Commerce (Formerly BigCommerce) Launches AI Product Data Enrichment: What Makes Product Data Selectable by AI Agents, and How Merchants Should Prepare
Commerce launched Feedonomics Enrichment and BigCommerce Catalog Enrichment. This article explains what AI-ready product data actually means, how it maps to Google and OpenAI feed specs, the open questions, and what merchants should fix in their own catalogs.
Key Takeaways
- On September 29, 2026 (U.S. time), Commerce (formerly BigCommerce) announced Feedonomics Enrichment and BigCommerce Catalog Enrichment. Beyond product titles, descriptions and FAQs, they use generative AI to create the structured facts and Q&A that AI uses to interpret what a product is, who it is for and why it fits
- AI agents cannot recommend products they cannot read. Rebuilding catalogs written for human readers into sets of machine-readable facts is becoming a precondition for being found
- Feedonomics Enrichment supports English only, and pricing and performance figures are undisclosed. Merchants do not need to wait for tools: they can start by turning specs buried in images into text and building Q&A from their support inquiry history
The Two Enrichment Products Commerce Announced
New capabilities help B2C and B2B merchants create richer, AI-ready product data at scale and form the foundation of Commerce’s broader agentic commerce...
www.globenewswire.comOn September 29, 2026, Commerce (NASDAQ: CMRC), the parent company of BigCommerce and Feedonomics, announced two new capabilities in a press release. Feedonomics Enrichment is for companies that distribute product data across many channels, and BigCommerce Catalog Enrichment is built into the BigCommerce control panel.
Both start from the product information a merchant already has and generate product titles, descriptions, feature bullets, FAQs and SEO metadata. On top of that, they produce structured facts, snippets and Q&A fields that generative search and answer experiences use to interpret what a product is, who it is for and why it matches a question. The destinations named include advertising channels and marketplaces such as Google, Meta, Amazon and eBay, and conversational AI such as Gemini, ChatGPT and Copilot.
The announcement's premise is clear. If an AI answer engine cannot access the relevant information about a product, that product will not appear in the answer. Traditional catalogs were written for people, not for AI agents.
AI agents can only answer questions about your products as well as your data allows.
What 'Product Data That AI Agents Choose' Actually Means
"AI-ready product data" is a convenient phrase, but it is often used without a clear definition. Breaking down this announcement, what gets generated falls into three layers of different character.
The first is text for people. Titles, descriptions and feature bullets belong here, and they extend the SEO work merchants already do.
The second is facts machines can match. This means holding attributes such as material, dimensions, compatible models and target age as fields, rather than burying them in prose. When answering a request like "machine-washable, 100% cotton indoor shoes for kids," matchable facts are a more reliable signal for AI than the wording of a description. This is the layer the announcement calls structured facts.
The third, and the newest part of this announcement, is material for answering questions. Q&A and snippets fall here, preparing answers to "who is this product for?" and "why does it fit this question?" in a form AI can easily cite. Commerce explains that for a conversational experience to match a shopper's question to a specific product and give a credible reason to choose it, contextual product content is required.
These three layers already show up in the AI platforms' own specifications. In May 2026, Google added conversational attributes to Merchant Center. According to PPC News Feed, there are six, all optional: question and answer, document link, related product, item group title, variant option and popularity rank. They do not affect product approval status.
OpenAI's product feed spec looks different. Attributes such as material, dimensions and review count are defined as optional fields, but the spec states that raw review entries and question-and-answer lists are not part of the current discovery contract. The same "Q&A for AI" may or may not be accepted depending on the destination.
| Data layer | What Commerce's two products generate | Google Merchant Center | OpenAI product feed spec |
|---|---|---|---|
| Text for people | Titles, descriptions, feature bullets, SEO metadata | Basic attributes such as title and description | Title and description are required (aim for at most 150 and 5,000 characters) |
| Facts machines can match | Structured facts and structured fields | Existing attributes such as material and size, plus conversational attributes such as variant options and related products | Material, dimensions, weight, category and more defined as optional fields |
| Material for answering questions | FAQs, Q&A, snippets | The 'question and answer' conversational attribute (optional) | Q&A lists are not part of the current discovery spec |
Gee gave a footwear catalog as an example in an interview with Digital Commerce 360. A retailer might translate it into 15 languages, send it to Amazon, eBay, Google, Meta, TikTok and Walmart, and then also deliver it to answer engines such as ChatGPT, Claude, Copilot, Gemini and Perplexity. She said each needs UCP (Universal Commerce Protocol, a common commerce specification for agents promoted by Google and others) support, and called it "a one-to-many problem."
That one-to-many structure is the real difficulty of product data work. Without a single source of truth for facts, content drifts apart from one destination to the next.
