How buyers find suppliers in ChatGPT, and how a manufacturer gets on the shortlist

This week I asked ChatGPT for a website and marketing agency for manufacturers in Central Texas. Uwindi came up third of four. I asked a similar question about Texas as a whole. We were gone.

Nothing about our work changed between those two questions. What changed was how specific the question was. That is how more B2B buyers and procurement teams now find suppliers: they ask ChatGPT, Perplexity or Gemini to recommend a few companies and build a supplier shortlist. The answer usually names three to six. If yours isn't one of them, you never see the visit, the click or the RFQ you didn't get.

For ten years I ran lead generation for industrial and B2B companies: ads, SEO, CRM, sales teams. Now we run a fixed set of buyer questions through AI search every month, on our own name. Here is what buyers ask, what we saw, and what a manufacturer can actually fix.

What buyers ask

Buyers don't type "CNC machine shop". They describe the job:

• "Who can machine 6061 aluminum housings, 500 pieces a year, ISO 9001, in Texas?"
• "Sheet metal fabricators near Austin with powder coating in-house"
• "Contract manufacturers for medical device components, small batches, lead times under four weeks"

Every one of these prompts carries qualifiers: a process, a material, a quantity, a certification, lead times, a place. The assistant looks for companies whose public information matches those qualifiers and that other sources agree on. A supplier whose homepage says "we do precision machining" matches none of them.

What I saw when we tested it ourselves

Every month we ask ChatGPT the same set of buyer-style questions about our own business. Logged out, a new chat for each question, from an Austin location. We write down who gets named and what is said. One run is a snapshot, not a ranking: answers change from run to run. Still, a few patterns held.

The qualifier decides. With Austin or Central Texas in the question, ChatGPT named us and showed our map listing with its rating. With "Texas" alone, it named agencies that focus on manufacturing. The answer followed the most specific fact it could match.

It repeats what you wrote. ChatGPT described us in sentences very close to our own pages: where we're based, who we work with, our fixed price. Whatever your site states plainly is what the assistant will say about you. Whatever it only implies, the assistant can't use.

It takes labels literally. In one answer it cited a smaller project of ours as our manufacturing example and skipped the case I would call our strongest. I think the label decided: the smaller project read as manufacturing at a glance, and the stronger one didn't say so as plainly. ChatGPT is a very literal reader. If a fact matters, write it out.

Old data lingers. A week after our relaunch, a search engine was still showing the project count and the pages of the old site. AI answers draw on these indexes, so a stale description in a directory or an old page can keep coming back for months.

It works fast when the facts are there. On one site we launched this year, the first leads from AI answers came within days of launch. That site said what it does, where, and for whom, on every key page.

Why a supplier gets named or skipped

From these checks and the sites we build, the companies that get named have a few things in common:

1. Capabilities in plain words, on their own pages. Process, materials, tolerances, part size, volumes, certifications, industries served. A PDF equipment list is a fine document for an engineer. For an assistant it is close to invisible.
2. One clear sentence about who they are. "[Company] is a [process] shop in [city, state] that makes [parts] for [industries]." Assistants quote sentences of this shape almost word for word.
3. The same facts everywhere. The website, the Google Business Profile, industry directories and review sites say the same thing about what you make and where you are. When sources disagree, the assistant hedges or leaves you out.
4. Proof from third-party sources. Reviews, directory listings, a case study, a mention in a trade publication. An assistant names a supplier with more confidence when it isn't taking the supplier's word alone.
5. A local signal for local questions. For "near me" and city questions, a complete map listing with reviews carries a lot of weight.

This work goes by names like generative engine optimization (GEO) or AI search optimization. In practice it comes down to clear, consistent facts.

How to check where you stand

You can run the same test in half an hour:

1. Write five to ten questions the way your buyers would ask them when they look for a supplier, with the qualifiers: process, material, certification, region. Add one with no location.
2. Ask each one in ChatGPT logged out, in a new chat, so your own history doesn't shape the answer. Then ask the same questions in Google and read what AI Overviews says.
3. Write down whether you're named, in what position, what is said about you, which citations (source links) the answer shows, and who else appears.
4. Repeat monthly with the same questions. The trend over two or three months tells you more than any single answer.

Read closely what the assistant says about you. Wrong facts, an old address or an outdated service list usually point to the source it is reading.

What to fix first

If you're missing from answers you should be in, this is the order we work in:

1. Capability pages. One page per process and per industry you serve, with the facts a buyer filters by. In my experience, this is the biggest single change for both Google and AI search.
2. The defining sentence. Put one plain sentence about who you are and what you make on the homepage and the about page. Use the same sentence in your directory profiles and in your structured data (schema markup).
3. Directory and map listings. Correct the company description, address, phone and services on your Google Business Profile, Bing Places and the industry directories you're listed in. Remove duplicates.
4. The RFQ path. Being named only helps if the buyer can act. A buyer collecting quotes sends the same drawing to several plants at once. If your form throws an error or the page loads slowly, they don't call to ask what happened. They move on to the next plant. The site has to take a drawing in one step and confirm that it arrived.
5. Outside proof. Ask your best customers for reviews, publish one case study with real facts, and say yes when a trade publication asks for an expert view.

Sometimes the website is the problem. Sometimes the website is fine and the directories tell a different story. The check above shows you which.

What nobody can promise

No one can guarantee that ChatGPT or Google's AI will name you as a supplier. The answers change from run to run and model to model, and they depend on sources you don't control. What you can control is whether the facts about you are clear, consistent and easy to find. In our checks, that is what decided who got named.

Frequently asked questions

Is this different from SEO?

Mostly no. The same work helps both: capability pages, plain facts, consistent listings and proof from other sources. The difference is in what you check. For AI search, you read the answers and see who gets named, not only where your pages rank.

How long until AI answers change?

It varies. A week after our own relaunch, a search engine still showed our old pages. On a site we launched this year, the first leads from AI answers came within days. Plan for months, check monthly with the same questions, and watch the trend.

We build websites and run SEO for manufacturers around this path, from a buyer's question to an RFQ in your CRM. See how we work with manufacturers and CNC machine shops, or tell us about your project.

Konstantin Tesov, Founder, Uwindi