The vocabulary of a machine-readable catalogue

Product data that an AI agent can retrieve, parse and cite without a browser, a human, or guesswork — reachable at a stable address, present in the served response rather than assembled afterwards, typed rather than described in prose, and attributable to a source.

Why it mattersAn assistant that cannot retrieve your data does not fall back to your brochure; it falls back to whatever a third party wrote about your part, or to a competitor whose data it could read.

See alsoStructured dataServer-side renderingModel Context Protocol

An automated client that fetches web pages on behalf of an AI system, either to build training corpora or to answer a user's question live. Each operator publishes its own user-agent names, and the two purposes are often served by different agents from the same operator.

Why it mattersBlocking or allowing them is not one decision. A site may reasonably want to refuse training crawls while remaining fully readable to the agent answering a buyer's question right now — and that distinction is expressed per user-agent.

See alsorobots.txtServer-side rendering

Server-side rendering means the page's content is present in the HTML the server returns. Client-side rendering means the server returns a shell and the content is assembled by JavaScript in the browser afterwards.

Why it mattersPublished analysis of AI-crawler traffic has found the major assistants' crawlers fetch JavaScript files but do not execute them. To those crawlers, a client-rendered catalogue is an empty page — regardless of how complete it looks to a human.

See alsoAI crawlerAgent-ready product data

The practice of making content retrievable and citable by AI assistants rather than only rankable in a list of links. It emphasises being reachable, parseable, unambiguous and attributable, since the outcome sought is a citation inside an answer rather than a click.

Why it mattersFor a catalogue the work is mostly not editorial. It is reachability, rendering, structure and identifiers — the same things that make product data usable anywhere else.

See alsoAgent-ready product dataServer-side rendering

A fixed set of questions with known-correct answers, used to measure whether a system still answers correctly after it changes. Each run is scored against the same expected answers, so a regression shows up as a number rather than as an anecdote.

Why it mattersWithout one, 'the assistant gets our parts right' is an impression. With one it is a measurement, and you find out the day it stops being true rather than the quarter after.

See alsoAgent-ready product data

An open protocol for exposing tools and data to AI assistants over a defined interface, so an assistant can call a capability directly instead of scraping a web page and inferring what it found. A server declares the tools it offers; a client discovers and calls them.

Why it mattersFor a catalogue, it is the difference between an assistant reading a rendering of your product and asking your system a question — parametric search, part lookup, alternates, compliance — and getting a typed answer back.

See alsoAgent-ready product dataParametric search

A proposed plain-text file at the root of a site that gives language models a short, curated index of what the site contains and where the substantive content lives. It is advisory, not a control: it tells a model what is worth reading rather than what it may read.

Why it mattersIt is the cheapest signal a catalogue can publish, and adoption is still low enough that publishing one is a differentiator rather than table stakes.

See alsorobots.txtAgent-ready product data

The file at the root of a site that tells automated clients which paths they may request. Standardised by the IETF as RFC 9309 in 2022, it is a statement of permission that well-behaved crawlers honour — not an access control, and not enforcement.

Why it mattersMost industrial sites wrote theirs before AI crawlers existed, so their policy toward those crawlers is whatever a years-old wildcard rule happens to imply. That is a decision by accident.

See alsollms.txtAI crawler

Machine-readable markup embedded in a page that states what the page is about in a shared vocabulary — for a product, typically schema.org's Product type carrying identifiers, price, availability and specifications as named fields rather than as sentences.

Why it mattersIt removes the inference step. Without it a machine has to work out that a number on the page is a price; with it, the page says so.

See alsoGTINManufacturer Part NumberAgent-ready product data

Selecting a part by filtering on its measured characteristics — voltage, thread pitch, ingress rating, temperature range — rather than by browsing categories or matching keywords. It requires each characteristic to exist as a typed value with a unit, not as text in a description.

Why it mattersIt is how engineers actually specify. An assistant asked for an IP67 8-pole M12 connector is performing a parametric query, and can only answer it against data typed well enough to filter.

See alsoETIMModel Context Protocol

The identifier a manufacturer assigns to a part. Unlike a global trade identifier it is only unique within that manufacturer, and in practice it is written inconsistently — with and without separators, suffixes, packaging codes and regional variants.

Why it mattersMost cross-referencing failures are MPN normalisation failures. If the same part appears under four spellings, an assistant will treat it as four parts, or fail to match it to the one a customer asked about.

See alsoEntity resolutionGTIN

A globally unique trade item identifier administered under the GS1 system, encompassing the number encoded in a retail barcode. Unlike an MPN it is unique across all manufacturers.

Why it mattersIt is the strongest key for matching a product across systems, which is why AI shopping surfaces and marketplaces ask for it first.

See alsoManufacturer Part NumberStructured data

Deciding that several records describe the same real-world thing, and collapsing them into one canonical entity with the variants recorded against it. In a parts context this covers part numbers, variants, packaging options and the same part listed by multiple distributors.

Why it mattersIt is the difference between a catalogue and a graph. Alternates, cross-references and 'what else fits' questions are all downstream of getting this right.

See alsoManufacturer Part NumberParametric search

An open, international classification model for technical products. It defines classes of product and, for each class, the features that describe it and the units and value lists those features may take — so a technical specification becomes comparable across manufacturers rather than being a free-text table.

Why it mattersIt is already the shared language of electrical, HVAC and building-products supply in much of Europe. A catalogue mapped to it is one an assistant, a wholesaler and a regulator can all read the same way.

See alsoECLASSBMEcatParametric search

A cross-industry data standard for classifying and describing products and services, widely used in procurement and industrial automation. Like ETIM it supplies both a classification hierarchy and standardised property definitions.

Why it mattersWhere ETIM dominates electrical and building products, ECLASS is more common in industrial automation and process industries. Many manufacturers need both.

See alsoETIMBMEcat

A widely used XML format for exchanging electronic product catalogues between suppliers and buyers, commonly carrying ETIM or ECLASS classified data as its payload.

Why it mattersIt is how catalogue data already moves between trading partners. A catalogue that can produce clean BMEcat usually has the underlying structure an agent interface needs.

See alsoETIMECLASS

A structured, machine-readable record of a product's characteristics — materials, compliance, repairability, provenance — accessible from a data carrier on or with the product. It is introduced under the EU's Ecodesign for Sustainable Products Regulation, with obligations set per product group through delegated acts.

Why it mattersIt makes machine-readable product data a legal requirement rather than a marketing advantage, for a widening set of goods sold into the EU. The work overlaps substantially with making a catalogue agent-readable.

See alsoETIMStructured data