Marketplaces and agentic commerce: seller reputation becomes a buying criterion
By Daniel Pisano, Partner & CGO at Vurdere
On a marketplace, the same product is often sold by several sellers at similar prices and delivery times. When an AI agent makes the choice, the product rating is identical for all of them. What separates one offer from another is the seller's reputation, and agentic commerce protocols already have dedicated fields for it.
Why is the marketplace where agents most need to choose between sellers?
Picture a specific blender model, same barcode, listed by six shops on the same marketplace. Same photo, same spec sheet and the same product reviews, because those describe the blender, not the shop selling it. A person usually picks the lowest price with acceptable delivery. An AI agent buying on that person's behalf is left with a question the product page doesn't answer: which of these six shops can be trusted?
Shoppers expect to hand over more of these decisions. Global Payments' annual Agentic Commerce Report, based on 8,027 consumers in seven markets including the UK and France, found that consumers now expect AI agents to make 15% of their purchases within five years, up from 9% a year earlier. The same release warns that retailers will have to compete for visibility inside AI-driven shopping, and names trust as one of the likely differentiators.
Trust is also where European shoppers hesitate. In an Ecommpay survey of 1,750 consumers in five European countries, published in July 2026, only 33% of UK respondents said they would trust an agent with their card details, and 31% would not let AI spend any money at all.
What do the protocols already say about the seller?
This is the point that made me write this piece. Agentic commerce protocols treat the seller as data separate from the product.
Specification | Seller fields | What it means |
OpenAI product feed (ChatGPT) | seller_name (required), seller_url, marketplace_seller, store_review_count and store_star_rating (the last two need specific setup) | The store rating is its own field, on a 0 to 5 scale, and must not be mixed with product reviews. |
UCP (Google), draft catalogue spec | Optional seller field on each variant, described as seller context | The protocol already reserves space for who sells each variant. Its sub-fields aren't defined on that page of the spec yet. |
OpenAI's rule is explicit: the store review count must come from the same population as the store rating, and it is not combined with product reviews. The protocol separates two questions that retailers tend to blur. One is whether the product is good. The other is whether the seller delivers what it promises.
Why isn't the product rating enough on a marketplace?
Because it is shared. If six sellers offer the same blender, all six show the same 2,000 product reviews. That rating helps the agent decide whether to buy that model. It doesn't help decide who to buy it from.
Seller reputation answers the second question, and it has one important property: it can't be inherited. Product reviews can come from the manufacturer, other channels or years of sales. A seller's reputation is built order by order, by that shop, with those customers.
What makes a seller reputation that a machine can use?
Today the protocols ask for two numbers: the store rating and the number of reviews behind it. For those numbers to mean something, I would look at these points:
Point | Why it matters |
Volume of store reviews | A 4.9 from 15 reviews is fragile next to a 4.6 from 3,000. |
Reviews tied to real orders | A store rating only makes sense if the reviewer bought from that store. |
Recency | A shop that served customers well two years ago may run differently now. |
Comments on delivery and service | This is what separates sellers of the same product: on-time delivery, packaging, returns. |
Replies to negative reviews | They show how the seller fixes problems, which is what worries people most when they delegate a purchase. |
Does seller identity matter under EU rules?
Yes. Under the EU Digital Services Act (Regulation 2022/2065), marketplaces that let consumers conclude distance contracts must collect information on each trader before it can sell: contact details, an ID document, payment account, trade register number and a self-certification of compliance with EU law. They must make best efforts to check it and suspend traders who don't correct inaccurate information. Consumers must be shown the trader's contact details, registration and that commitment.
Traceability isn't a rating, but it points the same way as the protocols: knowing who sells is becoming part of the buying decision.
What can sellers and marketplaces do now?
Separate product reviews from store reviews. Ask for both after every order, with different questions.
Ask for reviews on the channel your customers answer. In Vurdere's survey of more than 120,000 consumers, review requests sent by WhatsApp converted up to 15%, against 5% for email.
Reply to negative store reviews with what was done, not with a template.
If you run a marketplace, make each seller's rating available in structured form, so it fits the fields agents already read.
Track your store rating the way you track price and delivery time. It will weigh in the same comparison.
How does Vurdere help marketplaces and sellers?
Vurdere has modules built for this problem. Alongside product reviews, the platform includes seller reviews, which treat the reputation of whoever sells as content in its own right, and a Seller Portal. The Seller Showcase lets shoppers check a seller's rating and other buyers' experiences before they commit.
In A/B tests with current clients, Vurdere recorded up to 3x conversion rate, +15% incremental sales and up to 10x more reviews collected. Installation takes about 3 hours, with one week of validation.
My view
I think marketplace sellers will feel agentic commerce before brands with their own stores do. On its own site, a brand controls the whole page. On a marketplace, a seller shares one listing with competitors selling exactly the same item, and the only thing that is truly theirs is their reputation.
That's why I disagree with treating the store rating as an internal marketplace metric, mainly useful for penalising late deliveries. It is turning into decision data, published in its own protocol field. Sellers who fight over pennies on price should fight just as hard for real, recent store reviews. When an agent looks at six identical offers, that is what it will find to break the tie.
Frequently asked questions
What's the difference between product reviews and seller reputation?
Product reviews say whether the item is good and are the same for everyone selling it. Seller reputation says whether that shop delivers, responds and fixes problems, and belongs to that shop alone.
Do AI shopping agents consider store ratings?
OpenAI's product feed has fields for store rating and store review count, separate from product data, and UCP includes a seller field per variant. How each agent weighs that data is not public.
How can a small seller compete with a large one?
With verifiable reputation. A seller with fewer orders but recent, genuine and well-answered reviews has an argument that doesn't depend on size.
What does EU law require of marketplaces regarding sellers?
The Digital Services Act requires marketplaces to collect and check information on each trader and to show consumers the trader's contact details, registration and commitment to comply with EU law.
Comments