Who Each Product Is For
The audiences are clearly split. Feedonomics Enrichment is aimed at enterprise and multichannel brands with complex catalogs. Gee said it gives "brands on any platform" a data pipeline, so companies running stores outside BigCommerce are in scope too. BigCommerce Catalog Enrichment lets BigCommerce merchants do the work entirely inside the control panel.
| Item | Feedonomics Enrichment | BigCommerce Catalog Enrichment |
|---|---|---|
| Intended users | Enterprise and multichannel brands (including stores on other platforms) | BigCommerce merchants |
| Delivery model | Both self-managed and managed service | Self-serve inside the control panel |
| Quality checks | A scorecard for accuracy, consistency and brand adherence, with an explainable issue list | A proprietary scorecard for accuracy, consistency and adherence to brand voice |
| Distinctive features | Marketplace and advertising fields, mapping to the brand's own taxonomy, analytics across onsite traffic and paid media | Select products, provide brand inputs, review, then apply across the catalog in one click |
| Language | English only for now (multi-language coming) | Not mentioned in the announcement |
| Availability | Listed as 'Now Available' on the official site | Listed as 'Early Access' on the official site |
Availability deserves attention. The press release presents both products side by side, but on Commerce's agentic commerce page, BigCommerce Catalog Enrichment is labeled Early Access.
This is not Feedonomics' first use of generative AI to process product data. In July 2025, it announced product data enrichment using Gemini on Google Cloud. In April 2026, it also released the Agentic Commerce Engine, which converts product data into agent-oriented formats and distributes it. The distribution pipes came first; now the company has productized the content that flows through them.
Why Commerce Is Putting Product Data Front and Center Now
Company circumstances matter here too. In July 2025, BigCommerce renamed itself Commerce.com and changed its ticker to CMRC, describing the move as preparation for an "agentic commerce era" in which AI researches, recommends and completes transactions on behalf of consumers.
On September 10, 2026, it announced an operating plan expected to cut $60 million to $80 million in annualized costs. While reducing costs such as staffing and professional services, it said it would protect investment in B2B, payments, Feedonomics, product intelligence and agentic commerce. The two new products sit squarely in the areas it pledged to protect.
Gee listed four things merchants must get right: be discoverable by agents, put agents to work running the business, convert inside the conversation, and build on open standards so they are not locked in. Data enrichment serves the first of these.
Caveats: Who Vouches for Product Information Written by AI?
The biggest issue is the accuracy of generated content. When AI writes descriptions and FAQs, the risk of it filling in specs that are not in the source data cannot be eliminated. Commerce's design, with a scorecard that evaluates accuracy and merchant review before changes go live, assumes this risk. However, the evaluation criteria and how well errors are detected have not been published.
AI makes mistakes on the other side too. According to a benchmark by Product.ai's data team, which compared major AI systems on 220 shopping questions, the paid tier of Gemini produced at least one likely fabricated claim on 21% of questions. Richer data does not guarantee that AI will use it correctly.
Other points remain unconfirmed.
- Performance figures are undisclosed. The announcement includes no data on how much enrichment increased AI-driven visibility or sales
- Pricing is undisclosed. No pricing structure has been given, including for the managed service
- How generated Q&A is handled depends on each destination's spec. As noted above, OpenAI's current spec has no place to receive Q&A
There is also a brand question. If many companies use the same generation tools, product copy will start to sound alike. What differentiates a merchant will not be polished prose but how accurately it publishes facts only it knows.
What Merchants Should Fix in Their Own Catalogs
Feedonomics Enrichment is English only, and few Japanese merchants run on BigCommerce. For merchants in Japan, the value of this announcement lies less in the tools themselves than in the map it provides of what to prepare so AI can read their products.
The first thing to tackle is turning information buried in images into text. In Japanese e-commerce, many product pages bake materials, dimensions and usage instructions into long vertical images. These are easy for people to read but simply do not exist for an AI reading a feed. The contents of those spec tables need to be re-entered as product data fields.
Next, check whether "who it is for" can be written as facts rather than impressions. That means breaking "recommended for sensitive skin" down into matchable facts such as "fragrance-free, alcohol-free, patch-tested." Recording, as facts in the same way, who the product does not suit and when it should not be used also reduces the room for AI to make wrong recommendations.
The material for Q&A already exists inside the company: the history of inquiries sent to customer support. Real questions and answers about fit, compatibility, shipping and return terms are a firmer basis than FAQs that generative AI writes from scratch. They are also the quickest place to find candidates for Google's conversational attributes.
If you mass-produce copy with generative AI, decide on a pre-publication review process first. For cosmetics and health foods, AI may write claims that violate Japan's Pharmaceuticals and Medical Devices Act or the Act against Unjustifiable Premiums and Misleading Representations. Regardless of whether AI wrote it, the seller is responsible for the representation. The same thinking Commerce built in, quality evaluation plus human review, needs to be brought into your own operations.
Finally, consolidate facts into a single source of truth. If the material or dimensions of the same product differ across Rakuten Ichiba, Yahoo! Shopping, Amazon, your own site and your Google feed, no amount of polished copy will keep AI's judgment steady. The "one-to-many problem" Gee describes is especially serious in Japan's marketplace-centered selling structure.
Conclusion
Commerce's announcement redefines product data for AI agents as more than better titles and descriptions: it is the work of preparing facts machines can match and material for answering questions. A company that built the distribution pipes first has now productized the generation of what flows through them.
What to watch next is the timing of multi-language support, pricing, and figures showing what changed before and after enrichment. How far Google's and OpenAI's product data specs go in accepting Q&A and "who it is for" information will also shape priorities. Even while waiting for the tools, merchants can start aligning the facts in their own catalogs.
