AI Product Discovery Starts With Product Data, Not Search UI

Retailers are starting to say the quiet part out loud: AI product discovery is not only a search-box problem. It is a product-data problem.

A July 2026 Retail Gazette report said Marks & Spencer partnered with Lily AI to improve product discovery across Google, organic search, and emerging AI-powered shopping channels. The article framed the work around structured product data at scale: richer attributes, better categorization, faster product setup, and cleaner feeds for paid and organic channels.

That is a useful signal for any retailer or marketplace team. AI discovery does not begin when a shopper types a prompt. It begins when a product record is created, attributed, classified, validated, and exported to the systems that discovery layers read.

Why the M&S signal matters

The interesting part of the M&S example is not that a large retailer is using AI. That is no longer surprising. The interesting part is where the AI is being applied: before discovery, at the product-content and feed layer.

For years, product discovery projects often started downstream. Teams tuned on-site search, rewrote product titles, adjusted ad campaigns, or optimized category pages. Those still matter, but AI shopping channels change the dependency chain. A shopping assistant, search engine, marketplace ranker, or retail media platform needs structured facts it can interpret without guessing.

That pushes attention upstream to the product record:

• Is the product categorized correctly from launch?
• Are the attributes complete enough for the channel to understand the item?
• Are values normalized into units, colors, materials, sizes, and controlled terms the platform can compare?
• Are identifiers, variants, availability, and price aligned across feeds?
• Can enriched claims be traced to a source?

When those answers are weak, the AI layer has to infer from thin titles and inconsistent descriptions. When they are strong, the AI layer can retrieve and rank products with much less ambiguity.

AI discovery is a feed-quality problem

The phrase “product discovery” can hide several systems that all consume the same underlying truth: Google Shopping, organic search, marketplace search, retail media targeting, internal site search, recommendation systems, and AI shopping assistants.

Each channel has its own format, but the inputs overlap. They all benefit from accurate identifiers, typed attributes, clean category mapping, current price and availability, variant relationships, and product descriptions that do not contradict structured fields.

That is why feed enrichment should not be treated as a last-mile marketing task. It is infrastructure for every discovery surface that reads product data.

Product attributes are becoming merchandising controls

Traditional merchandising teams thought in pages, collections, banners, and campaign copy. AI discovery adds another control plane: attributes.

A dress that has occasion, fit, fabric, neckline, sleeve_length, color_family, and care attributes can be discovered through many more natural-language paths than a dress with only a title and short description. The same is true outside fashion. A fastener with grade, material, standard, thread, head type, finish, and pack quantity can be found by exact technical intent. A pump with flow rate, voltage, connection type, material, and certifications can be compared safely.

The key is that attributes must be structured, not merely mentioned in prose. AI systems can read prose, but production discovery systems work better when the important facts are discrete fields with consistent values.

What to fix before chasing another discovery layer

Before adding another search tool or AI shopping integration, audit whether the product data can support it.

• Resolve duplicate and near-duplicate products so every channel sees one canonical product identity.
• Define required attributes by category, not as a generic enrichment checklist.
• Normalize units, colors, materials, sizes, standards, and controlled values before feed export.
• Validate identifiers such as GTIN, brand, MPN, and supplier part numbers instead of inventing missing values.
• Keep enriched values tied to evidence so claims can be reviewed, reversed, and trusted.
• Write approved changes back to the PIM, ERP, ecommerce platform, and feed tooling so fixes persist across exports.

This is the layer Claro focuses on. We do not treat AI discovery as a prompt problem. We resolve product identity, enrich missing attributes, normalize values, validate feeds, preserve provenance, and write clean records back to the systems your discovery channels already use.

Book a demo to see how Claro prepares product data for AI discovery

Sources and article inspiration

Related Claro resources

FAQ

What does AI product discovery need from product data? +

AI product discovery needs stable product identity, structured category-specific attributes, normalized values, current commercial fields, and provenance for enriched data. Without those inputs, search and shopping agents must infer too much from titles and descriptions.

Is product feed enrichment the same as better product copy? +

No. Better copy can help humans, but feed enrichment for AI discovery means turning product facts into structured, validated fields that search engines, ad platforms, marketplaces, and shopping agents can parse reliably.

How does Claro improve AI product discovery? +

Claro resolves duplicate product records, enriches missing attributes from trusted sources, validates values, normalizes feeds, and writes clean data back to PIM, ERP, ecommerce, and marketplace systems so discovery channels receive trustworthy product facts.