Fanplayr Launches Catalog Intelligence to Make Product Catalogs AI-Readable, and What the AI Visibility Score Actually Measures
Starting from Fanplayr's Catalog Intelligence launch, this piece maps the eight AI Visibility Score areas against the product data requirements Google and OpenAI have already published, and where to start fixing a catalog.
Key Takeaways
- On August 25, 2026, Fanplayr announced Catalog Intelligence, which converts product catalogs into structured data that AI can interpret. Every engagement opens with an AI Visibility Assessment that scores readiness across eight areas, from product completeness to category attributes
- The same concern already shows up in platform specifications. Google added Merchant Center attributes for conversational surfaces, and OpenAI's product feed accepts Google-compatible formats. Product data work is becoming a requirement set by the destinations, not a vendor talking point
- AI visibility scores are measured differently by every vendor, so absolute values do not compare. With pricing and customer results undisclosed, the practical move is to start with identifiers and variant structure, which you can fix yourself
Product catalogs were never built for AI in the first place
Fanplayr today announced Catalog Intelligence, a new solution that transforms traditional product catalogs into AI-ready product intelligence.
pressreleasehub.pa.mediaThe claim that poor product data costs retailers AI-sourced revenue is not new. What makes this launch worth reading is that it starts the commercial relationship with a diagnosis and a score.
Fanplayr, which works in AI-powered conversion optimization and personalization, announced Catalog Intelligence on August 25, 2026. Built on the company's Verada AI platform, it analyzes, enriches, and organizes product data so that AI search, recommendation engines, and marketplaces can understand a retailer's products more easily. The solution is available globally, and Fanplayr operates across North America, Europe, Asia, and Latin America.
What it generates is category-specific attributes and classifications, product facts, specifications and identifiers, audience and use-case signals, semantic search language, variant relationships, and export-ready structured data. The company frames this as more than filling in missing fields. The stated goal is to give AI the context to understand what a product is, who it is for, how it should be classified, and when it should be recommended.
Simon Yencken, CEO of Fanplayr, put it this way.
We're entering a new era of online shopping where AI is becoming a primary way consumers discover, evaluate, and purchase products. Most product catalogs were never designed for that world.
Source: Simon Yencken, CEO of Fanplayr
Fanplayr has been moving in this direction for the better part of a year. In December 2025 it announced Verada AI Tags, which automatically generate contextual tags by interpreting products the way a human shopper would. Catalog Intelligence reads most naturally as an extension of that tagging work into full-catalog structuring and assessment.
The eight areas behind the AI Visibility Score
Every Catalog Intelligence engagement begins with an AI Visibility Assessment. Products are evaluated against the information AI expects for their category, producing a score that represents readiness for AI-powered product discovery.
The assessment covers product completeness, category attributes, classification, identifiers, images and media, audience and search signals, variant structure, and overall AI readiness. A complimentary AI Visibility Snapshot is offered at cia.fanplayr.com, which analyzes a representative sample of products within minutes.
One caveat deserves stating plainly. This is a vendor press release. Pricing, named customers, and measured revenue impact from score improvement are all undisclosed. The scoring methodology is described only as those eight areas, with no published detail on weighting or thresholds.
Platform requirements are already written down as specifications
Setting the vendor's claims aside, it is faster to look at what the destinations actually ask for. Over the past year, product data requirements have moved past advice and into published specification.
OpenAI publishes a product feed specification that calls for a UTF-8 tab-delimited or comma-delimited file with one product or variant per row. Required fields include id, title, description, link, and image_link. The notable part is that a Google-compatible product data feed can be uploaded without renaming its columns to OpenAI field names. In practice, feed operations are consolidating on top of Google's product data specification.
On variants, OpenAI's best practices go into real detail. Use a stable product ID for the parent and a unique variant ID for each purchasable option. Keep title, url, description, media, availability, and price variant-specific when those values differ. Put user-facing option dimensions such as color and size into variant_options. The "variant structure" area in Fanplayr's assessment points at exactly this class of requirement.
