traditional SEO remains useful because crawlable pages, clear information and authority support discovery across several systems.
AI visibility adds another question: can an AI system interpret the offer well enough to include it in an answer or transaction?
Adobe found that U.S. retail product pages had an average machine-readability score of only 66% in early 2026. Roughly one-third of product-page content remained difficult for AI systems to read. Angarum found a related gap in local discovery in its 2026 Local Search and AI Discovery Report.
Attribution gets harder
An agent might read a guide, compare a merchant feed, check third-party reviews and complete a purchase through another interface. The retailer may see the final referral or transaction without seeing every source that influenced the recommendation.
Teams will need to track agent referrals, feed eligibility, product inclusion, assisted conversions and the quality of AI-referred sessions. Last-click reporting will miss part of the journey.
What businesses should do now
Most companies can start with an audit rather than a platform rebuild.
1. Test real buying prompts
Ask ChatGPT, Gemini and other relevant tools to recommend products in your category. Use the details a customer would provide: budget, location, intended use, delivery date and constraints.
Record whether your products appear, which competitors receive recommendations and what evidence the answer cites. Repeat the test because responses can vary.
2. Clean the product catalog
Check titles, descriptions, categories, variants, identifiers, prices and inventory. Remove contradictions between the product feed and website. Add the attributes customers use when comparing options.
A catalog that causes errors in Google Merchant Center will not become easier for an AI agent to understand.
3. Make important content machine-readable
Keep core product details in crawlable text. Use descriptive headings and HTML tables for specifications. Add useful alt text to product images. Avoid placing essential information only inside images or scripts that prevent reliable rendering.
Adobe’s machine-readability findings show that this basic issue remains common among large retailers.
4. Add and validate structured data
Use appropriate schema for products, offers, availability, reviews, organization details and FAQs. Structured data does not guarantee an AI recommendation. It gives machines a cleaner representation of information that already appears on the page.
Keep the markup consistent with visible content. False prices, outdated stock or review markup that users cannot see creates risk.
5. Publish comparison-ready information
Answer the questions a buyer asks before choosing: Who is this for? Which use case does it suit? What are the dimensions? What does the warranty cover? Which accessories work with it? What will shipping cost?
Useful comparison content helps a shopper and gives an agent facts it can use.
6. Strengthen third-party trust signals
Keep business listings accurate. Earn reviews on platforms customers use. Pursue credible editorial coverage and expert mentions. Correct false or outdated descriptions of the company when possible.
An agent may learn about the offer from sources the business does not control.
7. Review commerce and payment readiness
Ask the ecommerce platform and payment provider which agentic commerce standards they support, where those features operate and who remains the seller of record. Review fraud, refund, customer-data and support implications before enabling a new checkout path.
Do not install a protocol for the label alone. Connect it to a channel where eligible shoppers and merchants can transact.
8. Protect the customer relationship
Decide what happens after an agent sends the order. The customer still needs confirmation, tracking, support and a workable return process. Loyalty benefits and account identity should carry across supported channels when possible.
The transaction may start outside the website. The merchant still owns the fulfillment experience.
Does agentic commerce matter to service businesses?
Product retailers face the clearest immediate impact because they already maintain catalogs, inventory and checkout systems. Service businesses should still watch the same change.
A consumer can ask an agent to compare insurance options, find a hotel, identify three HVAC contractors with strong reviews or book an appointment within a certain area and time window. The agent needs structured facts about service areas, availability, pricing approach, qualifications, reviews and booking rules.
Many service purchases involve a quote, inspection or regulated advice, so the agent may stop at a shortlist or appointment request. That still gives it influence over which businesses the customer considers.
Local companies can prepare by maintaining accurate business information, building clear service pages, publishing useful pricing context and explaining who each service suits. These steps support local SEO and AI discovery at the same time.
Risks businesses should plan for
Incorrect recommendations
An agent can misunderstand a product, use old pricing or omit a suitable option. Businesses need a way to monitor how their offers appear and correct the source information.
Less control over presentation
The agent may reduce a detailed brand story to a sentence and a few attributes. Consistent information across the website, feeds and trusted third parties lowers the chance of distortion.
