When an engineer asks an AI assistant to spec a component, the assistant does not browse. It fetches, parses, and cites whatever it can actually read — and if that is not your catalogue, it is a competitor's. We graded 984 manufacturer and distributor catalogues worldwide on one question: can an AI agent read them? The median scored 50 out of 100. Not one reached an A. Partsgraph is the layer that fixes it. We read from what you already run — PIM, ERP, your website, your datasheets — and build one canonical record for every part you sell, with each specification stored as a value an agent can filter on rather than a sentence buried in a PDF. That record is published where agents actually look: an address they can query directly, specifications written into your own pages, plain-text versions, product feeds, and the exports the EU Digital Product Passport rules will ask for. Then we measure it. Every week we ask the assistants the questions your buyers ask, and check the answers — how often your parts are named, whether the figures quoted are yours and correct, and which competitor is named instead. Your website does not change. Nothing enters your critical path. It is an overlay, not a migration. Grade your own catalogue free at partsgraph.ai/audit — no signup, 60 seconds.