Alexa for Shopping can only recommend what a product page tells it. When we started looking closely at Amazon listings, we kept finding the same gaps: missing attributes, thin bullets, details that contradicted each other. They were easy to spot and slow to fix by hand. So we built agents that find the gaps and write the fix, and we built the review around your team.
This post is part of All Eyes on Amazon, our two-week series on getting your listings ready for Alexa for Shopping before Black Friday. Today we're launching Enrich in ReFiBuy for Amazon. Here's how it works, and why we built it the way we did.
Our initial audit found errors on 22% of Amazon product pages. Some are obvious, like a waterproof boot with no waterproof attribute. Others are quieter: a width that appears in the title but not in the width field, an ingredient list that contradicts a "vegan" claim, bullets that describe the brand but not the product.
It's not hard to see how it happens. Many listings were built by hand, often under pressure to go live, and then left alone. Shoppers still read those pages, and now Alexa for Shopping reads them too. Alexa matches a shopper's question to the details on the listing, and it uses those details, like fit, materials and use case, to explain why it picked a product. If a detail isn't on the page, Alexa has nothing to match or explain.
Before suggesting anything, ReFiBuy for Amazon evaluates each product page across four areas: Be Available, Be Findable, Be Trusted and Be Chosen. Every product gets an Amazon Agentic Commerce Optimization (ACO) Score, with the specific gaps behind it and fixes ranked by estimated impact. That's what tells the agents where to start.
A product page carries two kinds of information, so Enrich has two agents.
The content agent works on the words a shopper reads and Alexa draws on: titles, highlights, bullets, descriptions and search terms. It rewrites them to say clearly what the product is for, who it fits and what it's made of, the same things Alexa uses to explain a pick. You see the current value and the suggested value side by side.
The attribute agent fills in the structured fields Alexa matches on, like width, closure type, material, age range and intended use. These are the fields most often left blank, and the ones that decide whether a product can match a specific question like "waterproof trail shoes in a wide fit."
We didn't want a single model writing straight into your listing. Each agent is paired with a reviewer agent that checks its work before you see it. Every suggestion that reaches you carries a Confidence Score, showing how sure the agent is that the change improves on what's there, plus citations to the source fields it drew from, so your team can see where each detail came from.
Good listing copy for one brand is wrong for another. In AI Configuration, you give the agents your brand context, then set guidelines for each field, a glossary of the terms you use and exclusions for the claims you never make. The agents also remember what your team accepts and rejects, so suggestions get closer to your standards over time.
Every suggestion lands in the Enrichment Queue. You see the existing value next to the recommended one, for the parent product and each variant, and you can accept, edit, reject or revert. Work one suggestion at a time, or apply a whole field at once.
Reviewing thousands of listings by hand before Black Friday isn't realistic, so review is built in and configurable. With Enrichment Queue Automation, you set two bars for each field: suggestions above the high bar are accepted, suggestions below the low bar are rejected, and everything in between goes to a person. By default, titles auto-accept at 95% confidence and attributes at 80%. Teams that need more sign-off can add multi-step approval workflows.
Once your team approves its changes, you choose how they go live: export a feed of enriched data into the Seller Central, Vendor Central or product information management (PIM) workflow you already use. What it looks like on one product
In our demo workspace, one trail running shoe started with an evaluation score of 78 out of 100. The content agent suggested changes to 10 fields and the attribute agent suggested 9 attributes, so 19 suggestions were waiting for review, with confidence scores mostly in the mid-80s to mid-90s. Many were details the listing simply didn't state, like the width of each size variant and the activity the shoe is built for.
Enrich is available now in ReFiBuy for Amazon, built for Alexa for Shopping. The easiest place to start is one product. Get your free Amazon ACO Report before Black Friday, and we'll show you what Alexa can't read yet and what to fix first.