Fraud and disputed intent
Merchants need proof that the buyer authorized the agent and the purchase. Payment tokens, agent identity and approval records address part of this problem, but businesses still need clear refund and support processes.
Privacy and personalization
An agent can make better recommendations when it knows a buyer’s history, preferences and location. Companies should limit data collection to what they need and explain how they use it. Permission should remain visible to the customer.
Dependence on large platforms
If discovery and checkout move into a few AI interfaces, those platforms gain influence over visibility, fees and customer access. Businesses should keep their website, customer list and direct relationships healthy while testing new channels.
How to measure agentic-commerce readiness
Standard traffic reports do not show the whole picture. Add a small set of measures that reflect machine discovery and agent-driven sales:
- AI referral traffic: Sessions and revenue from known AI platforms.
- AI-referred conversion rate: The share of those sessions that produce a lead or sale.
- Product inclusion rate: How often your eligible products appear in a defined set of shopping prompts.
- Recommendation quality: Whether the agent presents the correct price, attributes and use case.
- Feed health: The share of products with complete, current and approved data.
- Machine readability: Whether AI systems can access the information customers need.
- Assisted conversions: Purchases where AI influenced research but did not receive the final click.
- Returns and support issues: Problems connected to incorrect agent recommendations or expectations.
Start with a fixed prompt set and a monthly review. Agent answers change too often for one screenshot to serve as a benchmark.
Your next shopper may be an AI agent
The customer remains human. The software standing between the customer and the seller is changing.
An AI agent may read the catalog, compare the offer and decide whether a product deserves the buyer’s attention. It may also prepare the cart and complete the transaction after approval. Businesses that supply accurate, useful and accessible information give both the agent and the person a reason to choose them.
Start with the work that improves the current customer experience: clean up the data, answer comparison questions, make policies easy to find and test how AI systems describe the brand. Those fixes create value before agentic purchasing reaches every category.
Angarum Media helps businesses improve search visibility, AI discovery and the content systems behind both — and, when you’re ready to advertise inside the AI conversation itself, our ChatGPT Ads management. Explore Angarum Media or ask us to review how well AI systems can understand your offer.
Primary sources used
- Adobe: AI traffic grows but retail sites lag in AI search visibility
- Adobe: AI travel and retail traffic report, June 2026
- Google: New tools and Universal Commerce Protocol for retailers
- Google: Universal Commerce Protocol updates
- Stripe: Agentic Commerce Protocol documentation
- Visa: Intelligent Commerce
- Visa and OpenAI: Building the future of agentic commerce
Frequently Asked Questions
Is agentic commerce the same as ecommerce?
Ecommerce gives customers a digital store where they browse and buy. Agentic commerce lets an AI agent complete parts of that work on the customer’s behalf, including discovery, comparison, cart creation and, in supported cases, payment.
Is agentic commerce the same as conversational commerce?
Conversational commerce uses chat or messaging to help a customer shop. The customer still drives most of the process. An agentic system can act on an approved goal and complete tasks across the purchase journey.
Can AI agents buy products without permission?
Responsible agentic payment systems use buyer authorization and controls. The buyer can approve a purchase or set rules such as spending limits. Visa, Stripe and Mastercard are developing identity, tokenization and intent controls for this purpose.
Which companies support agentic commerce?
Google, Stripe, Visa, Mastercard, Shopify, Salesforce and several large retailers have announced standards, integrations or payment tools. Availability depends on the market, platform, merchant and use case.
Will AI shopping agents replace websites?
Websites will remain important for product information, direct relationships, support and transactions. AI agents may reduce the number of pages a shopper visits before choosing a product. Businesses should design for human customers and make the same information easy for machines to interpret.
How can a small business prepare for agentic commerce?
Start with accurate product or service data, crawlable pages, structured markup, clear policies and consistent third-party information. Test realistic prompts on AI platforms and track whether the business appears with correct information.
Does agentic commerce replace SEO?
No. Crawlability, useful content, authority and clear site structure still support discovery. Agentic commerce adds product feeds, live catalog data, machine-readable policies and transaction readiness to that work.
Let’s build what’s next.
Tell us where your brand is headed — we’ll reply within one business day with where to start.