Google has gone further still. In January 2026 the company added a large set of product data attributes aimed at AI Mode, Gemini, and Business Agent, and published the Universal Commerce Protocol, an open standard co-developed with Shopify, Etsy, Wayfair, Target, and Walmart for agents to discover products and complete checkout. Among the new attributes, popularity rank, document link, question and answer, and related product carry a Merchant Center help note stating they are primarily intended for conversational experiences such as AI Mode, as PPC Land reported. That attributes are now being designed for conversational interfaces by name is the part that matters.
Feeds are not the only path. Shopify offers Storefront MCP, which lets any AI assistant connect directly to a store's catalog, cart, and policies. The shift is less about having more places to send a feed and more about product data being read directly by external agents.
| Destination | How data is handed over | What reportedly matters most |
|---|---|---|
| Google Merchant Center | Product feed (attribute-based) | Identifiers such as GTIN, variant attributes including color / size / material, plus popularity rank, question and answer, and related product added for conversational surfaces |
| OpenAI (ChatGPT product feeds) | UTF-8 TSV / CSV upload or API. Google-compatible formats are also accepted | Required fields such as id, title, description, link, image_link, separation of parent product ID from purchasable variant ID, and variant_options |
| Shopify Storefront MCP | Connection to catalog, cart, and policies through an MCP server | Structured product information usable directly for product search and cart operations |
| Your own site (AI crawlers) | schema.org Product structured data | Price, availability, reviews, ratings, and consistency with the values in the feed |
Lined up this way, what Catalog Intelligence offers looks less like an invention and more like a single service that absorbs requirements the platforms have each issued separately.
How to read a product sold as a score
The number of players quantifying AI visibility has grown sharply over the past year, which makes it worth understanding the limits built into the format itself.
The fundamental issue is that LLM output is probabilistic. As a comparison of LLM visibility tracking tools notes, the same prompt does not return the same answer every time, and results shift with phrasing, model version, and the content environment at query time. A single snapshot is unreliable; repeated sampling of the same prompts is what produces a statistically stable picture. On top of that, some tools track live sessions while others use synthetic prompts. Different methodologies produce meaningfully different numbers for the same brand.
Treating a free snapshot score as an absolute measure is therefore the wrong use of it. Where it earns its keep is in relative comparison before and after your own changes, or in the qualitative list of what is missing.
There is also an incumbent field here. Product information management vendors including Salsify, Akeneo, and Syndigo each position their platforms around AI readiness for product data, and Akeneo's 2026 spring release introduced a feedback loop that improves product data continuously from real-world signals. Judged purely as catalog enrichment, Fanplayr is entering a crowded market. Its differentiation will likely rest on how the catalog layer combines with the behavioral data and personalization platform the company already runs.
Where to start
Working through what you can fix yourself, the order of priority is fairly clear.
Check identifiers first. GTIN is one of the strongest matching signals available, and missing or incorrect values drop products out of the clusters where listings from multiple retailers are aggregated. If this layer is broken, adding attributes downstream will not compensate.
Variant structure comes next: separating parent products from purchasable units, making color and size dimensions explicit, and reflecting price and availability that differ by variant. As noted above, OpenAI and Google ask for the same shape here.
Category-specific attributes and audience or use-case signals follow. Conversational queries run longer than traditional search and read more like questions. Without data describing who a product is for and when it is used, those queries will not find it.
Finally, confirm that feed values and the schema.org Product structured data on your own site agree. When the two disagree, both lose credibility as signals. That order holds whether or not you run an assessment tool.
Conclusion
The announcement itself is one vendor's new product. The pattern behind it is far broader. Product data has long been raw material for building pages that humans read. It is now becoming input for machines that read and recommend.
The proliferation of readiness scores will likely continue for a while yet, and checking where a number came from and how it was measured will remain necessary work whichever tool you use. Cleaning up identifiers, variant structure, and category attributes, on the other hand, does not go to waste regardless of which score turns out to be right. That is the place to begin.
