A narrow lead in machine building, stated as evidence a buyer can check

Why "the leading supplier" gives a reader nothing to check

A homepage that calls the firm "the leading supplier of [machine type]" states an opinion. Even when it is true, a buyer cannot check it, and an answer that repeats it can only attribute it to you. The facts behind a real lead can be checked: the tasks your machines do, how many are running and where, in which industries, built to which standards, and where people outside the firm have written about them.

Nothing here makes a niche a shortcut to AI visibility, and no one outside the companies that build the assistants knows how an answer's list of suppliers is put together. What a niche firm controls is whether its pages give a buyer's question, and any answer to it, a checkable fact about the niche to repeat. On the benchmark in the GEO paper by Aggarwal et al. (KDD 2024), sources that added statistics, quotations or citations were more visible in generated answers, by up to about 40% for the best methods. A dated count of machines in service is a statistic of that kind.

Whether you may call yourself the leader at all is a separate question. Rules on comparative and superlative claims in advertising differ between the US, Canada, the EU and the UAE, so the word itself is for your counsel.

Applications solved, written up one at a time

An application is a case study of one problem your machine solved for one customer. Give each its own page or headed section, with the facts an engineer asks for on the first call:

• the task, in the customer's terms, such as pressing [n] bearings into a housing or packing [product] into cases;
• the part and its material, with the size range the machine handles;
• the rate the customer asked for and the rate measured at acceptance, with how and when it was measured;
• the constraint that made it hard: floor space, changeover time, a clean room, a line that could not stop;
• the machine delivered, by family or model, and the month it was commissioned;
• the industry and the region, with the customer's name only where they gave written permission.

Put the whole claim in the first sentence, so that a reader who stops there still has it: "We built a [machine type] that [task] at [rate] for a [industry] manufacturer in [region], commissioned [month year]." Quote a customer only in words they approved, attributed as they agreed. The write-up in general is covered in how to write up a reference project, and what a confidentiality agreement lets you publish in confidential work as evidence.

The installed base, counted, dated and split by region

A lead in a niche rests on a count, so publish it as one: how many machines of the type are in service, where, as of when, and counted how. Separate delivered from in service, say how retired machines are treated, and count from delivery records and service contracts rather than from memory.

Update the table once a year, in the same month, and change the date in the caption. A count written as "[n] machines, as of [month year]" stays true after it is overtaken; "[n]+ machines worldwide" does not. Publishing your own figures with the method attached is the subject of figures a reader can count.

Industries and standards, named the way they name themselves

List the industries you have delivered to and link each one to its applications, so that "we build for [industry]" has its evidence one click away. An industry named with no application behind it is as unchecked as the word "leading".

For standards, write the designation, the title and the edition exactly as the standard's publisher prints them, and say what the relationship is: designed to, built to, tested against, or certified under. Those are four different claims. "Designed and built to [designation], [title], [edition]" is your statement about your own process. A certificate is a third party's statement, and belongs on the page only if you hold it, with the issuer, the number, the scope in its own words and the dates; in markup it is hasCertification with a Certification, no wider than that scope.

Leave out any summary of what a standard requires. The standard's own text is the authority on that, and a paraphrase loses its precision. Which standards apply to a machine, and what conformity marking a buyer's market expects, differ by jurisdiction; that is for your counsel or the authority where the machine is sold. The page says what you did.

Show, paper and patent records a buyer can follow

A mention on someone else's site is evidence that does not rest on your word, as long as a reader can follow it to the source. Give each one a line and a link to the page where it lives:

• a trade show: the show's name, the year, the booth and what you exhibited, linked to the organizer's exhibitor page;
• a technical paper or talk: the title, the authors, the venue and the year, linked to the proceedings or the publisher's page;
• a patent: the number, the title, the grant date and the patent office, linked to the office's public record;
• a trade press article: the headline, the publication and the date, linked to the article itself.

List only what exists and opens. A patent number that leads nowhere, or a paper that turns out to be a brochure, undoes the credit the rest of the list earned. These are brand mentions in the plain sense, and earning mentions honestly covers how to get more of them without paying for placement.

A before-and-after test for the niche pages

The test asks one thing: once the evidence is on your pages, do answers to questions about the niche say anything they did not say before? Run it with the same questions before the pages go up and after.

1. Write four to six questions that describe the niche the way a buyer would, by task, part, rate and industry, without your firm's name. The bracketed panel in named for the part you can make shows how to build them.
2. Take the before reading: every question on each assistant you measure, more than once, recording whether you are named, what the answer says about the niche's suppliers, which figures it gives and which pages it cites. This is your baseline measurement.
3. Publish the evidence: the application pages, the installed-base table, the standards and certificates, the list of mentions. Write each page and its date into your change log, the way the monthly routine keeps one.
4. Leave the questions exactly as they are, and wait at least until your next monthly reading. Crawlers revisit pages on their own schedules, and no one can say in advance when an answer will reflect a change.
5. Take the after reading the same way and set it beside the before: named or not, what the answer now says beside your name, and whether any page it cites is one of the new ones.
6. Count a change only if it holds across runs and assistants. An answer that repeats a fact from the new pages, such as "[n] machines in service in [region]", shows the evidence reached it; a single run that names you once shows nothing yet.

Even a clean result proves less than it seems. Other pages changed in the same month, and the assistants may have too. What the test gives you is a dated before and after on fixed questions, read the way our methodology reads every panel, and that is as much as a supplier outside the assistants can measure. The overview for machine builders and machine shops has the other pages this evidence sits beside.