AI visibility for a US manufacturer or industrial supplier: the spec in text, the named standard, and the PDF that hides both

A US manufacturer or industrial supplier gets named by ChatGPT or Gemini when the specification sits in readable text on a page the engines can fetch. The buyer is not searching for your brand. They search with a process, a material grade, a standard designation, a tolerance or a part number, and the answer is assembled from whichever pages carry those strings in text. In our benchmark run on 15 October 2026, ChatGPT and Gemini each answered 102 buying questions in the United States, which is 204 US readings, and cited our own domain zero times on either engine. Across the whole run of 597 readings, 85.6 per cent of 10,652 citations went to somebody's own website and 9.7 per cent to official platform documentation. The reading for a shop floor is blunt: capability that lives in a PDF, a CAD viewer or a quote form is not in the text those answers are made from.

What the run covered, and what it did not

One dated measurement, two engines, three markets. ChatGPT and Gemini were read on 15 October 2026 in the United States, the United Kingdom and India, 597 usable readings of 612 attempted. These were general buying questions put market by market, not a study of each industry, so the run holds no machining number and no fabrication number. It holds the source mix an American answer is built from, which is the part a manufacturer can act on. The method and the citation table are in what AI assistants actually cite across 597 readings of ChatGPT and Gemini in three markets, with the American slice in the US cut of those 204 readings.

Five surfaces are named in the product: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Two of the five were in this run, so nothing below applies to Perplexity, Claude or AI Overviews. They also behave differently enough to test separately: ChatGPT's most cited domain appeared in 176 of its 306 readings, Gemini's in 19 of its 291.

How a sourcing question is actually phrased

What the buyer has	How they ask it	What has to be in your text
A print and a volume	Who can hold this tolerance on this process at this annual quantity	Process list, routine tolerance band, inspection equipment
A material callout	Supplier for this grade in this form and this temper	Grades and forms by name, not all metals
A standard	Shop certified to this standard for this application	Certificate, registrar, accreditation body, scope
A part number	Cross reference or equivalent for this part	Part number pages and a cross reference table
A contract requirement	Domestic supplier that meets this procurement rule	Registration details, country of melt and manufacture

Every row is a filter. Pass it and you are on the shortlist; fail it and nobody tells you, because a buyer who cannot confirm a fact moves to whoever published one.

The PDF is the most expensive habit in the trade

Most industrial sites hold the right facts in the wrong container. The capability statement is a brochure, the dimensions a drawing, the alloy table a scan, the tolerance chart an image in a download. A fetcher reads none of that reliably, so the sentence a model needs exists nowhere on your domain.

The fix is unglamorous and it works. Take the three documents your sales team emails most and republish their contents as HTML: a heading per section, the numbers in a real table, units stated, and a review date. Keep the PDF for printing and put the same content above the download link. Then view source on a product page and search for a dimension. If the number is not there, no assistant has read it.

One page per named standard, one page per part family

Two page types do most of the work here, and almost nobody publishes either.

• A page per standard you genuinely work to, named exactly as the buyer writes it: ASTM, ASME, SAE and AMS designations, ANSI and NEMA where they apply. Say what it governs, what you do to meet it, which processes it covers, and what paperwork ships with the part, from a certified material test report to a first article inspection report.
• A page per part family or part number, with the cross reference table on it. Interchange and equivalence questions are among the commonest things an industrial buyer brings to an assistant, and that answer usually comes from a distributor catalog rather than the maker.

Put the bands on those pages as bands: routine tolerance against the tighter one that costs more, lead time as a range. Stated numbers are what let a supplier be slotted into a shortlist.

The compliance facts an American buyer filters on

These separate a shortlist from a logo wall, and most already sit in the quality office.

• Quality registration with the detail attached: ISO 9001, AS9100 or IATF 16949, with the certificate number, the registrar, the accreditation body behind it, the scope and the expiry. Scope matters more than the logo: a certificate covering one cell is not a plant wide claim.
• Export control status, plainly stated, for anyone touching defense or dual use work, including whether you are registered for ITAR controlled articles.
• Procurement eligibility: Buy American Act, Berry Amendment or Trade Agreements Act work, your SAM.gov registration, CAGE code, NAICS codes, and country of melt for metal.
• Substance compliance where it applies, such as RoHS and REACH declarations, dated.

A certificate number with the registrar beside it is also the cheapest way to meet what the engines asked for: 102 of 102 build spec conversations demanded a date and a last verified line on any claim, and 80 of 102 asked for the sample size behind a number.

The layer you do not control

Industrial answers lean on distributors, catalogs and sourcing directories, because those pages are structured, filterable and current. You can influence that layer without owning it: make the distributor listing carry the same part numbers, grades and certifications as your pages, and check nothing in the catalog is two revisions behind your drawing. A buyer comparing a stale listing with your site reads the gap as risk.

Resist turning that into a directory campaign. Review directories took 0.8 per cent of the 10,652 citations in our run, and industrial directories were not counted separately, so treat listings as half a day of housekeeping and spend the week on the spec pages. For the organic search half of the same job, what actually changes by industry is a shorter list than most agencies suggest.

Reading your own position, in one afternoon

1. Write six sourcing questions as a buyer would, with a process, a grade, a tolerance and a volume in them, naming nobody.
2. Ask each one twice in ChatGPT and twice in Gemini, in a temporary chat with memory off.
3. Record which suppliers were named, which catalog was named instead, and save the sources.
4. Pick the two questions where a competitor was named and you were not, and publish those two pages this month.

We sell the fix, so read the argument with that in hand. AI Knows Us is our product, and on an industrial site most of the job is turning the quality folder and the capability brochure into pages, then getting those facts into the catalogs. 64 of the 102 build specs wanted that said outright, and the run behind this page records our own 3 citations in 597 readings, none American. For the shop floor version of the same walkthrough, see how a manufacturing company gets recommended by AI.

Questions about AI visibility for a US manufacturer or industrial supplier

We sell through distributors. Is this still our job?

Yes, because the distributor page quotes your specification and certifications, and whatever is wrong there came from you. Publishing the authoritative version on your own domain gives every catalog one current source to copy, and the engines a page with your name on it.

Our drawings are confidential. What can we publish?

The envelope rather than the design: process, size range, material grades, routine tolerance band, finishes, inspection equipment and lead time range. None of it reveals a customer part, and all of it is what a buyer filters on.

Why does the answer keep naming a distributor instead of a manufacturer?

Because the distributor published the specification as structured text while the manufacturers published a brochure. Across the run 85.6 per cent of citations went to somebody's own site, and the quoted pages are the ones carrying facts in text.

How do we show a price when everything is quoted?

Publish the structure instead of a rate: what you price per, what moves the number, tooling terms and who owns the tool, payment terms, and the date you reviewed all of it in dollars. All 102 build specs asked for pricing in the reader currency, and a structure answers that where a quote form cannot.

The page a design engineer would actually use

The test for all of this is one page and four minutes. Could an engineer who has never heard of you open a single page, read what you run, in what materials, to what tolerance, under which standards and with what documentation, and decide to send you a print? That page is also the one an assistant can quote, because both readers want the same thing: facts in text, with units and a date on them. Most industrial sites cannot pass that test, which is why the suppliers that publish it are the ones getting named.