Product Q&A as data for AI shopping agents: the format that answers before the customer asks
By Daniel Pisano, Partner & CGO at Vurdere
Customer questions and answers on a product page are the closest thing to what shoppers now type into an AI assistant. Agentic commerce protocols currently carry only the rating and review count. Everything else, from dimensions to compatibility and real-world use, the agent has to find on the page, in plain and checkable text.
Why have customer questions become data for AI agents?
People talking to a shopping assistant don't type keywords. They ask questions. "Will this air fryer fit under a kitchen cabinet?", "Does this sunscreen leave oily skin shiny?", "Are these trainers OK for wide feet?". These are the same questions customers have been leaving in the Q&A section of product pages for years.
The scale is already visible. In its Q3 2025 results, Amazon said 250 million customers had used Rufus, its shopping assistant, that year, and that shoppers who used it during a trip were 60% more likely to complete a purchase. In Europe, an Ecommpay survey of 1,750 consumers in Spain, Italy, France, Germany and the UK, published in July 2026, found that people are far more willing to use agents to compare products and find deals than to let them complete a purchase alone. Comparison means asking questions, and agents need someone to have answered them already.
What do agentic commerce protocols read today?
In the guide to agentic commerce protocols, I showed that the aggregate product rating is already a protocol field. It's worth looking at what else goes in and what stays out, because that is where Q&A comes in.
Specification | What it carries about the product | Questions and answers |
UCP (Google), draft catalogue spec | Title, description, price, variants, media and a rating field with value, scale and review count. | No dedicated field. |
OpenAI product feed (ChatGPT) | Offer data, seller name, review_count and star_rating, plus store ratings with specific setup. | The documentation says Q&A lists and individual reviews are not part of the discovery contract. |
In practice, the protocol carries the number, not the content. A 4.7 rating from 812 reviews travels in the feed. The answer explaining that the appliance is 32 cm tall and fits under a standard wall cabinet does not.
If the feed doesn't carry questions, where does the agent find them?
Amazon is the clearest example. The company says Rufus, renamed Alexa for Shopping in May 2026, was trained on its product catalogue, customer reviews, community Q&As and information from across the web. When the platform builds the agent itself, Q&A is a source.
Outside the big marketplaces, the logic is similar but depends on access. What isn't in the feed, an agent can only find if it can read the product page or other sources on the web. A Q&A section that loads only after a click, or only through JavaScript, risks not being read at all.
Which questions are worth answering on the product page?
Not every question helps. The useful ones settle a concrete doubt that the spec sheet doesn't cover. They usually fall into five groups:
Compatibility: does it fit my model, does it work with a UK plug, will it fit the bracket I already have.
Size and space: real height, weight, whether it fits a specific spot.
Use: which skin type, hair type, terrain or routine it suits.
Comparison: how it differs from the previous version or the model next to it.
After-sales: warranty, installation, returns, spare parts.
How you answer matters as much:
Weak answer | Answer that helps the customer and the agent |
"Hi! Yes, it fits. Let us know if you have any other questions." | "Yes. It is 32 cm tall and 28 cm wide. With the lid open it needs 45 cm of clearance above." |
"Please see the product description." | "It's designed for oily skin. It has a matte finish and SPF 50, as stated on the pack." |
A useful answer is short, specific and consistent with the spec sheet. The figures in the table are illustrative, but the rule applies to any category: the answer should make sense on its own, without the question next to it.
How should you organise Q&A for people and agents?
Answer fast and with facts. A question left unanswered for a week helps nobody and makes the store look careless.
Keep Q&A in the page source, loaded with the rest of the content.
Rank by relevance. The question ten people asked should appear before the one asked once.
Reuse what repeats. If the same doubt arrives every week, the answer belongs in the product description, which is a required field in UCP.
Let customers answer too. A buyer's experience complements the brand's official answer.
Review answers when the product changes. An outdated answer is worse than none, because an agent may repeat it.
How does Vurdere handle questions and answers?
At Vurdere, Q&A is one of the platform's modules, alongside product reviews, interactions and community, and Social SEO. Questions are ranked by relevance, so the most common doubt shows first, and customer content is structured so that both people and search engines can read it.
We also use AI to suggest replies to support teams. In an internal study published on our blog, intelligent reply suggestions were associated with 20% higher conversion and up to two-thirds lower support operating costs. In A/B tests with current clients, the platform recorded up to 3x conversion rate and +20% organic traffic. Installation takes about 3 hours, with one week of validation, on Shopify, VTEX, Magento and other platforms.
My view
I think Q&A is the most underrated content in ecommerce. Plenty of stores treat the questions section as a support inbox tucked at the bottom of the page, answered when someone has time. I see it the other way round: it is the only part of the page written in the customer's own words, with their real doubts, already in the question format they use with an AI assistant.
I also don't think it's wise to wait for the protocols to add a field for it. Today, the ChatGPT feed spec states plainly that Q&A lists are outside discovery. That may change, or it may not. Meanwhile, stores with good, fast and visible answers are writing the material that assistants like Amazon's already use. Stores that leave questions unanswered leave a gap, and the agent will fill it with a competitor's page.
Frequently asked questions
Do AI shopping agents read product page Q&A?
It depends on the agent. Amazon's assistant was trained on community Q&As among other sources. Open protocols such as UCP and the ChatGPT product feed have no Q&A field, so the information needs to be accessible on the page.
What's the difference between an FAQ and customer Q&A?
An FAQ is written by the brand, with the questions it expects. Customer Q&A comes from shoppers, with real doubts about a specific product. The two work together.
Is it worth answering old questions?
Yes, especially recurring ones. A clear answer is still useful to whoever lands on the page today, person or agent.
Who should answer: the brand or other customers?
Both. The brand guarantees the technical facts. Other customers explain how the product behaves in real use, which a spec sheet can't show.
Comments