How to make your catalog readable by AI agents — built on existing open standards, not a proprietary Xibrary format. Useful whether or not you ever work with us directly.
AI agents are increasingly the ones researching and sourcing parts on a person's behalf — see why AI agents will do the shopping. An agent can't reliably read a page designed for a human — marketing copy, inconsistent formatting, specs buried in a PDF datasheet — the way a person can. The gap between "a human could find this" and "an agent can act on this" is closing fast, and catalogs that stay purely human-readable risk becoming invisible to an entire, growing category of buyers.
The single highest-leverage change: add schema.org Product structured data (JSON-LD, in the page's <head> or inline) to every product page — name, SKU, price, availability, and additionalProperty for specs like voltage, torque, or dimensions. This is not a new or proprietary requirement; it's the same markup Google has used for rich search results for over a decade, and any AI system doing web-based retrieval already knows how to parse it.
A CSV, JSON, or XML export of your catalog — even a simple one, updated on whatever schedule is practical — is far more reliable for an agent to consume than parsing rendered HTML. This isn't a novel ask: electronics distributor aggregators already do exactly this in practice — Nexar (Octopart's parent) ingests manufacturer and distributor part data via structured CSV feeds as standard practice. Use GS1's GTIN for product identifiers where applicable — it's the existing global standard for exactly this, not something specific to us.
llms.txtA short, plain-text summary of what your site offers and where the important pages live, at /llms.txt — the emerging convention for orienting an AI system quickly, the same way robots.txt orients a crawler. Ours is at xibrary.havencommand.com/llms.txt as a working example.
robots.txtA lot of ambiguity for both AI systems and honest indexers comes from a robots.txt that's silent or overly broad. If you're open to AI/research crawlers indexing product data but not to competitors scraping pricing in bulk, say so explicitly with a scoped User-agent block, rather than leaving it to guesswork or a blanket disallow that blocks everyone equally.
Where it applies, export control classification (ECCN) and authorized-distributor status are exactly the kind of information an agent needs attached directly to structured data to act safely — see why compliance has to become machine-readable. A datasheet PDF a human might read once isn't a reliable channel for this if the goal is letting software use it correctly.
If your site already follows these guidelines, Xibrary can index your catalog cleanly with attribution back to you, at no cost during our current build phase — see the Xibrary homepage for what that looks like. If it doesn't yet, adopting them is worth doing on its own merits: they're existing, widely-used standards that make a catalog more useful to any AI system, not just ours.