What is AI Product Attribution? How Retailers Use AI to Close the Search Gap
When Marks and Spencer announced a partnership with Lily.ai in July 2026 to improve its product data and Google Shopping performance, a phrase appeared across retail technology coverage that many retailers had not encountered before: “product attribution”.
The concept is not new. Retailers have been grappling with the underlying problem for years. AI has changed how that problem gets solved. This article explains what AI product attribution actually means, how it works, and why it matters for any retailer selling online.
The merchant-speak vs consumer-speak problem
Every product in a retail catalogue has two descriptions. The first is the one the retailer or brand writes: precise, internally logical, often technical. The second is the one the shopper types into Google, or says to their phone, or enters into a site search bar: loose, colloquial, context-specific.
A home furnishings retailer might list a sofa as “three-seater upholstered occasional chair in stone bouclé”. A shopper looking for the same item might search “cosy cream textured sofa” or “chunky knit fabric corner seat” or “scandi-style settee light grey”. These queries describe the same product. But the product data as written does not contain the terms the shopper used, so the search fails.
This gap between merchant language and consumer language is one of the most costly structural inefficiencies in retail ecommerce. It causes failed onsite searches, missed paid search impressions, and products that exist in a catalogue but are invisible to the customers who want them.
How AI product attribution works
AI product attribution addresses this problem by analysing existing product data and images, then generating additional structured attributes that express the product in the language consumers actually use.
The process typically works in three stages:
Ingestion. The platform takes in existing product data: images, descriptions, titles, category assignments, and any structured attributes already present. It normalises this data, handling inconsistencies in format, completeness, or vocabulary.
Enrichment. Computer vision models analyse product images to identify visual characteristics: material appearance, colour nuances, silhouette, style cues, functional features. Natural language processing analyses existing text data for meaning and category context. The combination generates a set of consumer-oriented attributes drawn from a large proprietary taxonomy, covering terms like occasion, fit, aesthetic, material feel, and use context alongside the standard technical specifications.
Distribution. The enriched attribute data is then pushed to the destinations where it creates value: Google Merchant Center for Shopping and Performance Max campaigns, onsite search indices, personalisation and recommendation engines, and increasingly the structured data layers that AI discovery platforms (ChatGPT Shopping, Perplexity, Gemini) read to surface products in conversational and agentic commerce contexts.
The result is a product that was described one way internally now becoming discoverable across many more of the queries its buyers actually use.
AI product attribution vs AI product description generation
These two capabilities are sometimes grouped together under “AI product content”, but they are distinct approaches that address different moments in the product content lifecycle.
| AI product attribution | AI product content generation | |
|---|---|---|
| Starting point | Product data already exists in some form | Product images and raw supplier data only |
| What it produces | Enriched consumer-language attributes added to existing records | Descriptions, structured attributes, taxonomy classifications, and imagery created from scratch |
| Best used when | Catalogue is live but underperforming in search and paid | Products have no content at all, or existing content is thin, copied, or underperforming |
| Commercial outcome | Improved ranking, click-through, and conversion for existing products | Products become findable for the first time |
Neither approach is superior in all circumstances. They address different stages of the same problem.
A retailer launching 200 new products every month needs content generation: descriptions and attributes where none exist, so those products can be published, found, and sold. A retailer with 50,000 attributed products that are live but underperforming in paid search needs attribution optimisation: a richer consumer language layer on data that already exists.
Many retailers need both. The order matters: generation before optimisation. You cannot enrich what has not yet been created.
What AI product attribution means for SEO and Google Shopping
The commercial impact of product attribution shows up most directly in two places.
Google Shopping and Performance Max. Google’s shopping surfaces rank products based on the quality and completeness of their product feed data. A feed that contains only technical merchant attributes will match fewer search queries than one enriched with consumer vocabulary. Attribute enrichment directly expands the range of queries a product can rank for without changing the product itself. M&S reported improved visibility, higher click-through rates, and measurable revenue lift from their Lily.ai deployment for exactly this reason.
Organic search and SEO. Product pages that contain the specific terms shoppers search for perform better in organic search results than pages using only merchant vocabulary. Enriched attributes improve the semantic coverage of a product page, making it more likely to rank for the long-tail queries that aggregate to significant traffic volume. The same terms that help in Google Shopping also strengthen the organic product page.
