Why Agents Need Structured Product Data

AI agents are increasingly being asked to help with purchase decisions. “Find me the best deal on noise-cancelling headphones.” “Is this laptop cheaper anywhere else?” “What’s in stock right now under £300?” These are exactly the kind of tasks agents should be good at - they involve comparing structured information across multiple sources and presenting a clear answer.

The problem is that the information isn’t structured. Not in any form an agent can reliably use.

What agents actually need

When an agent is asked to find the best price on a specific product, it needs a few things to do the job well. It needs current prices from multiple retailers. It needs to know whether the product is actually in stock. It needs to be confident that it’s comparing the same product across different listings, not two similar-sounding variants. And it needs all of this in a consistent, machine-readable format that it can process without guessing.

None of this exists today in a form that agents can access cleanly.

Retail product data lives on individual merchant websites, formatted for human eyes. Product names vary between retailers. Prices are embedded in page markup, sometimes only rendered by JavaScript. Stock status is often ambiguous or absent. There’s no shared identifier system that reliably links the same physical product across different stores - or rather, there is one (the GTIN barcode number), but it’s inconsistently applied and rarely exposed in a queryable way.

The result is that an agent asked to compare prices has to either scrape multiple websites - which is slow, fragile, and frequently against the retailer’s terms of service - or fall back on a vague response and suggest the user do the comparison themselves. Neither outcome is what the user asked for.

Why this matters now

Two things have changed that make this problem worth solving.

First, agents are getting better at following multi-step processes. A modern AI agent can take a user’s request, break it into sub-tasks, call external tools, process the results, and present a synthesised answer. The infrastructure for agent-mediated commerce is maturing rapidly. What’s missing is the data layer - a reliable source of structured product information that agents can query without having to solve the web scraping problem themselves.

Second, people are comparison shopping more aggressively than at any point in recent memory. Cost of living pressures mean that buyers - particularly in categories like electronics, where average order values are high - are increasingly unwilling to purchase from the first retailer they find. They want to know they’re getting the best available price. This is tedious, repetitive work that agents are perfectly suited to automate, if they have the right data.

What “structured” actually means

It’s worth being specific about what structured product data looks like in this context. It’s not just a list of products with prices. It means:

Consistent schema - every product follows the same data format, regardless of which retailer it comes from. An agent that can parse one response can parse them all.

Reliable identity - when the same product appears from multiple merchants, there’s a clear way to group those listings together for comparison. Ideally through a shared identifier like a GTIN, not through fuzzy name matching.

Explicit freshness - every data point carries a timestamp showing when it was last verified. An agent can assess whether a price is current or potentially stale, and communicate that confidence level to the user.

Machine-first formatting - the data is designed to be consumed by software, not read by humans in a browser. JSON, not HTML. Predictable fields, not free-text descriptions that require natural language processing to extract meaning.

This is the gap that exists today. Retailers publish product data for human shoppers. Agents need it in an entirely different form. Someone needs to do the translation work - ingesting messy retail data, normalising it, and serving it through a clean interface that agents can use without friction.

That’s what Aether does. But regardless of whether the solution comes from Aether or someone else, the need is real and growing. As agents take on more commercial tasks, the demand for reliable, structured product data is only going to increase.