Blog
What we measured
Findings on how AI systems read product data: what they get right, where they invent, and what makes a product record verifiable. Every model is named and every observation dated.
Your Product Can Be Correct Everywhere — and Still Be Wrong in AI
The FCC Record Was Correct. The Manufacturer Was Correct. The AI Answer Still Needed Review.
Why ProductKnowledgeGraph Pilots Start With Measurement, Not Integration
FDA Data Can Be Correct While the AI Interpretation Is Wrong
AI Doesn't Read a Product Page. It Reconstructs a Product.
Enterprise AI14 August 20262 min read
Your Product Can Be Correct Everywhere — and Still Be Wrong in AI
AI can retrieve the right facts and still build the wrong product.
Enterprise AI11 June 20263 min read
The FCC Record Was Correct. The Manufacturer Was Correct. The AI Answer Still Needed Review.
A correct regulatory record is not enough. AI also needs to know what the record is a record of.
Product Announcements03 May 20263 min read
Why ProductKnowledgeGraph Pilots Start With Measurement, Not Integration
Before trying to improve how AI understands a million products, prove that you can change what it understands about ten.
Enterprise AI21 April 20262 min read
FDA Data Can Be Correct While the AI Interpretation Is Wrong
For regulated products, knowing the right fact is not enough. AI must know which product the fact belongs to and what the authority actually said.
LLM14 March 20262 min read
AI Doesn't Read a Product Page. It Reconstructs a Product.
An answer is assembled from manufacturer pages, regulatory databases, retailer listings and component documentation. The AI is not reading your page — it is reconstructing an identity.