AI-generated article. This article was researched and drafted using AI tools and published automatically, and its featured image was generated by AI. Facts are drawn from the sources cited in the text.
Picture a shopper asking an assistant for waterproof hiking boots under a hundred dollars, reading three suggestions, and clicking through to a store they had never heard of. They buy. The store owner sees a referral from a chat product in their analytics and has no idea how that recommendation happened, or how to make it happen again. That gap, between being recommended and understanding why, is the practical shape of agentic commerce for small business right now.
It is worth being precise about what changed, because the version of this story told a year ago turned out to be wrong.
The checkout everyone predicted did not arrive
Through 2025 the assumption was that people would buy inside the chat window. That has not happened. OpenAI deprecated Instant Checkout in March 2026, and analysis from Digital Applied describes the industry quietly regrouping around a different model: discover in the assistant, buy on the merchant’s own site. Google’s AI Mode did add agentic checkout, with Wayfair, Chewy and Etsy among the early participants. But the dominant pattern is still recommendation followed by redirect.
The fight over who gets to do the recommending is much louder. Amazon won a temporary injunction in March 2026 blocking Perplexity’s Comet agent from its site, and the Ninth Circuit reversed it, finding that the balance of equities and the public interest favored Perplexity. Modern Retail frames Amazon’s motive plainly: it wants to control the shopping experience, including which ads a shopper sees.
For a business that is not Amazon, that fight is mostly good news. The traffic has to land somewhere. It is worth knowing what is actually doing the recommending, though, because the word agent covers a lot of ground right now, from genuinely autonomous systems to search with a friendlier interface.
What agentic commerce for small business actually looks like now
The numbers are the interesting part. Adobe found AI-referred traffic to retail sites doubled year over year, and first-quarter 2026 figures put growth at 393 percent. Volume on its own would not mean much. Quality does.
In March 2025, traffic arriving from AI assistants converted 38 percent worse than other channels. In March 2026 it converted 42 percent better, and by May 2026 Adobe’s figure had moved to 54 percent better. Revenue per visit from AI referrals ran 37 percent above non-AI traffic. Shoppers arriving this way spent 48 percent more time on the page and browsed 13 percent more pages per visit.
The explanation is not mysterious. Someone who has spent ten minutes comparing options conversationally arrives already decided. The browsing happened somewhere else. What lands on the site is the final step of a journey the merchant never saw.
The part that should give anyone pause
All of those percentages come off a small base. AI referrals are still a modest share of most stores’ total traffic, and a 393 percent increase on a small number is still a small number. Anyone moving a real budget on those figures alone is moving too fast.
There is a harder problem underneath. Nobody buys placement here. There is no bid, no ad slot, no way to pay for a recommendation the way a business pays for search results. What gets recommended depends on what an assistant can read and verify about a product, and that is a genuinely different discipline from advertising.
The standards have not settled either. The Universal Commerce Protocol, published by Google and Shopify with Etsy, Wayfair, Target and Walmart in January 2026 under an open license, defines how agents discover products, negotiate capabilities, check out and hand off after purchase. OpenAI’s Agentic Commerce Protocol takes a different route, bundling checkout and payment through a shared payment token. Two standards, both live, neither obviously winning. For a small team that turns this into a build or buy question with an unusually expensive wrong answer, and it is worth remembering how many agent projects never make it out of the pilot stage even when the technology works.
A smaller place to start
Rebuilding a store around a protocol is not the first move. The first move is duller and much cheaper: make the business legible to a machine.
That means product and service information a system can read without a human interpreting it. Prices in text, not baked into an image. Specifications that are complete and honest. A returns policy, a shipping timeline, a service area, written plainly on the page rather than implied. Google opened simplified onboarding through Merchant Center in March 2026 specifically to bring in retailers of every size, which is a reasonable signal of where the effort is expected to come from.
Service businesses are not exempt. An assistant recommending a bookkeeper or a physiotherapist is doing the same work with different inputs.
None of this requires understanding agents deeply, and anyone who audits their own product pages this month is already further along than most. It also overlaps almost entirely with plain good web hygiene, which means the work is not wasted even if agentic commerce stalls again.
What is less clear is what happens to the relationship. When the comparison, the questions and the decision all happen inside somebody else’s assistant, and the merchant sees only the final click, who does the customer actually belong to?
