Named for the part you can make: AI answers for machine builders and shops

What a technical buyer's question carries

Compare two questions. "Who are the best electricians in Brampton?" sets one condition, a trade in a town. "5-axis machining of titanium parts up to [size], [quantity] a year, near Toronto" sets five: a process, a material, a size limit, a volume and a place. An engineer with a drawing open can type the second kind, because the numbers are in front of them, and add a tolerance, a certificate or a lead time on top.

Each condition is a test a supplier passes or fails, and an answer can say you pass it accurately only where some text says so. With web search on, an AI assistant reads a handful of the pages its searches return before it writes. Google's AI features may run several related searches across subtopics, which Google's documentation calls query fan-out, and a question with five conditions has several subtopics to search. A page that states one of them as a figure gives a search something to find; a brochure line about "precision machining" states none of them.

So for an engineering firm, AI visibility is read per capability rather than per company. A shop can be named for large aluminum parts and missing for titanium, named for turning and never for grinding. A single yes or no across every question hides that; a reading per capability shows which pages are thin.

Named, recommended and first, defined for a supplier

Named means the firm appears anywhere in the answer, under any form of its name, whatever the answer says about it. For a supplier, write down which capability the answer tied the name to, because an answer to a milling question can name your shop for its turning, and that is a different finding from being named for milling.

Recommended means the answer puts the firm forward as suitable for the job in the question, rather than listing it or mentioning it in passing. An AI recommendation may give a reason, and for a supplier the useful reasons are specifications: a tolerance held, a material cut, an envelope, a certificate, a record in an industry. A recommendation whose reason is wrong is recorded twice: as a recommendation and as an error.

First means the firm is the first supplier the reader meets in the answer, at the top of a list or first in prose. The order of names moves between runs, so first place is read across runs and months rather than taken from one answer. Our methodology builds its visibility score from the first of the three readings alone, the share of the twelve questions that named the firm, and reports first place beside it. What separates the three for any business is the subject of the three rungs of an AI shortlist.

The figure beside your name, checked against your own

An engineering answer can name you and still get you wrong in a way that costs work. It can give an envelope larger than your largest machine, state a tolerance you hold only on request as your routine one, name a certificate that has since lapsed, or put you down for a material you do not take. A buyer who reads that sends a drawing you have to decline, or drops you at the first review.

So every figure an answer states about your firm is checked against the figure you publish, one at a time. There are three results: it matches; it differs, and the answer's value is written down beside yours; or you publish no figure for it, and the answer took it from another page or from nothing at all.

A difference is traced to the page that still states the old figure, whether a datasheet PDF, a directory entry or a trade-show listing, then corrected there and dated; the steps are in tracing a wrong fact to its source. A missing figure is a page to write, and the tables that hold it are laid out in capability pages a buyer can compare.

A bracketed panel, one condition per question

The panel is a short list of questions written the way your buyers type them, each built around one capability you want to be named for, with its figures in brackets. Fill the brackets from your own RFQ history: the sizes, tolerances, materials and volumes buyers sent you, including the jobs you turned down. A question built on your own trade name for a process, one no buyer would type, measures nothing.

Two or three questions of each kind are enough for a first reading. That reading is your baseline measurement, and the list then stays as written, so that every later reading compares with it. Questions that contain your firm's name answer a different question, whether what is said about you is right, and belong on a list of their own; the monthly routine keeps the two apart.

What to record for each answer

Every answer gets one row, filled in by a person who read all of it. The first columns are the ones any panel records. The capability, the figures and their match are what a supplier adds, because an engineering answer can name you with figures, and a figure can be wrong.

Keep the whole answer as well, pasted unedited with its date. A match is only as good as the text it was read from, and months later the pasted answer is how you check a verdict. The general record, without the supplier columns, is in the prompt monitoring guide.

Reading the first round, capability by capability

Sort the rows by capability before you count anything. Five patterns are worth telling apart, and each points to a check on your own pages rather than to a cause you can be sure of.

• Never named for a capability you hold. Look for a page that states the process with its figures. If the only mention is a line in a brochure or a PDF, the answer had nothing of yours to repeat.
• Named for a different capability than the one asked. Check which of your pages the answer cited; a page about a process you have moved away from can outlast the change.
• Named with a figure that differs. Trace the figure to the page that states it, correct it there, and date the correction.
• Named, with a directory, a PDF or a show listing as the cited page. Your own pages may state the capability less plainly than that listing does; read the two side by side.
• Recommended for a region or an industry you do not serve. State the service area and the industries in text, and say which you do not take.

Then look at the suppliers named instead of you, capability by capability, as competitor analysis in AI answers describes, and put the fixes in the order the overview for machine builders and machine shops gives. The reading changes nothing by itself; it tells you which capability page to write next.