A website can look clear to a person and still be vague to an AI agent. The next practical layer is not only a company profile. It is a machine-readable description of what the business sells, what each offer is good for, where the limits are, and what the safe next action should be.
A human can look at a product page and understand a lot without reading every word. The photo, price, layout, category, brand feeling, review count and personal context collapse into a quick decision. An AI agent does not have that lived context. It needs explicit facts.
That is why a normal product description is often not enough. "Stylish leather sneakers for the city" may work for a shopper, but an agent still has to infer material, fit, weather limits, return risk, use case, size behavior, shipping, stock, and who this product is not for.
If an agent decides using only the facts you wrote down, which facts did you forget to write down?
In the agent-readability experiment we split 1,500 noindex, follow business profiles into five cohorts: minimal, structured, trust/source-map, agent guidance, and super profile. The goal was deliberately separated into two layers.
Can AI/search/agent crawlers discover the profile layer, follow files, and receive stable 2xx responses?
Does the profile help an AI explain, compare, verify, and route a business with less guessing?
That is technically positive, but it is not yet proof of business value. More reads can simply mean more available files. The stronger question is whether richer machine-readable profiles make an AI agent more accurate, less hallucinatory, better at citing sources, and clearer about next actions.
Company identity is the first layer: who you are, what you do, why you can be trusted, and where official facts live. But buying does not stop at identity. For stores, clinics, agencies, consultants and local services, agents also need offer-level facts.
Name, category, summary, variants, use cases, exclusions, and target customer.
Constraints, geography, availability, budget range, service limits, and reasons not to recommend it.
Source URLs, freshness, owner-confirmed fields, policy links, reviews, credentials, and provenance.
Route to checkout, booking, quote request, contact form, messenger, or human confirmation.
Agntbase Product Layer is a machine-readable product or service description package for AI agents. It does not replace your storefront, catalogue, booking page or payment provider. It gives agents a clearer factual layer to read before they compare, shortlist or route a buyer.
Crawler access proves the layer can be reached. The next benchmark must test interpretation quality. For the same business, compare answers with and without Product Layer and measure whether the agent can extract services, identify constraints, cite official sources, avoid unsupported claims, and choose the correct next action.
Does the agent describe the offer correctly and avoid inventing details?
Does it see fit, limits, price range, location, availability and trust signals?
Does it send the buyer to checkout, booking, quote request or human confirmation?
The commercial question is not "did a bot read more files?" It is: if a buyer-side AI agent has to shortlist three options from ten similar businesses, which one is easier to understand, verify and route to safely?