AI discovery surfaces. ChatGPT Shopping, Perplexity, and Google’s AI Overviews are increasingly reading structured product data to answer shopping queries directly. Products with richer, consumer-oriented attribute data are more likely to be surfaced in these environments because the AI can understand and match them to the query. Attribute enrichment is effectively AEO (answer engine optimisation) applied at the product data level.
How retailers access AI product attribution capabilities
The challenge historically has been that the most sophisticated attribute enrichment platforms, including Lily.ai, are built for and priced at enterprise scale. Entry points of $50,000 or more per year place them out of reach for the majority of UK retailers.
The same underlying capability (generating structured, consumer-language product attributes from images and existing data) is now available through AI product content platforms built for mid-market and independent retailers.
merchi.ai generates structured product attributes, descriptions, taxonomy classifications, and lifestyle imagery as part of a single configured pipeline. The output is built around each retailer’s specific data schema and attribute model, not a generic template. Grosvenor Flooring used it to clear a 1,000-product backlog and achieve 976% online revenue growth by getting products correctly attributed and findable for the first time.
The platform supports 40+ languages, integrates with the major ecommerce platforms, and is available from £99/month on self-serve terms. It also includes AI lifestyle imagery generation and handles both ends of the product content problem: creating content from scratch where none exists, and optimising existing content that is thin, supplier-copied, or underperforming in search.
For retailers evaluating which approach is right for their catalogue, the starting question is the one above: do your products have content that needs optimising, or do they need content created in the first place? For many catalogues, the honest answer is both, and a platform that handles both in the same pipeline avoids the need to stitch together separate tools for each stage.
If you are not sure where your catalogue sits, a 30-day free trial lets you run merchi.ai across a subset of your own products to see what it produces, or book a 20-minute walkthrough to assess it against your specific product data.
Frequently asked questions
What is AI product attribution?
AI product attribution is the process of using machine learning to analyse existing product data and imagery, then generate additional structured attributes that describe products in the language consumers use when searching. It bridges the gap between how a retailer internally categorises a product and how shoppers describe it in search queries, onsite search, and AI discovery surfaces.
Why is product attribution important for retail ecommerce?
Most retail catalogues are described in merchant language: precise, internally logical, but not aligned with how shoppers search. This mismatch means products are invisible for many of the queries that would find them. Attribute enrichment closes that gap, improving performance in Google Shopping, organic search, onsite search, and AI discovery platforms. M&S’s partnership with Lily.ai was specifically aimed at closing this gap in their product feed.
What is the difference between AI product attribution and AI product description generation?
Attribution enriches existing product data with consumer-oriented attributes. Description generation creates product content from scratch, typically from images and raw supplier data, for products that have no copy at all. Both capabilities address the product content problem; they operate at different stages. Generation is the first step; attribution optimisation follows once the content foundation exists.
Does AI product attribution improve Google Shopping performance?
Yes. Google Shopping ranks products partly based on the richness and relevance of their feed attributes. A product feed enriched with consumer-language terms matches more search queries than one relying on technical merchant vocabulary alone. M&S reported stronger visibility, higher click-through rates, and measurable revenue lift from AI-enriched product attribution in their Google Shopping campaigns.
How does AI product attribution relate to AI search and answer engines?
AI discovery platforms (ChatGPT Shopping, Perplexity, Google AI Overviews) read structured product data when responding to shopping queries. Products with richer, consumer-oriented attribute data are more legible to these systems and more likely to be surfaced in AI-mediated shopping experiences. Attribute enrichment is effectively product-level AEO: making individual products machine-readable for AI recommendation and discovery.
Can smaller retailers access AI product attribution?
Enterprise-only platforms like Lily.ai are priced at $50,000+/year, which limits access to large retailers. The underlying capability (generating structured, consumer-language attributes from product images and data) is now available through platforms built for the mid-market. merchi.ai provides AI product content generation and optimisation, including structured attribute output, from £99/month with no enterprise sales process required. It handles both content creation and content optimisation, so retailers with existing catalogues that need improving are not excluded from the capability.