When a shopper asks an AI assistant for waterproof hiking boots for wide feet under $250, the assistant does not look at your homepage banner or your lifestyle photography. It reads your product data. If the answer is not in the data, your product is not in the answer.
That makes product data quality the most practical thing an ecommerce brand can work on for AI shopping agents right now. It is also the least glamorous, which is why most brands have not done it.
Why product data decides whether AI agents recommend you
AI shopping agents compare products by reading structured information such as attributes, variants, prices, stock and policies. Copy written to persuade a human often buries the facts an agent needs. A description that says a jacket is perfect for any adventure tells an agent nothing. A field that says waterproof rating, fill weight and fit tells it a lot.
The platforms are moving quickly. In January 2026 Shopify introduced the Universal Commerce Protocol and Agentic Storefronts, which let merchants manage how their products appear across AI channels such as ChatGPT, Microsoft Copilot and Google's AI Mode from the Shopify admin. In September 2026, the company behind BigCommerce and Feedonomics launched catalog enrichment tools built around the same idea, that an agent can only answer questions about your products as well as your data allows.
The channels are getting easier to reach. The data you send them is still on you.
The product data checklist for AI shopping agents
Work through this list for your highest revenue categories first.
- Attributes live in fields, not prose. Material, dimensions, fit, compatibility, care and use cases should be structured fields an agent can read, not sentences hidden in a description.
- Values are consistent. Pick one unit, one spelling and one format for each attribute. Navy, Navy Blue and NVY are three different colours to a machine.
- Variants are modelled properly. Each size and colour should be its own variant with its own SKU, price and availability, not a note in the description.
- Products carry identifiers. GTINs or UPCs, manufacturer part numbers and brand names help agents match your product to what the shopper asked for.
- Price and stock are current. An agent that recommends something out of stock, or at last week's price, sends an annoyed customer to your checkout.
- Policies are machine-readable. Shipping times, costs, return windows and warranty terms should exist as clear, structured information, not only in a footer page.
- Product pages answer real questions. Use your support tickets and on-site search terms to find what people ask, then answer it on the product itself.
- Images are described. Descriptive alt text helps accessibility, search engines and agents that cannot see the photo.
- Structured data matches the feed. The schema markup on your product pages should agree with your feeds. Conflicting prices or availability erode trust with every system reading them.
- There is one source of truth. Product information should flow from one place, your platform or a product information system, to every channel, so a fix happens once.
How to audit your catalog in a week
You do not need a six-month project to find out where you stand.
- Export your catalog and pick the five categories that drive the most revenue.
- For each category, list the attributes a shopper uses to choose, such as size, fit, material or compatibility.
- Measure how many products actually have each attribute filled in, and how consistent the values are.
- Compare that list with the questions in your support tickets and the terms in your on-site search logs. The gaps show you what is missing.
- Fix the template and the process, not just the products. If new products keep arriving incomplete, the catalog will decay again within a quarter.
Who owns your product data?
Most catalog problems are not technical. They are ownership problems. Someone has to decide what a field means, which system is right when two disagree and when a product is ready to sell. In most retail and ecommerce teams, nobody has that job.
Rick Watson made this point in a recent episode of the Watson Weekly with Abhi Sachdeva, co-founder and CTO of EKYAM. Abhi estimates that only about five in every hundred retailers he talks to have a real product data owner, someone who sits between the business and the technical team and decides what a customer or a margin actually means. A person in a meeting knows which number to trust. An agent does not. It takes the first answer it finds and repeats it with confidence.
They also make the case that data readiness is never finished, and we agree. A catalog you clean up this quarter starts drifting the moment new products, suppliers and channels arrive. Treat product data as an ongoing role with a named owner, not a one-time project.
Can AI fix your product data?
Partly. AI is very good at drafting descriptions, extracting attributes from supplier documents, standardizing values and flagging products with thin content. Used well, it can do in days what would take a team months.
It is also capable of inventing a fabric blend or a weight limit that sounds plausible and is wrong. Wrong attributes create returns, complaints and in some categories real liability. Treat AI as the first draft and keep a person accountable for facts, especially specifications, safety information and anything regulated. An agent skill that encodes your product content rules, as we describe in what agent skills are, helps keep that work consistent.
This work pays off even if agentic commerce takes longer
Better product data improves on-site search and filtering, product listing ads, marketplace feeds and organic search. A search platform such as Algolia can only rank and filter on attributes that exist. The same fields that help an AI agent help a shopper on your own site today.
That is what makes this a safe investment. You are not betting on one AI channel. You are fixing the foundation every channel reads from. If your platform makes structured data hard, that is often a sign of the architecture issues we covered in why AI agents need a composable architecture.
Common questions
Do we need a PIM to be ready for AI shopping agents?
Not always. Many brands on Shopify Plus or BigCommerce can get far with well-structured metafields and disciplined processes. A product information system starts to pay off when you have many channels, many suppliers or many people editing product data.
Who should own product data in an ecommerce team?
One named person per area, usually in merchandising or ecommerce operations, with the authority to set definitions and the support of the technical team to enforce them. The title matters less than the role existing and having time for it.
Is schema markup enough?
No. Structured data on your product pages helps, but agents also read feeds and platform catalogs. All of them need to be complete and agree with each other.
Which products should we fix first?
Start with the categories that drive the most revenue and the most questions. Those are where missing data costs you the most, both in AI answers and on your own site.
How often should product feeds update?
As close to real time as your platform allows for price and stock. Descriptions and attributes can update less often, but should change whenever the product does.
Will AI shopping agents replace our on-site search?
Not soon. Many shoppers will still arrive at your site, and often an agent sends them there to buy. Good product data improves both, which is why it is the first thing to fix.
Where to start
Run the one-week audit on your top category and see what is missing. If you want help prioritizing the gaps, or deciding whether your platform is holding your data back, get in touch. For the bigger picture on the standards behind AI shopping, read our guide to UCP, ACP and MCP.
