AI Agents Are Doing the Shopping Now: How to Get Your Business Ready for Agentic Commerce

Not long ago, AI in online retail meant a chatbot in the corner of the screen answering questions about delivery times. That stage is quietly ending. The newest AI agents browse catalogues, compare offers across dozens of shops, fill the basket and increasingly pay. The industry calls this agentic commerce, and sometimes the customer never even opens the merchant's website.

The numbers behind this shift are getting hard to ignore. A SEON survey of 2,026 US adults, released this week, found that 68% already use an AI agent at least once a week. At the same time, the payment industry is racing to give those agents real spending power, with certified agent cards and new checkout protocols coming from OpenAI, Google, Stripe, Visa and Mastercard.

So this time we are looking at agentic commerce: what already works, where consumers still draw the line, and what business owners should change on their websites before AI agents become a normal sales channel. Let's sink in!

People Are Already Shopping With AI Agents

The habit arrived faster than most merchants expected. In the SEON study mentioned above, 68% of American adults said they use an AI agent at least weekly, and a large share of that activity is commercial: researching products, hunting for discounts, comparing specifications. Ravelin's global study from late 2025 put the figure even higher, with 73% of shoppers using AI somewhere in their shopping journey.

Riskified's quarterly Agentic Commerce Pulse adds a useful detail. In its Q1 2026 edition, 61.5% of consumers said they had used AI tools for product discovery and recommendations. Discovery is the part of the funnel that used to belong to your category pages, your blog and your Google rankings. It is moving into chat windows.

Even more telling, 31.2% of respondents told Riskified they prefer doing this inside general AI platforms such as ChatGPT or Gemini, slightly ahead of the 27% who prefer retailer websites and apps. For the first time, the place where buying decisions are made is not a website you control, and not even a search results page you can optimise in the classic way.

The Payment Rails Are Being Built Right Now

Discovery is one thing; payment is another. Over the past year the industry has produced a whole alphabet of standards for letting agents transact safely. OpenAI and Stripe introduced the Agentic Commerce Protocol, Google pushed its Universal Commerce Protocol along with the AP2 payments standard, and Visa, Mastercard and AWS have been piloting agent-driven checkout since early 2026.

This week brought a small but symbolic milestone: Sunrate launched a commercial virtual card designed specifically for AI agents, certified under Mastercard's Agent Pay programme. The card lets an agent pay within preset rules, such as spending limits, approved merchant categories and time windows, so a company can let software buy things without handing it the corporate card.

None of this is mass-market yet. But the direction is clear enough: the card networks are not waiting to see whether agentic commerce happens. They are building the rails now, and they expect merchants to plug in.

Trust Is the Bottleneck, Not Technology

Consumers, meanwhile, are enthusiastic about delegation and nervous about money. In the SEON survey, half of respondents said they are not comfortable letting an agent use their credit card. Riskified found much the same mood: 55% are uncomfortable with AI agents making purchases for them, and 46.5% do not trust any company at all to manage purchases on their behalf.

People also have firm views on who should pay when something goes wrong. Asked about fraud liability, 65% of SEON's respondents with an opinion pointed at the AI company, and only 11% at the merchant. Shoppers may blame the platform, but in practice the chargeback, the refund and the support ticket still land on the merchant's desk.

Expectations around safeguards are concrete too: almost three quarters of Riskified's respondents want strong authentication, such as biometrics or one-time passwords, before an agent completes a purchase. Trust, not technical capability, is what will decide how quickly this channel grows.

What Agentic Commerce Changes on Your Website

An AI agent does not care about your hero banner, your pop-ups or your carefully designed brand storytelling. It reads structured data. Product names, prices, availability, delivery terms, return policies: if this information is buried in images, PDFs or scripts that only render on click, an agent will either get it wrong or skip your shop entirely.

That is why advice around agentic commerce keeps circling back to the same basics. Clean schema markup, accurate product feeds, machine-readable policies and a checkout that works without human-only puzzles. The industry has even picked up a new acronym for it, answer engine optimisation, which in practice overlaps heavily with good old technical SEO.

There is also an awkward middle ground to manage. Many bot-protection tools cannot yet tell a scraper from a paying agent, and blocking everything non-human may soon mean blocking actual customers. Merchants will need a policy for agent traffic the same way they once needed one for mobile.

Practical Steps for Businesses Building Digital Products

If you run an online store or are building one, start with an audit of how your product data looks to a machine. Check your structured data in Google's testing tools, make sure prices and stock levels are accurate in your feeds, and write your delivery and returns pages in plain, parseable language. This costs little and pays off in regular search as well.

Next, ask your payment provider what their agentic roadmap looks like. Stripe, PayPal and the card networks are all shipping agent-related features, and if your checkout is built on one of them, some of this capability will arrive almost for free. For custom platforms, it is worth planning early where agent authentication and spending rules would fit.

Finally, decide your own policy before the traffic forces the question. Will you accept agent-initiated orders? Above what order value do you require a human confirmation step? How will you log and label agent transactions in analytics? Teams we work with find it far easier to answer these questions calmly now than in the middle of a dispute later.

Where This Is Heading

Analysts expect agent-led buying to concentrate first in retail, travel and finance, where purchases are frequent and easy to compare. Further out, industry forecasts talk about algorithmic price negotiation and instant settlement becoming normal around 2030, with Know Your Agent identity checks playing the role that KYC plays in banking today.

If you sell through a major platform such as Shopify or Wix, part of this work will be done for you, since the platforms themselves are negotiating how their catalogues appear inside AI assistants. Custom-built shops have more freedom but also more homework, because nobody will expose their product data for them. That trade-off is worth weighing in any replatforming discussion over the coming year.

Regulation is coming along as well. The EU AI Act and updated payment rules will shape what agents may do without a human in the loop, especially for larger amounts. For most small and mid-sized businesses, though, the sensible move is not a big platform rebuild. It is making sure the data and checkout you already have do not lock agents out.

Final Notes

Agentic commerce is at that familiar early stage where the statistics look dramatic and the actual revenue is still small. Both things are true at once. Most of your customers already talk to AI agents every week, and most of them still will not let those agents touch their card. For now.

The businesses that win this transition will not be the ones with the flashiest AI announcement. They will be the ones whose product data is clean, whose checkout is flexible and whose policies were written before the first agent-driven order arrived. If you want help getting your website or e-commerce platform ready for that moment, the Davydov Consulting team is always happy to take a look.