How manufacturers show up in ChatGPT and AI search when buyers ask for suppliers

This is a how-to for the people who run industrial companies and their websites: owners, engineers, sales and marketing leads at job shops, OEMs, coaters, connector makers, distributors and labs. Everything here you can do yourself. If you'd rather have it done, our page on AEO for manufacturers covers scope, testing and measurement.

AI assistants are built to answer. When your site doesn't state a fact, the assistant reaches for what's typical of companies like yours and writes it down with the same confidence as a fact it found. A shop that never states its lifting capacity gets the capacity shops like it tend to have. A coater whose site says only 'plating' can be credited with a process it doesn't run, or miss the one it runs every day.

The buyer can't tell the guess from the fact. You never see the question, so you never get to correct the answer. Sometimes the assistant does say 'I don't know'. That costs you too: the buyer moves to the next name.

The fix is the same either way. Write the fact down, where an assistant can read it.

In September 2026 we asked AI assistants about two manufacturers. One described an equipment maker's dryer as belt-driven (it's belt-free) and put the company in the wrong city. Another gave a large-part shop a lifting capacity it doesn't have and said it held no certifications. Both companies had the facts. Neither had written them where AI could read them.

The method, the assistants we queried, the sample and the dates are on the results page.

Buyers type the words of the problem they're fixing: the feature, the spec, the failure. An engineer with wet caps jamming a capper searches for drying bottle caps. A drying-equipment maker's pages say 'bottles' throughout and never use his words, so thinner pages that do get read first. That's the story told in Won on Vocabulary.

To find the words, read your RFQs, your quote notes and the questions your sales team answers on the phone. Engineers write the spec into the RFQ. Put those exact strings on the page that answers them: the class, the grade, the thread, the feature name. Keep your own category names too. Assistants match the question to the words on the page, so the page needs the buyer's.

AI assistants read more than your website. They read directories, distributors' listings, association member lists, trade-show exhibitor pages and the PDF capability statement you sent out years ago. When those disagree with your site, the assistant has to pick one, and it may pick the old one.

Start with the basics: name, address, phone, and certifications with their scope. Then the capability claims. A directory that still lists a certification you let lapse, or a process you sold off, is a fact AI can repeat to a buyer. The listings side of this work, including citation cleanup, is covered in our page on local SEO.

AI assistants read a page much as a careful engineer does: title, headings, sentences, tables. A few habits make your facts easy to find.

One capability per page. A single 'Capabilities' page with forty bullets gives an assistant forty fragments and no context. A passivation page with the spec, the alloys and the inspection method gives it an answer.

Facts in text. Numbers inside a photo, a logo wall or a scanned PDF are easy for machines to miss. Put them in sentences, or in a table with units.

A plain first sentence. Say what the page covers at the top, the way an answer would.

Structured data for the basics. Schema markup that states your company, address and services helps machines read the facts. It doesn't replace the sentences.

Let the right crawlers in. Check that your robots.txt file and firewall aren't blocking the crawlers AI search tools use to read pages. Crawlers that collect training data are a separate decision, and yours to make.

You can run a useful check in an afternoon.

Write five to ten questions your buyers really ask, in their words: the part, the process, the spec, the region. Ask at least two AI assistants on the same day, with the same wording. Copy each answer word for word and write the date beside it.

Mark every statement about your company: right, guessed, wrong, or 'I don't know'. For each miss, find the sentence your site is missing, or the listing that disagrees. Fix those, then ask the same questions again on a new date.

Two cautions. Answers change between runs and without notice, so treat each one as a single sample. And being named once says nothing about next time. The date on every answer is what keeps the comparison honest.

Some of what's sold as AI search optimization doesn't hold up for an industrial company.

Pages written only for AI. Filler pages stuffed with phrases read badly to engineers and give assistants nothing new to quote.

Promises of placement. Nobody controls what an assistant says. Be wary of anyone who guarantees you'll be named or cited.

Claims you can't back. An assistant will repeat what you publish, and a buyer will check it at the first audit. Write only what the floor can prove.

Facts only in PDFs and images. They're easy to miss. Put the numbers on the page.

A separate 'AI page'. The pages that serve buyers serve AI. One page per capability, with the numbers, does both jobs.