When AI compares, what sets your store apart?
By Daniel Pisano, Partner & CGO at Vurdere
When an AI agent compares stores with similar prices, delivery times and catalogs, the tie-breaker is the reputation it can read and verify. To get recommended by AI shopping agents, a store needs reviews in volume, recent, honestly distributed and from verified buyers, published as structured data.
How does an AI agent build its shortlist?
Take a simple request: "beginner running shoes, under $100, delivered by Friday." The agent starts by filtering. It drops anything over budget, out of stock or unable to arrive in time. That part runs on objective catalog, price and logistics data.
The problem starts after the filter. Ten or twenty very similar options remain. To the agent, two $95 pairs arriving on Thursday are practically identical on everything objective. That is where it needs another criterion to decide what to show first.
A Salesforce executive has explained that language models prioritise structured product feeds over what they find by scraping pages. In other words, the tie-breaker has to exist in a format the machine can read.
Why aren't price and delivery enough to stand out?
Because the agent is very good at levelling whatever can be compared. If your price is $5 higher, it sees it. If a competitor delivers a day earlier, it sees it. The advantages a person sometimes forgives, like a nicer-looking page, don't exist for the machine.
In the catalog specification of the Universal Commerce Protocol, Google's shopping protocol, the product carries an aggregate rating field with its value, its scale and the number of reviews. It is one of the few quality signals that travels alongside price and stock. That is why reputation stops being a page detail and becomes a store's main argument against an almost identical rival.
What makes a review trustworthy to a machine?
A 4.8 score says little on its own. For the agent, four characteristics carry weight:
Signal | Why it matters | What to do |
Volume | A rating from 300 reviews is more stable than one from 6. | Ask for a review after every purchase, on every channel. |
Recency | Reviews from two years ago don't describe today's product or service. | Keep collection continuous, not only during campaigns. |
Distribution | An average of 4 can hide half fives and half ones. | Publish every review, including negative ones, and reply to them. |
Verification | A review from a confirmed buyer is worth more than one from any visitor. | Flag and show which reviews come from verified buyers. |
The last two are already under discussion in the protocol itself. In May 2026, a proposal in the UCP repository asked for the rating to include the star distribution and the number of reviews from verified buyers. Its reasoning matches this article: an average and a count alone can't show whether there is consensus or polarisation. The proposal has not been approved yet.
Are fake reviews now a legal risk too?
Yes. In the United States, the FTC's Rule on the Use of Consumer Reviews and Testimonials took effect on 21 October 2024. It bans fake reviews, including AI-generated ones, and the suppression of negative reviews, with civil penalties of up to $51,744 per violation. In the European Union, the Omnibus Directive (EU 2019/2161), applied since 28 May 2022, prohibits presenting reviews as genuine without taking steps to verify they come from real consumers.
The direction is the same on both sides of the Atlantic: the review that counts is the one that can be verified. AI agents tend to follow the same rule.
How does Vurdere help a store stand out?
At Vurdere we built the platform around one idea: a good review comes from real behaviour. Product reviews, Q&A ranked by relevance and seller ratings on marketplaces start from real purchases and interactions, and the data is structured and analysed with AI.
Volume also depends on the collection channel. In Vurdere research with more than 120,000 consumers, review requests sent by WhatsApp converted at up to 15%, against 5% for email. In A/B tests with current clients, Vurdere recorded up to 10x more reviews collected and up to 3x higher conversion rate.
Shiseido shows why this matters. In beauty, shoppers want to know how a product worked for someone with similar skin and a similar routine. The agent reasons the same way: it needs specific, trustworthy reviews to recommend the right product to that person.
Checklist: getting your store recommended by AI shopping agents
Publish the aggregate rating and review count as structured data, not only in the widget.
Ask for a review after every purchase, on the channel your customers actually answer.
Don't filter out negative reviews. Reply to them.
Identify reviews from verified buyers.
Answer customer questions on the product page.
Track recency: how many reviews each product received in the last 90 days.
My view
For years, the industry treated reviews as social proof to win over the undecided visitor. That still holds, but there is now a second reader, and it is far more demanding. An agent isn't impressed by a high average. It wants to know how many people reviewed, when, how the scores are spread and whether those people actually bought.
That is why I don't see fake reviews as a risky shortcut. I see them as a shortcut that no longer works. A store inflating its score today is building a reputation the machine will discount and regulators in the US and Europe already penalise. A store collecting real reviews every day is building the one advantage an agent can't level by comparing prices.
Frequently asked questions
How do I get my products recommended by AI shopping agents?
Keep a complete catalog, up-to-date price and stock, and reviews in structured data. Among similar stores, agents tend to favour the one with plenty of recent, verifiable reviews.
Do negative reviews hurt AI recommendations?
A realistic distribution, with some low scores the store has replied to, signals authenticity. Suppressing negative reviews is prohibited under the FTC rule in the US and the Omnibus Directive in the EU.
What is a verified review?
A review from someone who demonstrably bought the product. The exact definition of verification is left to each store or platform.
What does the FTC fake reviews rule prohibit?
Among other things, it bans fake or AI-generated reviews, buying reviews the brand knew or should have known were misleading, undisclosed insider reviews and suppressing negative reviews.
Comments