When the buyer asks an AI instead of a search engine
Industrial buyers are starting to ask assistants, not search boxes. What that changes for a supplier's website, what it does not, and what we did to ours.
A buyer who once typed bandsaw blade suppliers into a search box and worked down ten blue links now increasingly asks an assistant a whole question: what blade do I need to cut 60mm 304 stainless bar, and who sells one? The assistant answers in prose and cites a handful of pages. For the supplier there is no page two to sit on. Either your page is one an answer engine can read, extract and cite, or you are not in the conversation at all.
What actually changes
Less than the noise suggests, and the honest baseline comes from Google itself: its documentation says the established SEO fundamentals are what its AI features build on, and that a page needs nothing special beyond being indexed and eligible for a snippet. What shifts is the unit of consumption. A search engine ranks whole pages; an answer engine quotes passages. A page whose sections each answer one question in plain, complete sentences, with the concrete numbers in tables rather than trapped in marketing prose, gives an engine something it can lift and attribute. A page that is one long unbroken claim gives it nothing to hold.
Structure is now legibility
Structured data used to be a garnish for rich results. For machine readers it is closer to a contract: an Organization node says who is speaking, a Product node with real offers says what is sold and at what price, an FAQPage says which questions a page answers, a BreadcrumbList says where the page sits. None of it is visible to a human visitor and all of it removes guesswork for a machine one. The same goes for boring hygiene: one canonical URL per page, an honest sitemap, hreflang that tells the truth about which languages actually exist.
The polite conventions are young
Beyond the established fundamentals sits a newer, less proven layer. llms.txt is a proposed convention, a curated map of a site for AI systems at a well known URL, with a full text companion for the sites that want to offer one, and there is no public evidence yet that any major engine consumes it. We publish both anyway, for the same reason a shop labels its shelves before the auditor asks: the cost is an afternoon, the file is honest whether or not anyone reads it, and the crawlers that do read sites this way are named in our robots file and welcomed rather than blocked. Graded claims matter here: this is a bet on a convention, not a measured channel.
The same discipline as quoting
The deeper change is one industrial suppliers are well placed to win, because it is the discipline their own trade already runs on. An answer engine deciding what to cite is doing what a buyer's engineer does with a datasheet: looking for concrete figures, stated sources and claims that do not exceed their evidence. Pages that carry real specifications, real prices and real limitations, the load table with its reference, the price with its currency, the roadmap item marked as a roadmap item, are the pages a machine can safely repeat. Puffery is not just unpersuasive now. It is unquotable.